# Lantern: full article text > Complete text of every Lantern article. The index with links is at https://lantern.is/llms.txt. --- # AI shopping agents are already visiting stores, and Walmart and BigCommerce just let them in Source: https://lantern.is/resources/ai-shopping-agents-are-already-visiting-stores Author: Lantern Briefing Published: 2026-10-21 Requests from AI agents that browse stores for shoppers rose from 12.5 million in August to 58.2 million in September 2026, and on October 21 Walmart and BigCommerce opened the door to them. Image: An aerial view of a highway interchange where several elevated roads cross Image source: https://storage.googleapis.com/lantern-website-493621-content/responsive-media/webp-v1-q80-effort4-autorotate/3da7fb0cc1d1c3ea509678a18df71e25f25967b174eb257db40c677afd89532c/320-640-960-1280-1600/original.jpg Credit: Photo by Denys Nevozhai on Unsplash AI agents that shop on a person's behalf are already visiting online stores in volume. DataDome, which runs bot protection for websites, counted 12.5 million requests in August 2026 from AI agents that work through a browser, and 58.2 million in September, according to its [AI Traffic Report Q3 2026](https://datadome.co/threat-research/ai-traffic-report-q3-2026/), published October 15. Meta's Muse became fully available only on September 22, and without it the September total was still 38.7 million, about three times August's. ChatGPT's cloud browser accounted for most of that rise, going from 6.1 million requests to 27.4 million, while Claude's browser extension went from 3.6 million to 10.5 million. These visits look different from the shoppers and crawlers a store is used to. Each Muse user's agent runs in its own cloud virtual machine, so DataDome found that 99.3% of sampled Muse requests came from cloud and CDN infrastructure rather than from phones or home connections, and Muse does not identify itself as an AI agent. A bot filter that trusts home internet traffic and challenges data-center traffic treats a customer's agent the same way it treats a scraper. When Jefferies pointed Muse at several hundred leading US shopping and travel sites, about a third blocked it and another third challenged or limited it, [according to its October 8 note](https://finance.yahoo.com/technology/ai/articles/metas-muse-ai-agent-blocked-170000815.html). DataDome sells tools to manage this traffic, and its counts cover only the sites on its own network. On October 21, 2026, three companies made their own answer to these agents public. Walmart launched Agent Gateway, which lets approved AI agents search Walmart and Sam's Club inventory and buy for their users over the Model Context Protocol, [TechCrunch reported](https://techcrunch.com/2026/10/21/while-amazon-blocks-ai-agents-walmart-rolls-out-the-welcome-mat/). Walmart's terms for agents say an agent may not act beyond what the shopper intends or see the shopper's payment information, that Walmart's returns process applies when an agent gets an order wrong, and that agents breaking the terms will be blocked. Agent Profiles, rolling out in November 2026, will give each agent its own profile in a customer's account with a per-purchase approval or a spending limit. Amazon took the opposite position on September 20, when it [began blocking Muse](https://www.geekwire.com/2026/amazon-blocks-metas-muse-ai-assistant-in-new-standoff-over-agentic-shopping/) from shopping on Amazon.com. Commerce, the parent company of BigCommerce, and Cloudflare made the call for every eligible BigCommerce store. They [turned on WebMCP by default](https://www.globenewswire.com/news-release/2026/10/21/3385458/0/en/commerce-and-cloudflare-bring-ai-shopping-assistants-to-bigcommerce-merchants-online-stores-with-webmcp.html), which gives a browser agent a way to ask the store directly for its products, prices, sizes and stock, build a cart, and hand the shopper a link to check out on the merchant's own site. Merchants can switch it off in a new settings menu that also manages Commerce's other AI connections. The [WebMCP specification](https://webmachinelearning.github.io/webmcp/) is still a draft published by a W3C community group. Anthropic added another agent of this kind on October 21 with a research preview of [Claude Duet](https://claude.com/resources/articles/claude-duet), an agent for professionals and small business owners that works from a virtual machine with its own browser. It is available by waitlist to Pro and Max subscribers in the U.S. Before it pays for anything, Claude Duet shows the user what it is about to buy and waits for approval, then pays with a virtual card through Stripe's Link wallet or checks out with Shopify's Shop Pay. On a BigCommerce store, the platform has now decided how these agents are received. On other stores that decision often sits with the bot protection in front of the site. Ask whoever manages it what happens when an AI agent arrives from a cloud address and tries to open a product page or add an item to the cart, and settle on the answer you want before Black Friday on November 27, 2026. ## Common questions ### How much AI agent traffic are online stores seeing? On DataDome's network, AI agents that browse sites on a person's behalf made 58.2 million requests in September 2026, up from 12.5 million in August. Meta's Muse made up a third of the September total after becoming fully available on September 22. ### Why do bot filters block AI shopping agents? Agents such as Muse and Claude Duet run in cloud virtual machines, so their requests come from data-center addresses rather than shoppers' own devices. DataDome found that 99.3% of sampled Muse requests came from cloud and CDN infrastructure, and Muse does not identify itself as an AI agent. ### What did Walmart and BigCommerce launch on October 21, 2026? Walmart launched Agent Gateway, which lets approved AI agents search Walmart and Sam's Club inventory and buy for their users under terms Walmart sets. Commerce and Cloudflare turned on WebMCP by default for eligible BigCommerce stores, so browser agents can search a store's catalog and build a cart. ### Can a BigCommerce merchant turn WebMCP off? Yes. Merchants can switch it off in a new settings menu that also manages Commerce's other AI connections. ## Sources - [DataDome: The AI Traffic Report Q3 2026](https://datadome.co/threat-research/ai-traffic-report-q3-2026/) - [Help Net Security: Half the AI bot traffic hammering your website comes from Meta](https://www.helpnetsecurity.com/2026/10/19/datadome-ai-agent-traffic-report/) - [Yahoo Finance: Meta's Muse AI agent blocked by one-third of websites tested, Jefferies says](https://finance.yahoo.com/technology/ai/articles/metas-muse-ai-agent-blocked-170000815.html) - [TechCrunch: While Amazon blocks AI agents, Walmart rolls out the welcome mat](https://techcrunch.com/2026/10/21/while-amazon-blocks-ai-agents-walmart-rolls-out-the-welcome-mat/) - [GeekWire: Amazon blocks Meta's Muse AI assistant](https://www.geekwire.com/2026/amazon-blocks-metas-muse-ai-assistant-in-new-standoff-over-agentic-shopping/) - [GlobeNewswire: Commerce and Cloudflare Bring AI Shopping Assistants to BigCommerce Merchants' Online Stores with WebMCP](https://www.globenewswire.com/news-release/2026/10/21/3385458/0/en/commerce-and-cloudflare-bring-ai-shopping-assistants-to-bigcommerce-merchants-online-stores-with-webmcp.html) - [Anthropic: Introducing Claude Duet](https://claude.com/resources/articles/claude-duet) - [W3C Web Machine Learning Community Group: WebMCP draft specification](https://webmachinelearning.github.io/webmcp/) --- # Shopify hands the catalog to AI agents while shoppers still verify before buying Source: https://lantern.is/resources/shopify-hands-the-catalog-to-ai-agents-while-shoppers-still-verify-before-buying Author: Lantern Briefing Published: 2026-10-20 Since October 13, 2026, any AI agent can search Shopify's catalog and check out with Shop Pay. A survey the next day found nine in ten AI shoppers check a recommendation elsewhere before buying. Image: Hands holding a smartphone Image source: https://storage.googleapis.com/lantern-website-493621-content/responsive-media/webp-v1-q80-effort4-autorotate/6247f45d0e7b26062bdce4272585cfc32d13bd4e75ed9ebac4cec44dce4ea3fc/320-640-960-1280-1600/original.jpg Credit: Photo by Erik Mclean on Unsplash On October 13, 2026, Shopify opened its catalog to any AI agent. Through its Shop MCP connection, an agent or app can search more than a billion products in Shopify Catalog and take the shopper through a Shop Pay checkout. The shopper confirms the total and payment method before the order goes through, unless they have turned on a setting that lets the agent pay within a budget they set. Orders placed this way appear in the merchant's Shopify admin. A day later, on October 14, Product.ai published a survey of 1,493 U.S. adults that shows how shoppers treat those recommendations. Among the 550 who use AI for product research, 89.9% had checked an assistant's recommendation through another source before buying, up from 85.7% in June. Only 21.6% had let an assistant complete a purchase from start to finish. Agents can now get from search to checkout on Shopify, while most of the shoppers who use them still look for a second opinion before paying. The survey also shows which assistants those shoppers use to research a purchase, and two of them sit far ahead of the rest. The same two lead on trust. Asked which assistants they trust most for shopping, 43.5% named Gemini and 42.2% named ChatGPT. A wrong answer has a cost: six in ten AI shoppers who had caught an assistant's mistake changed their plans, by skipping the purchase, buying a different product or giving up on AI for that kind of shopping. Merchants decide how much of this reaches their store. In the Agentic section of Shopify admin, turning off the default "Allow Shopify to manage for me" setting lets a merchant set catalog access and direct checkout channel by channel. Turning off catalog access stops third-party agents from reaching products through Shopify Catalog, although products can still appear through ordinary web crawling. The second look is where your product facts get tested. Ask Gemini and ChatGPT about a product you sell the way a shopper would, then compare the answer with your product page and your listings elsewhere. Wherever a price or a specification differs, correct the source the assistant is reading. ## Common questions ### What did Shopify launch on October 13, 2026? Shop MCP, a connection that lets any AI agent or app search the more than a billion products in Shopify Catalog and complete a Shop Pay checkout. The shopper confirms the total and payment method before the purchase completes. ### Do shoppers check what AI assistants recommend? Most do. In Product.ai's October 2026 survey, 89.9% of U.S. shoppers who use AI for product research had verified a recommendation through another source before buying, up from 85.7% in June. ### Can a Shopify merchant keep AI agents out of its catalog? Yes. With "Allow Shopify to manage for me" turned off in the Agentic section of Shopify admin, a merchant can turn off catalog access for each AI channel. Products can still appear through web crawling. ## Sources - [Shopify developer changelog: Add shopping and Shop Pay checkout to any agent with Shop MCP](https://shopify.dev/changelog/posts/add-shopping-and-shop-pay-checkout-to-any-agent-with-shop-mcp) - [Shop Help Center: personal agents](https://help.shop.app/en/shop/shopping/personal-agents) - [Shopify Help Center: agentic storefronts](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts) - [PR Newswire: Product.ai survey finds four in five AI shoppers haven't let an AI assistant buy for them](https://www.prnewswire.com/news-releases/productai-survey-finds-four-in-five-ai-shoppers-havent-let-an-ai-assistant-buy-for-them-nine-in-10-check-its-advice-before-purchase-302907868.html) --- # Gen X shoppers still start at Google, Walmart and Amazon, with AI in a supporting role Source: https://lantern.is/resources/gen-x-shoppers-still-start-at-google-walmart-and-amazon Author: Lantern Briefing Published: 2026-10-19 Half of Gen X shoppers who researched online started at Google and about four in ten at Walmart or Amazon, while ChatGPT and Gemini each reached about one in five, PYMNTS Intelligence found. Image: An open sign hanging on a shop's glass door Image source: https://storage.googleapis.com/lantern-website-493621-content/responsive-media/webp-v1-q80-effort4-autorotate/d01c0087e23ed7253e3bcfca801bbbb109b3e7385a8915a79b42391209b25a43/320-640-960-1280-1600/original.jpg Credit: Photo by Tim Mossholder on Unsplash Gen X shoppers still begin most product searches where they always have. Among Gen X consumers who researched online before their most recent purchase, 50% used Google, 40% used Walmart and 39% used Amazon, while ChatGPT reached 21% and Gemini 17%, according to [PYMNTS Intelligence's Gen X report](https://www.pymnts.com/report/study/gen-x-clicks-then-commits-how-to-win-the-store-sale/), published in October 2026. AI plays a supporting role for this generation. In July 2026, 18% of Gen X consumers said they typically started online retail research with an AI tool, and 56% said they would rather shop in a physical store. Among the Gen X consumers who did use AI to discover products, 67% used Gemini and 46% named it the most helpful, ahead of ChatGPT at 48% and 34%. Millennials ranked the two the other way around. A second development, from October 9, 2026, concerns how a business can be read by more than one assistant at once. Sierra, which is developing the Personal Agent Protocol with Meta, published a draft of the protocol and named 35 more design partners, among them OpenAI, Visa, Mastercard, PayPal and Target. A business publishes a file at /.well-known/poppy.json on its website so a customer's personal agent can find it, and the customer signs in through OAuth to grant the agent access. One token then works across the company's APIs, its website and conversations with its own agent. The draft is an early step. Sierra says it will run design workshops over the following month and then publish a reference implementation. Its direction suits a customer base spread across several assistants, because one published description of a business can serve any agent that supports the protocol. The near-term work is in the PYMNTS numbers. If your customers skew Gen X, make sure your products read well on Google, Walmart and Amazon first, then ask Gemini and ChatGPT the questions those customers ask about your category and judge your product pages by both answers. ## Common questions ### Where do Gen X shoppers start looking for products? Among Gen X consumers who researched online before their most recent purchase, 50% used Google, 40% Walmart and 39% Amazon. ChatGPT reached 21% and Gemini 17%, according to PYMNTS Intelligence. ### Which AI assistant do Gen X shoppers use most? Among Gen X consumers who have used AI to discover products, 67% used Gemini and 46% named it the most helpful, compared with 48% and 34% for ChatGPT, according to PYMNTS Intelligence. ### How many Gen X shoppers start their research with AI? In July 2026, 18% of Gen X consumers said they typically started online retail research with an AI tool. More than half, 56%, said they would rather shop in a physical store. ### What does the Personal Agent Protocol draft cover? Sierra, which is developing the protocol with Meta, published a draft on October 9, 2026, with 35 additional design partners. A business publishes a /.well-known/poppy.json file so personal agents can find it, and a customer grants an agent access by signing in through OAuth. ## Sources - [PYMNTS: Gen X clicks, then commits: how to win the store sale](https://www.pymnts.com/report/study/gen-x-clicks-then-commits-how-to-win-the-store-sale/) - [Sierra: the Personal Agent Protocol draft and new design partners](https://sierra.ai/blog/poppy) --- # AI shopping has two gates: product data and agent checkout Source: https://lantern.is/resources/ai-shopping-has-two-gates-product-data-and-agent-checkout Author: Lantern Briefing Published: 2026-10-13 Algolia bought a product data company on October 6, 2026, and Forter launched checkout controls for AI agents on October 13. An AI-assisted sale has to clear both. Image: A woman checks a parcel with her phone beside a laptop and boxes Image source: https://storage.googleapis.com/lantern-website-493621-content/responsive-media/webp-v1-q80-effort4-autorotate/230db681e07b5c1fd19f54cd46a97c13079b8d538bd505944db82510db86765f/320-640-960-1280-1600/original.jpg Credit: Photo by Rifki Kurniawan on Unsplash On October 6, 2026, Algolia announced that it had acquired Velou, a New York company that turns product imagery and catalog content into structured, retail-specific product data. A week later, on October 13, Forter launched its Agentic Commerce Engine for merchants selling to AI shopping agents. One move works on the product record and the other on checkout, and an AI-assisted sale has to clear both. Algolia's case for the deal rests on the record itself. "When the underlying product record gets richer, every experience reasoning over it gets better," the company wrote, adding that the enriched record that fixes a store's search bar is the same one that makes its catalog readable to an agent. For a brand, the attributes an assistant needs to match a product to a shopper's question have to be in the product data before any assistant can use them. Forter's numbers show why checkout is getting its own tools. AI agent orders across Forter's network have more than quadrupled since August, and automated fraud doubled in the two months before the launch. Eran Vanounou, Forter's chief technology and product officer, said every merchant is now effectively opted in to AI shopping, whether or not they have prepared for it. The engine separates legitimate AI assistants from fraud and abuse bots in real time and supports headless checkout, so a merchant can set the conditions under which agents buy. A dashboard shows which AI agents are transacting on the site, which lets the merchant choose the agents it accepts. The product data gate is the one a marketing team controls. For the questions your shoppers ask most often, confirm that the attributes an answer depends on, such as size or material, are written into your catalog data as well as shown in photos. The checkout gate usually sits with whoever runs payments and fraud, and the question for that team is whether a legitimate agent's order would clear the fraud rules you run now. ## Common questions ### What did Algolia buy? Velou, a New York-based product intelligence company that turns product imagery and catalog content into structured, retail-specific product data. Algolia announced the acquisition on October 6, 2026. ### What does Forter's Agentic Commerce Engine do? Launched on October 13, 2026, it separates legitimate AI assistants from fraud and abuse bots and supports headless checkout for agents. A dashboard shows merchants which AI agents are transacting on their site. ### How fast are AI agent orders growing? On Forter's network, AI agent orders more than quadrupled between August and mid-October 2026, while automated fraud doubled over the two months before October 13. ## Sources - [Algolia: Algolia acquires Velou](https://www.algolia.com/blog/algolia/algolia-acquires-velou) - [PR Newswire: Forter turns the AI shopping agent surge into a secure revenue channel for merchants](https://www.prnewswire.com/news-releases/forter-turns-the-ai-shopping-agent-surge-into-a-secure-revenue-channel-for-merchants-302906333.html) --- # AI shopping splits into a paid track and an access-gated track Source: https://lantern.is/resources/ai-shopping-splits-into-a-paid-track-and-an-access-gated-track Author: Lantern Briefing Published: 2026-10-12 Jefferies found a third of leading US shopping and travel sites block Meta's Muse agent, while ChatGPT Ads tripled its advertiser count between August 31 and October 9, 2026. Image: A gray steel gate closed with a padlock Image source: https://storage.googleapis.com/lantern-website-493621-content/responsive-media/webp-v1-q80-effort4-autorotate/7bab7bebc901414ddc3218a2df15dced668fe6ba49cfe61d4ed47ec3caabd110/320-640-960-1280-1600/original.jpg Credit: Photo by Masaaki Komori on Unsplash On October 8, 2026, Jefferies reported what happened when it pointed Meta's Muse shopping agent at several hundred leading US shopping and travel websites. Roughly a third of the sites denied it access, and another third challenged or limited it. The rest let it through. Jefferies said new restrictions were appearing almost daily across the sites it tested. A day later, on October 9, New Street Research analyst Dan Salmon reported that ChatGPT Ads had passed 60,000 advertisers, a figure shown on a slide at Advertising Week New York. OpenAI had cited more than 20,000 on August 31. Salmon wrote that OpenAI is well on its way to passing 100,000 advertisers by the end of the year. Those two reports describe two separate ways a brand now reaches an AI shopper. The paid route is growing quickly. The organic route depends on whether the assistant can reach your product pages in the first place. Jefferies named website access as an increasingly important bottleneck for AI shopping agents. Meta has responded with sanctioned connectors to Walmart, Best Buy, Wayfair, Sephora and Dick's Sporting Goods, and Jefferies notes that many of those integrations are early or not yet live. An agent that can open your product pages, or reach your catalog through a connector, can describe your products when a shopper asks. The paid route is filling up. Commentary from tipsheet.ai noted that OpenAI's measurement and bidding tools suggest it is competing for Meta's performance budgets. The two routes call for different checks, and they usually belong to different teams. For the organic route, ask the assistants your customers use about a product you sell and see whether the answer can state its price, availability and fit from your own pages. For the paid route, run a small ChatGPT Ads test against one specific buying need, measured against your other acquisition channels, before holiday budgets are locked. ## Common questions ### What did Jefferies find when it tested Meta's Muse agent? Across several hundred leading US shopping and travel websites, roughly a third blocked Muse and another third challenged or limited it. The remaining sites let it through. Jefferies said new restrictions were appearing almost daily. ### How fast is ChatGPT Ads growing? ChatGPT Ads passed 60,000 advertisers by October 9, 2026, up from more than 20,000 on August 31, according to figures cited by New Street Research's Dan Salmon. He expects OpenAI to pass 100,000 advertisers by the end of 2026. ### Which merchants have sanctioned connectors with Meta's Muse? Walmart, Best Buy, Wayfair, Sephora and Dick's Sporting Goods, according to the Jefferies report. Many of those integrations are early or not yet live. ## Sources - [Yahoo Finance: Meta's Muse AI agent blocked by one-third of websites tested, Jefferies says](https://finance.yahoo.com/technology/ai/articles/metas-muse-ai-agent-blocked-170000815.html) - [tipsheet.ai: ChatGPT Ads passes 60k advertisers](https://tipsheet.ai/news/chatgpt-ads-60k/) --- # ChatGPT referrals show up in GA4, and most shoppers still want to buy for themselves Source: https://lantern.is/resources/track-chatgpt-referrals-ga4-shoppers-cautious-ai-purchases Author: Lantern Briefing Published: 2026-10-11 A GA4 filter isolates the visitors ChatGPT sends your store and the pages they land on. A YouGov survey says most of them want AI to compare and find, and plan to buy themselves. Image: A laptop on a sofa showing a web analytics report with charts Image source: https://storage.googleapis.com/lantern-website-493621-content/media/0bf2bfa1c64d93aa5ffbdab61793d1269573d248e62aeeb1388e1c4f84a5cb30/photo-1504868584819-f8e8b4b6d7e3 Credit: Photo by Lukas Blazek on Unsplash You can see what ChatGPT sends your store with a report you already have. In [a guide published by Practical Ecommerce on October 11, 2026](https://www.practicalecommerce.com/how-to-track-chatgpt-traffic-in-ga4), Ann Smarty walks through it in GA4. Open Acquisition, then Traffic acquisition, choose "Session source / medium" in the drop-down below the graph, and type "gpt" in the search field. What remains is your ChatGPT referral traffic. The number will probably look small beside organic search. Smarty's argument is about quality. Drawing on client sites she has optimized for AI answers, she writes that "ChatGPT traffic is often the most engaging source, even more than organic search." She presents that as her own experience and says she knows of no study on the question, so use it as a reason to look at your own engagement figures for that segment. The second report in her guide is the more useful one. ChatGPT keeps the prompts people typed to itself, but GA4 records where those visitors landed. Under Engagement, then Landing pages, add a filter for "Session source / medium" containing "gpt". Each page in that list is one ChatGPT chose to send someone to, and as Smarty puts it, "Each page solves a problem." A visitor who lands on a comparison page was most likely weighing options. A visitor who lands on a single product page had probably been pointed to that product already. Read the top pages as a list of the questions ChatGPT believes you answer. A survey published two weeks earlier says what those visitors expect to do once they arrive. YouGov polled more than 3,300 consumers in the UK and the US for ACI Worldwide, and [The Paypers reported the results on September 25, 2026](https://thepaypers.com/payments/news/aci-worldwide-finds-shoppers-cautious-about-ai-led-purchases). More than half, 53%, said they were uncomfortable letting an AI assistant make purchases for them, and 7% would allow it under conditions they set in advance. Asked which AI shopping features they want, respondents put help with price at the top. Interest in price drop alerts was higher among people aged 45 to 54, at 40%, than among those aged 18 to 24, at 27%. So the shopper arriving from ChatGPT has usually used the assistant to compare and to find, and intends to finish the purchase on your site. The pages in your landing page report carry that sale. Check that each one gives the price, the availability and the answer to the question that brought the visitor there near the top of the page, because that visitor has already done the comparing. ## Common questions ### How do I see ChatGPT traffic in GA4? Go to Acquisition, then Traffic acquisition, choose "Session source / medium" in the drop-down, and type "gpt" in the search field. The rows that remain are sessions referred by ChatGPT. ### Can I see what people asked ChatGPT before they visited? ChatGPT does not share prompts with the sites it sends visitors to. The landing page is the best available clue. Filter the GA4 Landing pages report by a session source containing "gpt" and read each page as the answer to a likely question. ### Is ChatGPT traffic better than organic search traffic? Ann Smarty writes that it is often the most engaging source on the client sites she works with, and she notes there is no study behind that yet. Compare engagement for the two sources in your own GA4 property before drawing a conclusion. ### Are shoppers ready to let AI buy for them? Most say they are not. In the YouGov survey for ACI Worldwide, 53% were uncomfortable with an AI assistant making purchases for them, and 7% would allow it under preset conditions. Price drop alerts and price comparisons were the most wanted features, at 35% each. ## Sources [Practical Ecommerce: How to track ChatGPT traffic in GA4](https://www.practicalecommerce.com/how-to-track-chatgpt-traffic-in-ga4) [The Paypers: ACI Worldwide finds shoppers cautious about AI-led purchases](https://thepaypers.com/payments/news/aci-worldwide-finds-shoppers-cautious-about-ai-led-purchases) --- # Shopify shows AI shopping agents your markdowns, and Amex proposes cover for agent errors Source: https://lantern.is/resources/shopify-compare-at-prices-ai-agents-amex-purchase-protection Author: Lantern Briefing Published: 2026-10-10 Updated: 2026-10-20 Shopify now shares compare-at prices with AI shopping agents, and American Express has proposed protection for charges an AI agent makes in error. Both turn a default into a decision. Image: A red shop window sign reading Sale in large white letters Image source: https://storage.googleapis.com/lantern-website-493621-content/media/0827f99d9239be09f8405747dc683e8dce7ff2af93347fb1a5184488f8f7cd34/photo-1643378813320-9eebae4833cb Credit: Photo by Markus Spiske on Unsplash Since October 9, 2026, Shopify has shared compare-at prices with AI shopping agents through Shopify Catalog. [Shopify's changelog](https://changelog.shopify.com/posts/compare-at-prices-now-displayed-to-agentic-channels) says agents reading the catalog previously saw only the current price, so a markdown that was obvious on your storefront never reached the assistant recommending the product. With compare-at prices included, an assistant can present a discounted product to the shopper as a discount. You control this in Shopify admin, in the Compare-at section of the Shopify Catalog mapping page. A compare-at price is a claim, and an assistant will repeat it to a shopper as fact. Before holiday promotions start, look through the products you have on sale and confirm that each compare-at price is one the product really sold at. On October 6, 2026, American Express released its Amex Business Playbook for Agentic Commerce. According to [Yahoo Finance's report](https://finance.yahoo.com/technology/ai/articles/american-express-telling-merchants-prepare-134818873.html), it proposes Agent Purchase Protection, which would cover eligible Card Members and merchants when a registered AI agent makes a charge in error, and it sets up a pilot Merchant AI Advisory Council. The protection is an announced intent for now. Amex has yet to publish its terms or say how an agent becomes registered. The report cites a figure that explains why a card network would make the offer: 93% of merchants believe AI providers should pay for an agent's mistakes. A week later, on October 13, 2026, American Express published its Intent Intelligence framework. [Digital Commerce 360](https://www.digitalcommerce360.com/2026/10/16/american-express-ai-framework/) reports that the framework describes how to record a shopper's purchase instructions before an AI agent transacts. Card issuers can use those records to detect fraud, approve payments, and resolve disputes. For a retailer, the framework could help determine whether an ordering mistake originated with an AI agent. Amex confirmed the framework has no live integrations yet. The company said publishing it is intended to let other payment and technology providers adopt its approach. The same day, JPMorgan Chase drew a harder line. CEO Jamie Dimon told analysts the bank will not allow outside AI agents to access customer accounts and operate on their behalf, [CNBC reported](https://www.cnbc.com/2026/10/13/jpmorgan-ai-agents-account-access.html). Dimon cited the risk of rogue agents at scale, saying there could be thousands of them. He said customers using outside agents must understand the risks and potential transfer of responsibility. A JPMorgan spokeswoman confirmed the bank currently blocks external AI agents from operating on behalf of retail customers. Agents can read some data, like balances, through open banking providers, but they cannot send money or pay a bill. The two positions point to different answers for how an AI-assisted purchase clears. Amex wants to record what a shopper authorized before an agent buys. JPMorgan wants to keep outside agents out of customer accounts entirely. For a brand, the practical question is which of the payment methods your customers use will work when an assistant buys on their behalf, and the answer now depends on the bank and the card network behind each one. Another development concerns which agents reach a store at all. [A Bernstein analysis reported by Investing.com on October 10, 2026](https://au.finance.yahoo.com/news/consumer-ai-agents-could-upend-151409484.html) notes that Amazon blocked Meta's Muse shopping assistant within two weeks of its launch, while Shopify, Walmart, Best Buy, Gap, Sephora and six other companies partnered with it for shopping and checkout. Bernstein expects companies to grant access selectively, one agent at a time. Cloudflare's figure in the same analysis shows the volume behind that choice. AI agent requests on its network rose more than 1,700% in a year. Taken together, these are decisions that used to be made for you by default. Which prices an assistant sees is now a setting. Who pays for an agent's mistake has a first proposal from a card network. Which agents may read your catalog or check out is a choice other large sellers are making agent by agent. Whether an assistant can actually complete a purchase depends on the payment rails and account access behind it. Start with the one that is live. The compare-at setting takes a few minutes to review, and it shapes how every assistant reading Shopify Catalog describes your sale prices. ## Common questions ### How do I control whether AI agents see my compare-at prices? Shopify's changelog points to the Compare-at section of the Shopify Catalog mapping page in Shopify admin. Review the setting there, and check that the compare-at prices on your discounted products are accurate. ### Is Agent Purchase Protection available now? American Express announced it on October 6, 2026 as a proposal inside its Amex Business Playbook for Agentic Commerce. Terms and the process for registering an agent are not yet published. ### What does the Amex Intent Intelligence framework do? Published on October 13, 2026, it describes how to record a shopper's purchase instructions before an AI agent buys. Amex confirmed it has no live integrations yet and is meant for other providers to adopt. ### Can an AI agent access a Chase account to make a purchase? JPMorgan Chase blocks external AI agents from operating on behalf of retail customers. Agents can read some data through open banking providers, but they cannot send money or pay a bill. ## Sources - [Shopify Changelog: Compare-at prices now shared to agentic channels](https://changelog.shopify.com/posts/compare-at-prices-now-displayed-to-agentic-channels) - [Yahoo Finance: American Express is telling merchants how to prepare for agent shopping](https://finance.yahoo.com/technology/ai/articles/american-express-telling-merchants-prepare-134818873.html) - [Investing.com via Yahoo Finance: Consumer AI agents could upend online shopping](https://au.finance.yahoo.com/news/consumer-ai-agents-could-upend-151409484.html) - [American Express launches Intent Intelligence framework for AI shopping agents](https://www.digitalcommerce360.com/2026/10/16/american-express-ai-framework/) - [JPMorgan blocks outside AI agents like Muse, Instinct from account access](https://www.cnbc.com/2026/10/13/jpmorgan-ai-agents-account-access.html) --- # ChatGPT ads stay unproven for the holidays while AI checkout tools arrive Source: https://lantern.is/resources/chatgpt-ads-unproven-holidays-ai-checkout-tools Author: Lantern Briefing Published: 2026-10-09 Advertisers are keeping holiday budgets on Meta and Google while ChatGPT ads stay in testing. Two October 8 launches make orders placed by AI assistants easier to complete and to vet. Image: A woman on a sofa holds a payment card while shopping on a laptop Image source: https://storage.googleapis.com/lantern-website-493621-content/media/8eedcc54ed26ed66c1ab9ff8b0378dc279a919cea5666fa823faf1cdfaf0e3fc/photo-1758874385215-c86fe62b446f Credit: Photo by Vitaly Gariev on Unsplash Brands are heading into the 2026 holiday season with their ad budgets on Meta and Google, and ChatGPT ads are staying in the test column. [Modern Retail reported on October 9, 2026](https://www.modernretail.co/marketing/why-chatgpt-ads-wont-likely-impact-holiday-sales-in-a-major-way-this-year/) that hundreds of brands are trying OpenAI's ad offering, and that advertisers treat it as a costly place to experiment with little expectation of significant holiday sales. The sharpest detail in that report comes from Terence Einhorn, head of insights at the measurement firm Measured. He told Modern Retail that a click on ChatGPT costs around six times its Google equivalent, that results so far have been "pretty bad", and that most of the sales those ads record would have happened anyway. Sensor Tower counted more than 1,400 advertisers running ChatGPT ads in August 2026, most of them large companies. Einhorn's advice to everyone else was a small learning budget. If your fourth-quarter plan is tight, that is a reasonable ceiling. The same week brought two launches for the other way an AI assistant can bring you revenue, which is by completing the purchase itself. On October 8, 2026, [Crossmint released its Agent Commerce Toolkit](https://www.prnewswire.com/news-releases/crossmint-launches-agent-commerce-toolkit-enabling-ai-agents-to-pay-with-any-card-and-check-out-online-through-one-integration-302901649.html), a self-serve developer kit that lets an AI agent store a shopper's card and check out at an online store through one integration. It uses tokenized credentials from Visa Intelligent Commerce and Mastercard Agent Pay where a card is eligible, and it chooses a checkout path for each purchase, including Shop Pay and the Universal Commerce Protocol. Crossmint says merchants do not need to add anything to accept these orders. Co-founder Alfonso Gómez-Jordana Mañas described what agent developers run into without it: "One card works and another doesn't. One store has an API and the next has a bot check." Also on October 8, [Riskified announced Agent Identity Risk Intelligence](https://ir.riskified.com/news-releases/news-release-details/riskified-launches-agent-identity-risk-intelligence-know-who), which identifies the AI assistant behind an order or a customer service request, connects it to a real customer, and returns a risk signal in real time. Riskified's announcement describes a change that matters to any store with a returns desk. An assistant keeps working after the purchase. It can compare prices across sites, check for a refund when a price drops, and look for no-fee returns, on every order it places. Put those together and an order placed by an assistant can reach your checkout without any work on your side. For the fourth quarter, the useful preparation is on your product pages and in your policies. An assistant comparing prices and reading return terms for a shopper acts on exactly what you publish, so your price, availability and return terms need to be accurate and easy to find. Ask your fraud provider how it treats an order placed by an AI assistant before holiday volume arrives. Modern Retail's sources described the same priority. Anders Bill of Superfiliate said brands are "very invested" in ranking at the top of AI recommendations on their own merits. The paid placement can wait for better evidence. The assistant's own recommendation, and the order that follows it, are already in play. ## Common questions ### Should my brand advertise on ChatGPT this holiday season? Most advertisers quoted by Modern Retail on October 9, 2026 are keeping holiday budgets on Meta and Google. Measured's Terence Einhorn suggests a small learning budget if you can afford one, and no more until the channel has a track record. ### How much do ChatGPT ads cost compared with Google? Einhorn told Modern Retail that the cost of a click on ChatGPT is around six times the Google equivalent. He also said most of the sales those ads record are ones the brand would have made anyway. ### Do I need to change my store for AI agents to buy from it? Crossmint says its toolkit completes checkout through Shop Pay, the Universal Commerce Protocol or browser automation, with no agent support required from the merchant. Accurate prices, availability and return terms matter more, because the assistant acts on what your pages say. ### How will I know an order came from an AI assistant? Fraud tools are starting to report it. Riskified's Agent Identity Risk Intelligence, announced October 8, 2026, identifies the assistant behind an order or service request and connects it to a real customer. Ask your own provider what it offers. ## Sources [Modern Retail: Why ChatGPT ads won't likely impact holiday sales in a major way this year](https://www.modernretail.co/marketing/why-chatgpt-ads-wont-likely-impact-holiday-sales-in-a-major-way-this-year/) [Crossmint: Agent Commerce Toolkit launch announcement](https://www.prnewswire.com/news-releases/crossmint-launches-agent-commerce-toolkit-enabling-ai-agents-to-pay-with-any-card-and-check-out-online-through-one-integration-302901649.html) [Riskified: Agent Identity Risk Intelligence launch announcement](https://ir.riskified.com/news-releases/news-release-details/riskified-launches-agent-identity-risk-intelligence-know-who) --- # Checkout inside AI is live; merchant fraud checks are the next test Source: https://lantern.is/resources/checkout-inside-ai-is-live-merchant-fraud-checks-are-the-next-test Author: Lantern Briefing Published: 2026-10-08 On October 8, 2026, Newegg switched on Universal Commerce Protocol checkout inside Google's Gemini app and AI Mode in Search. On October 8, 2026, Newegg switched on Universal Commerce Protocol checkout inside Google's Gemini app and AI Mode in Search. A shopper in the United States can discuss a product, decide, and pay with Google Pay without leaving the conversation. Newegg stays the merchant of record, fulfills the order, and handles support. The purchase happens inside Google's experience. For a brand, that changes the question. Getting named by an AI assistant was the hard part. Now the order itself has to survive a second set of systems: the merchant's payment rails, fraud checks, and returns logic. Visa's Pavan HS put it plainly on October 6, 2026. Merchants are treating AI agents as another sales channel that must work with existing checkout infrastructure. The same requirements around payments, fraud, loyalty, returns, and customer service apply. Tokens are one answer already on the table. A token can connect an agent-initiated purchase with later returns, refunds, and other transactions across channels. That gives the merchant a thread to pull when something goes wrong. Identifying legitimate agents without weakening existing fraud controls remains an unresolved merchant problem. Merchants work to keep controls strong while accepting real AI orders. Your product data has to be structured well enough that an assistant can connect a shopper's need to your SKU. Your catalog and order flow also have to be legible to the systems that stand between an AI recommendation and a completed sale. Clean product attributes and consistent identifiers give those systems what they need to approve a legitimate AI order. The Newegg launch makes the stakes concrete. A brand that gets recommended inside Gemini and AI Mode can now close the sale without sending the shopper anywhere else. A brand that gets recommended and then declined at the merchant's fraud screen has spent the same effort and earned nothing. The order has to clear the payment and fraud systems after the recommendation. Common questions What changed with Newegg on October 8, 2026? Newegg integrated Universal Commerce Protocol checkout on Google. U.S. shoppers can buy eligible Newegg products directly inside AI Mode in Google Search and Gemini without leaving the conversation, with Newegg as merchant of record. How can tokens help with AI shopping? Tokens can connect agent-initiated purchases with returns, refunds, and other transactions across channels. They give merchants a way to reconcile an AI order later without weakening fraud controls. What should a brand do now? Treat AI shopping as a sales channel that must work with existing checkout infrastructure. Ensure product data is structured for AI assistants, and prepare for token-based reconciliation so AI-initiated orders clear fraud and payment systems. ## Sources - [U.S. Shoppers Can Now Discover and Purchase Eligible Newegg Products Directly Within the Gemini App, and Google Search,](https://finance.yahoo.com/technology/ai/articles/u-shoppers-now-discover-purchase-123000305.html) - [PYMNTS | AI Agents Bring Shoppers, but Merchants Must Close the Sale](https://www.pymnts.com/news/artificial-intelligence/2026/ai-agents-bring-shoppers-but-merchants-must-close-the-sale/) --- # The agent infrastructure race now has a storefront test Source: https://lantern.is/resources/the-agent-infrastructure-race-now-has-a-storefront-test Author: Lantern Briefing Published: 2026-10-07 On October 5, DaVinci Commerce announced an expansion of its Agentic Commerce Experience Platform, introducing patent-pending Product Context Memory and a new… On October 5, DaVinci Commerce announced an expansion of its Agentic Commerce Experience Platform, introducing patent-pending Product Context Memory and a new platform architecture. The company says the expansion helps brands and retailers get products discovered, chosen, and bought as AI agents reshape how consumers shop. One day later, on October 6, Meta and Sierra published Personal Agent Protocol, an open standard for how personal AI assistants authenticate with businesses. Walmart and Shopify are founding partners. The protocol defines a direct connection between an assistant and a company, replacing the current workaround where assistants load pages and click forms the way people do. For a marketing leader, the protocol news is worth watching. The storefront numbers are worth acting on. Swap, a provider of AI-powered storefronts, reported that nearly 70 percent of online shoppers abandon their carts before checkout. Swap's CMO, Juan Pellerano-Rendon, said the brands that win Black Friday Cyber Monday are the ones where a shopper gets an answer fast enough to still be in the mood to buy. That answer often concerns sizing and fit. A shopper who asks whether a jacket fits is one answer away from buying or leaving. The protocol and the knowledge graph are infrastructure bets. They will matter over the next year. The storefront result is a proof point available now. A brand that can answer sizing and fit questions in a single exchange converts shoppers that a traditional product page loses. Product data is the smaller test. A product page that states fit, material, and use case in plain language gives an AI assistant something to answer with. The brands that get named in AI shopping answers are the ones whose pages a model can read and answer from. Lantern shows where your products rank in AI search and drafts the product data changes that move them up. Common questions What is Personal Agent Protocol? Personal Agent Protocol is an open standard from Meta and Sierra, announced October 6, that defines how personal AI assistants authenticate with businesses. Walmart and Shopify are founding partners. It gives companies visibility into what assistants do on their websites or APIs. What is Product Context Memory? Product Context Memory is a patent-pending technology from DaVinci Commerce, announced October 5, that helps brands and retailers get products discovered and bought as AI agents reshape how consumers shop. It is part of an expansion of the company's Agentic Commerce Experience Platform. Why do sizing and fit questions matter? Nearly 70 percent of online shoppers abandon their carts before checkout, according to Swap. Sizing and fit questions often block a purchase. A conversational storefront can answer those questions in one exchange, which keeps the shopper in the mood to buy. What should a brand do first? Test the smaller thing: make product data answer the shopper's actual question in one exchange. State fit, material, and use case in plain language on the product page. Watch the protocol announcements, but do not wait for them. ## Sources - [Introducing Personal Agent Protocol | Sierra](https://sierra.ai/blog/introducing-personal-agent-protocol) - [DaVinci Commerce Unveils Brand Agent Featuring Patent-Pending Product Context Memory For Personal AI Agents](https://www.prnewswire.com/news-releases/davinci-commerce-unveils-brand-agent-featuring-patent-pending-product-context-memory-for-personal-ai-agents-302898082.html) - [How agentic commerce helps brands like Willy Chavarria optimize for BFCM - Modern Retail](https://www.modernretail.co/sponsored/how-agentic-commerce-helps-brands-like-willy-chavarria-optimize-for-bfcm/) --- # AI discovery adoption hits 58.4M; merchants lag on AI platform handoff Source: https://lantern.is/resources/ai-discovery-adoption-hits-58-4m-merchants-lag-on-ai-platform-handoff Author: Lantern Briefing Published: 2026-10-06 PYMNTS reports that 58.4 million U.S. adults now begin online retail research with AI, up 18% since June. Image: A person relaxing on a couch holds a smartphone in a pale green case Image source: https://storage.googleapis.com/lantern-website-493621-content/media/d729a88ac890d1e8f32744f389b135cb89a8f27d3e4d54858ee2ac0f765b3399/photo-1543428390-52250776dd46 Credit: Photo by charlesdeluvio on Unsplash PYMNTS reports that 58.4 million U.S. adults now begin online retail research with AI, up 18% since June. The same research shows only 24% of consumers would let an AI agent shop and buy for them, while 52.3% prefer keeping their card outside the AI app. The discovery happens in AI, the checkout stays with the customer. That split matters for how you feed the channel. PYMNTS's Consumer AI Benchmark tracks the same shift with a stricter question. In its August survey of 2,191 U.S. consumers, fielded August 10 to 13, the share who name AI as their main starting point for retail research grew from 21.4% in June to 23.2%, about 60.9 million people. The share who have used AI to discover products rose from 22.9% to 25.8% over the same two months, reaching 67.8 million consumers. Across 60 U.S. merchants surveyed by PYMNTS, 53% felt prepared to share real-time inventory data, but only 23% felt prepared to connect with AI platforms or agent networks. Merchants are least ready at the handoff to AI platforms, and most ready on the catalog side. CarMax shows what the handoff looks like when a brand gets it right. On September 29, the company said it routes 100% of inbound store and customer experience center calls through AI voice agents built with Sierra. The agents answer questions like store hours and vehicle availability and pass the rest to associates. CarMax also puts more than 45,000 vehicles in front of shoppers through its ChatGPT app. CEO Keith Barr said every used car is an individual SKU, which makes the final transaction hard for AI to take over. The AI handles the discovery and the service, a person closes the sale. That is the pattern to watch in your category. DoubleVerify made the channel measurable on October 5. It launched [attribution measurement for advertising on ChatGPT](https://ir.doubleverify.com/news-events/press-releases/press-releases-details/2026/DV-Launches-Attribution-Measurement-and-Advances-Brand-Suitability-for-ChatGPT-Ads/default.aspx), so advertisers can incorporate ChatGPT Ads as a touchpoint within the broader customer journey and understand contribution to outcomes and ROAS. Weight Watchers added ChatGPT Ads as a new source of member acquisition and used DV Rockerbox to measure it. ChatGPT's attributed cost per acquisition was 15.3% below Weight Watchers' established benchmark during the measured period. A growth team can use that number to compare ChatGPT Ads against other acquisition channels. For a brand, the first move is to confirm that the assistants your customers ask can read your product data, because both the CarMax pattern and the new ChatGPT Ads measurement build on it. ## Common questions ### How many consumers actually use AI for shopping research? PYMNTS reports that 58.4 million U.S. adults now begin online retail research with AI, up 18% since June. In its August benchmark survey, 23.2% of U.S. consumers, about 60.9 million people, named AI as their main starting point for retail research, up from 21.4% in June. ### Are consumers ready to let AI agents buy for them? Only 24% of consumers in the PYMNTS research would let an AI agent shop and buy for them, and 52.3% prefer keeping their card outside the AI app. Discovery happens in AI, checkout stays with the customer. ### What does CarMax's AI rollout show? CarMax routes 100% of inbound store and customer experience center calls through AI voice agents built with Sierra, and it puts more than 45,000 vehicles in front of shoppers through its ChatGPT app. The AI handles discovery and service, a person closes the sale. ### How do I measure advertising on ChatGPT? DoubleVerify launched attribution measurement for ChatGPT Ads on October 5, letting advertisers incorporate the channel into their broader customer journey. Weight Watchers measured a CPA 15.3% below its benchmark using DV Rockerbox. ## Sources - [PYMNTS | Holiday Retail's AI Test Starts With Who Gets to Hit Buy](https://www.pymnts.com/study/holiday-retail-consumer-ai-shopping-behavior/) - [PYMNTS Intelligence | From Prompt to Purchase: Who Wins the AI Shopper?](https://www.pymnts.com/study_posts/from-prompt-to-purchase-who-wins-the-ai-shopper/) - [PYMNTS | CarMax Lets AI Take the Call, Not the Sale](https://www.pymnts.com/news/artificial-intelligence/2026/carmax-lets-ai-take-the-call-not-the-sale/) - [DoubleVerify Holdings, Inc. - DV Launches Attribution Measurement and Advances Brand Suitability for ChatGPT Ads](https://ir.doubleverify.com/news-events/press-releases/press-releases-details/2026/DV-Launches-Attribution-Measurement-and-Advances-Brand-Suitability-for-ChatGPT-Ads/default.aspx) --- # Checkout moved inside the AI assistant on October 5, and the recommendation now decides the sale Source: https://lantern.is/resources/checkout-moved-inside-the-ai-assistant-on-october-5-and-the-recommendation-now-decides-the-sale Author: Lantern Briefing Published: 2026-10-05 On October 5, 2026, TikTok launched an AI Shopping Assistant with a new in-app checkout feature that lets users buy directly from brands. Image: A woman on a couch holds a payment card and a smartphone while shopping online Image source: https://storage.googleapis.com/lantern-website-493621-content/media/70053463034a3ae93cdeb3524ad4704bb7031ef9b2e506999005a40b987a4171/photo-1758876201516-8a24429cbb3b Credit: Photo by Vitaly Gariev on Unsplash On October 5, 2026, TikTok launched an AI Shopping Assistant with a new in-app checkout feature that lets users buy directly from brands. Constructor announced Agentic Checkout the same day, embedding Stripe checkout inside its onsite AI shopping agents. Gap Inc. began rolling out AI-powered shopping tools across its brands that week. The common thread is checkout moving into the conversation where the recommendation happens, so the pay button belongs to the brand that conversation names. TikTok's assistant offers real-time product details, shipping information, sizing, availability, and help completing a purchase. It remembers a user's preferences through the conversation. The new checkout feature lets users buy directly from a brand from the For You feed with one click. TikTok is building these features with partners including Salesforce, Shopify, and Shoplazza. For a brand, the assistant is the new storefront. The assistant names the brands whose product details it can retrieve, and the one-click checkout follows the name. Constructor's Agentic Checkout makes the same move on a retailer's own site. Shoppers can discover products, ask questions, get recommendations, refine choices, and pay with Stripe inside a single conversation. Constructor positioned the launch against tools that help shoppers discover products then push them back into a traditional online flow to buy. [Four in 10 shoppers have already used AI shopping tools](https://www.unite.ai/constructor-embeds-stripe-checkout-in-its-onsite-ai-shopping-agents/), Constructor said, citing a Constructor and Shopify report. For a brand, structured product data decides whether it gets named. An assistant that can read sizing and availability can answer the shopper's question and carry the conversation to checkout. Gap Inc. is rolling out AI-powered shopping tools for all of its brands 30 years after launching its first e-commerce website. Starting Monday, customers can get recommendations from Alta Daily and Daydream. Banana Republic and Old Navy are launching conversational shopping agents on their websites and apps. A retailer that built its first website in 1996 is now putting AI assistants in front of shoppers across every brand. The assistants answer from the retailer's product data, and the quality of those answers depends on that data being complete and consistent. The assistant's answer is the recommendation. Checkout is the last step. A marketing or ecommerce leader should check which product facts an assistant can retrieve when it decides what to name. The sale goes to the brand the assistant can describe, which makes readable product pages and product data the place to start. ## Sources - [TikTok rolls out an AI shopping assistant and one-click checkout | TechCrunch](https://techcrunch.com/2026/10/05/tiktok-rolls-out-an-ai-shopping-assistant-and-one-click-checkout/) - [Constructor Embeds Stripe Checkout in Its Onsite AI Shopping Agents, Unite.AI](https://www.unite.ai/constructor-embeds-stripe-checkout-in-its-onsite-ai-shopping-agents/) - [Gap unveils AI-assisted shopping, three decades after its first website | Retail Dive](https://www.retaildive.com/news/gap-ai-assisted-shopping-three-decades-after-website/832070/) --- # AI shopping is still a question of which brand gets named Source: https://lantern.is/resources/ai-shopping-is-still-a-question-of-which-brand-gets-named Author: Lantern Briefing Published: 2026-10-04 On 29 September 2026, Federal Reserve Governor Christopher Waller told Sibos in Miami that AI shopping changes what banks must authenticate. Image: A hand holds a blank payment card over a card terminal on an orange table Image source: https://storage.googleapis.com/lantern-website-493621-content/media/f43e86d09b593e3a6a367efd15c6edbd418cb33cd89bfaa0abc47a52d3bc44f1/photo-1628527304948-06157ee3c8a6 Credit: Photo by Towfiqu barbhuiya on Unsplash On 29 September 2026, Federal Reserve Governor Christopher Waller told Sibos in Miami that AI shopping changes what banks must authenticate. An AI system may be acting for a consumer or business rather than merely assisting one. Issuers will need evidence that an agent has the authority to pay on a buyer's behalf, in addition to a valid payment credential. Waller named authentication, liability, and fraud as three major barriers to scaling delegated payments. Merchant usage is still small. Checkout.com's Agentic Commerce 2026 report, published in June, found that only 3% of UK and US merchant transactions involve AI agents, even though 89% of merchants surveyed say they are actively preparing and 42% are already testing. The infrastructure is arriving faster than usage. Mastercard expanded Agent Pay with a score that estimates how likely it is that a transaction was started by an AI agent. Stripe made every hosted checkout page agent-ready. Shopify switched agent checkout on by default for eligible merchants. Forrester analysts argue that consumer demand for autonomous purchasing remains unproven. They note that Amazon blocked Meta's Muse AI agent from making purchases, but advise retailers not to read that as a definitive signal about commerce. Meaningful adoption is expected first in narrow use cases like holiday promotions. Consumers might embrace automation to grab a limited edition item just as it drops, or to secure concert tickets in the best section when they go on sale. That points to a practical next move for brands. Win the AI recommendation in those narrow early use cases rather than wait for checkout standards. Checkout.com found that [59% of AI-assisted purchases still end on Amazon](https://www.fintechweekly.com/magazine/articles/agentic-commerce-launch-month-liability-risk-2026). Shoppers let the AI research, then go back to the checkout they already trust. The moment that matters is the moment the AI names a brand. If a shopper asks an AI assistant what to buy for a holiday promotion, the brand named in the answer gets the sale. A model names the brands whose pages and product data it can read and trust, which makes that data the place to spend effort through the holidays. Run the exact holiday query a shopper types, like "best gift for a runner under $50," and check which brand the assistant names. Then make the product data changes that put your brand in that answer. ## Common questions ### What did the Federal Reserve say about AI agents and payments? On 29 September 2026, Governor Christopher Waller said AI shopping changes authentication because an AI system may be acting for a consumer or business. Issuers will need evidence that an agent has the authority to pay on a buyer's behalf. ### How many transactions involve AI agents? Checkout.com's Agentic Commerce 2026 report found only 3% of UK and US merchant transactions involve AI agents, even though 89% of merchants surveyed say they are actively preparing and 42% are already testing. ### Where do AI-assisted purchases end up? 59% of AI-assisted purchases still end on Amazon, according to Checkout.com. Shoppers let the AI research, then go back to the checkout they already trust. ### What should brands do now? Win the AI recommendation in narrow early use cases like holiday promotions. Run the exact holiday query a shopper types and check which brand the assistant names. ## Sources - [Agentic Commerce Had Its Biggest Launch Month. Nobody Wants to Own the Risk., FinTech Weekly](https://www.fintechweekly.com/magazine/articles/agentic-commerce-launch-month-liability-risk-2026) - [Checkout.com | Agentic Commerce 2026](https://www.checkout.com/guides-and-reports/agentic-commerce-2026) - [Meta, Amazon, And The Real Question About Agentic Commerce | Forrester](https://www.forrester.com/blogs/meta-amazon-and-the-real-question-about-agentic-commerce/) - [PYMNTS | Fed's Waller: AI Shopping Changes What Banks Must Authenticate](https://www.pymnts.com/news/artificial-intelligence/2026/federal-reserve-waller-says-ai-shopping-changes-what-banks-must-authenticate/) --- # The AI shelf is consolidating, and the payment layer is being formalized Source: https://lantern.is/resources/the-ai-shelf-is-consolidating-and-the-payment-layer-is-being-formalized Author: Lantern Briefing Published: 2026-10-02 On October 2, Lightspeed Commerce published an analysis of 460,000 AI shopping responses across ChatGPT and Google's AI products. Image: A bicycle parked outside the window of an independent lamp shop Image source: https://storage.googleapis.com/lantern-website-493621-content/media/428985c3fec7c5fafa2789c089b9400e1f17ca3c14e3721884fe1eda0e916a0a/photo-1528698827591-e19ccd7bc23d Credit: Photo by Tem Rysh on Unsplash On October 2, Lightspeed Commerce published an analysis of 460,000 AI shopping responses across ChatGPT and Google's AI products. EMVCo closed public comments on its Agentic Payments Framework for Specifications on September 30. Read together, they show who gets named when a shopper asks an AI assistant what to buy, and whether the payment clears. Vaer AI and Lightspeed Commerce found that bigger merchants have an edge from the very start of AI search. When ChatGPT included retailer names in its own web searches, 97% were large or mid-sized chains. Small and local retailers accounted for just 1 to 4%. That pattern holds even when the assistant can see smaller stores. Large and small retailers each made up roughly 38% of the sources AI cited, but big chains received 52% of the top recommendations. Smaller businesses got 20%. One example from the study makes the point. A shopper asking for office supplies in New York saw three small local retailers among the six sources ChatGPT cited. The first recommendation was Staples, with the local options listed below it. The pattern repeats across assistants. Alexa for Shopping showed toys from 177 brands, and nearly every one came from Amazon's own store. The same toy request to ChatGPT produced recommendations from 24 retailers. Google's AI Mode went to 37. When the assistant's default answer draws from a narrow set of large merchants, a brand outside that set has to earn its place in the answer before its product pages matter. The EMVCo work addresses the second half: what happens after the assistant names a product. EMVCo's proposed Intent Services would let payment participants register, reference, retrieve, and manage consumer-authorized intent before, during, and after a transaction. The standards body may also develop know your agent capabilities and transaction indicators to identify agents and signal their involvement in payments. A shopper could tell an AI agent to buy a specific product, spend no more than a certain amount, or make recurring purchases. The framework would carry that authorization through the card payment. [PYMNTS Intelligence](https://pymnts.com/news/artificial-intelligence/2026/emvco-framework-targets-consumer-intent-agentic-payments) found that 56% of consumers would let AI agents search and compare products, but just 37% would let them authorize payments. Search and comparison are already a consumer behavior. Payment authorization is the next step, and the standards are being written now. For a marketing or ecommerce leader, the practical question is whether your brand can be seen and whether it can complete the sale. Ask the assistants the questions your buyers ask and note which retailers get named. Then check whether your payment flow can accept an AI agent's transaction. If the answer is a national chain or the assistant's default retail partner, the shelf you need to be on is the store the assistant already looks at. The EMVCo framework is moving from experiment to infrastructure. The brands that verify this now will be the ones whose authorization survives when the standards go live. ## Common questions ### What does the Lightspeed study actually show? Lightspeed Commerce and Vaer AI analyzed 460,000 AI shopping responses across ChatGPT and Google's AI products. They found that larger merchants get recommended more often, even when smaller retailers are present in the sources the assistant cites. ### Why does the EMVCo framework matter to a brand? EMVCo's proposed Intent Services would standardize how consumer-authorized intent moves through card payments. If an AI agent makes a purchase on a shopper's behalf, the framework defines how that authorization is registered, referenced, and managed. ### What should I do first? Ask the AI assistants the questions your buyers ask and note which retailers get named. If your brand is not among them, the next step is to be stocked where the assistant already looks. ### Is this a search problem or a payments problem? It is both. First, the assistant has to see and cite your product. Second, the buyer's authorization has to survive the transaction. The two developments address different halves of the same distribution challenge. ## Sources - [Lightspeed Commerce CEO: our study analyzed half a million AI shopping responses and revealed who gets to compete | Fortune](https://fortune.com/2026/10/02/lightspeed-commerce-study-ai-shopping-ecommerce-who-gets-to-compete/) - [Where brands stand when AI agents do the shopping | Digiday](https://digiday.com/marketing/in-graphic-detail-where-brands-stand-when-ai-agents-do-the-shopping) - [PYMNTS | EMVCo Framework Targets Consumer Intent in Agentic Payments](https://pymnts.com/news/artificial-intelligence/2026/emvco-framework-targets-consumer-intent-agentic-payments) --- # AI shopping now has a conversion premium and a checkout button Source: https://lantern.is/resources/ai-shopping-now-has-a-conversion-premium-and-a-checkout-button Author: Lantern Briefing Published: 2026-10-01 Adobe's holiday forecast puts a revenue number on AI traffic for the first time at this scale. Image: A pile of wrapped holiday gifts tied with ribbon and twine Image source: https://storage.googleapis.com/lantern-website-493621-content/media/70baa5a0acffac173398d44867b3e35dfe043b28feab65dca6fa62ea42b872a1/photo-1513885535751-8b9238bd345a Credit: Photo by freestocks on Unsplash Adobe's holiday forecast puts a revenue number on AI traffic for the first time at this scale. The company expects AI traffic to U.S. retail sites to more than double this holiday season, up 130% year over year, according to Adobe's annual online shopping forecast. Consumers arriving from AI assistance added items to their carts at a 32% higher rate than those not referred by AI, according to Retail Dive's report on the forecast. Adobe paired the forecast with a complementary survey of 5,000 U.S. consumers. That growth compounds on a much larger base each year. Adobe measured AI-referred visits to U.S. retail sites up 1,300% in the 2024 holiday season and 693.4% in 2025, so a 130% rise in 2026 still more than doubles last season's volume. Shoppers who arrive because an AI assistant named your brand are more likely to add to cart. For a marketing leader planning Q4, that changes the math on where to focus product data and page readiness before the spike. Google's Flipkart pilot shows the next step. Shoppers in India can now buy products from Walmart-owned Flipkart directly through Gemini and AI Mode. Users in the test see a Buy button on select Flipkart product listings that takes them directly to a Flipkart checkout flow without leaving the AI interface. Google plans to roll out the experience more broadly later in October, ahead of India's festive shopping season. The checkout button moves the transaction inside Gemini. A purchase that starts with a product recommendation can finish without a visit to the retailer's site, and the sale goes to the brand named in the answer. PYMNTS, in an October 1 analysis, argues that the emerging infrastructure for AI shopping is building a protocol stack. Multiple standards perform different jobs: one helps an AI agent discover inventory, another lets agents communicate, another records what a consumer authorized. The biggest mistake merchants could make in AI shopping, PYMNTS warns, may be waiting for one protocol to win. There probably will not be one. For a marketing leader, that warning means treating AI distribution as an infrastructure task now. The Adobe numbers give you the reason to audit what an AI platform can discover and transact from your product data. The Flipkart pilot shows the interface where that data gets used. The protocol stack means there is no single standard to wait for. Start with the audit. Pick a category question your shoppers ask, such as "best running shoes for flat feet." Check whether an AI assistant can name your product, show your price and availability, and complete a checkout. The channel is now both measurable and transactional. ## Common questions ### How fast is AI traffic to retail sites growing? Adobe expects AI-referred visits to U.S. retail sites to rise 130% this holiday season, after growth of 693.4% in 2025 and 1,300% in 2024. The percentage is smaller because each year starts from a much larger base. ### When will checkout inside AI interfaces be available beyond India? Google's Flipkart test is live for select users in India, with a broader rollout planned for later in October. The company has not announced dates for other markets. ### Do I need to wait for a single AI shopping standard before acting? PYMNTS argues there will not be one standard. The infrastructure is a protocol stack with multiple standards doing different jobs. The practical move is to make your product data readable to AI platforms now. ## Sources - [Adobe: U.S. Holiday Shopping Season to Hit Record $275.1 Billion Online, Rising 6.7% YoY](https://news.adobe.com/news/2026/09/adobe-us-holiday-shopping-season-to-hit-record) - [Adobe: Holiday Shopping Season Drove a Record $257.8 Billion Online with Consumers Embracing Generative AI Tools](https://news.adobe.com/news/2026/01/adobe-holiday-shopping-season) - [Adobe: Holiday Shopping Season Drove a Record $241.4 Billion Online](https://news.adobe.com/news/2025/1/adi-pr-full-season-recap) - [Consumer AI use a boon for the holidays as AI traffic doubles | Retail Dive](https://www.retaildive.com/news/consumer-ai-use-boon-holidays-ai-traffic-doubles/831790/) - [Google tests buying from Walmart-owned Flipkart through Gemini and AI Mode in India | TechCrunch](https://techcrunch.com/2026/09/26/google-tests-buying-from-walmart-owned-flipkart-through-gemini-and-ai-mode-in-india/) - [PYMNTS | A Vocab Playbook for Merchants Navigating Agentic Standards](https://www.pymnts.com/news/artificial-intelligence/2026/the-alphabet-soup-of-agentic-standards-a-vocab-playbook-for-merchants-and-psps) --- # DoorDash now texts carts; Mastercard now scores agent payments Source: https://lantern.is/resources/doordash-now-texts-carts-mastercard-now-scores-agent-payments Author: Lantern Briefing Published: 2026-09-30 On September 30, 2026, DoorDash launched a text-to-order AI agent inside Apple Messages. Image: A smiling woman at her front door receives a takeout food delivery Image source: https://storage.googleapis.com/lantern-website-493621-content/media/a6a2682101668af1fcec288a5987143f88ac662c7a04a12059862ba42870505e/photo-1752070182361-9fa562ed7f97 Credit: Photo by JESHOOTS.COM on Unsplash On September 30, 2026, DoorDash launched a text-to-order AI agent inside Apple Messages. A user can send a prompt like "order my usual," ask for a specific dish, or request a local recommendation. The agent searches local spots and suggests a cart from the prompt. It can text photos of the food it recommends and, for group orders, handle mixed dietary preferences and different quantities in one order. DoorDash opened a waitlist for U.S. users to try the feature. The agent takes over the two places where a brand normally wins or loses: the menu a customer scrolls and the checkout flow a customer completes. It builds the cart from the product data it can read, so the brands that land in the cart are the ones that describe what they sell in a form the agent understands. Mastercard added an AI-transaction probability score and a fraud intelligence layer to Agent Pay in September 2026, with Cloudflare and Skyfire as partners. The score is a trust signal for the transaction itself. When an agent initiates a purchase, the payment layer decides whether to approve it. What should you test? Start with the data an agent would need to assemble a cart from your catalog. That means product names, descriptions, prices, availability, and order minimums, all in a machine-readable format. Then test the payment handoff. Does your order data pass a fraud check when an agent submits it? Can an agent complete a purchase without a human stepping in? The person who owns revenue should own these tests. DoorDash's launch and Mastercard's update sit on the same line. An agent now assembles the order, and a payment network scores it, before a person reviews either one. ## Sources - [DoorDash launches an AI agent you can text to order food | TechCrunch](https://techcrunch.com/2026/09/30/doordash-launches-an-ai-agent-you-can-text-to-order-food/) - [Mastercard advances agentic commerce with new trust and intelligence services | Mastercard US](https://www.mastercard.com/us/en/news-and-trends/press/2026/september/new-trust-and-intelligence-services.html) --- # Commerce adds BigCommerce and Feedonomics product enrichment Source: https://lantern.is/resources/commerce-product-data-enrichment-bigcommerce-feedonomics Author: Lantern Briefing Published: 2026-09-29 Commerce’s new enrichment tools generate product copy and catalog fields. Retailers still need to check each claim against the product they sell. Commerce announced product-data enrichment tools for BigCommerce and Feedonomics on September 29, 2026. The tools generate catalog content from existing product information, giving retailers another way to prepare titles, descriptions, feature bullets and FAQs. The useful question for a merchant is whether the resulting copy helps a shopper choose the right product. According to [Commerce’s announcement](https://www.globenewswire.com/news-release/2026/09/29/3370746/0/en/commerce-introduces-data-enrichment-for-bigcommerce-and-feedonomics-to-power-agentic-commerce.html), Feedonomics Enrichment supports product content for multiple channels. BigCommerce Catalog Enrichment works inside the merchant control panel, where a user selects products, supplies brand guidance, reviews generated content and applies it. Those review steps deserve as much attention as generation. Suppose a storage-box listing includes external dimensions and a pack quantity of two, but says nothing about dishwasher use. A helpful description preserves the dimensions and makes the two-pack clear. A generated claim that the box is dishwasher-safe would need evidence from the manufacturer before publication. More detailed copy can still introduce a false buying reason. Review the fields that affect the purchase first: quantity, size, material, compatibility and included accessories. Keep dimensions labeled so a buyer can distinguish the box’s outside measurements from its usable interior. Where a fact is absent, ask the product owner to supply it. Rephrasing incomplete information cannot establish a missing specification. Also check the exact variant that a shopper will receive. A description for a black two-pack should agree with its image, selected variant and price. Copying an attractive description across a product family can obscure differences that matter at checkout. An assistant comparing options needs those differences to be explicit. For Google’s AI search features, [Google’s guidance](https://developers.google.com/search/docs/appearance/ai-features) says there are no special technical requirements beyond the foundations of Search. It also advises keeping structured data consistent with visible page content. Enrichment can help maintain useful information; using a particular tool does not establish that a product will be recommended. A sensible first release is a reviewed group of products with recurring customer questions. Ask merchandising to confirm the facts and customer support to check whether the copy answers those questions. Our [guide to product pages AI can interpret](https://lantern.is/resources/ai-legible-pdps) covers the information to make explicit on the page. ## Common questions ### What did Commerce launch? Commerce introduced Feedonomics Enrichment and BigCommerce Catalog Enrichment to generate and improve product content using existing catalog information. ### Should generated product descriptions be published automatically? Review claims that affect fit, compatibility, safety or what is included. Match the copy to an approved product specification and the exact variant being sold. ### Will enriched product data guarantee an AI recommendation? No. Accurate information helps a shopper evaluate a product, but no enrichment tool can promise inclusion in an AI answer. ## Sources [Commerce: Data enrichment for BigCommerce and Feedonomics](https://www.globenewswire.com/news-release/2026/09/29/3370746/0/en/commerce-introduces-data-enrichment-for-bigcommerce-and-feedonomics-to-power-agentic-commerce.html) [Google Search Central: AI features and your website](https://developers.google.com/search/docs/appearance/ai-features) --- # Shopify adds browser-agent checkout with buyer confirmation Source: https://lantern.is/resources/shopify-browser-agent-checkout-buyer-confirmation Author: Lantern Briefing Published: 2026-09-28 Shopify’s WebMCP checkout support lets browser agents assist with an order and return control when the buyer’s input is needed. Shopify added WebMCP support for checkout on September 28, 2026, giving browser-based AI agents tools to read and update the checkout a buyer has open. An agent can submit the order after buyer confirmation. Steps that need a person’s input return control to the buyer. [Shopify’s announcement](https://shopify.dev/changelog/posts/webmcp-support-for-checkout) names payment authentication and blocking checkout extensions as examples of those handoffs. It says the tools require no merchant configuration. The change concerns agents operating through the buyer’s browser; it should not be read as a promise that every assistant already supports the experience. For a merchant, the immediate task is to verify a complete journey through the store’s own checkout. A product recommendation can be accurate while the resulting order contains the wrong size, shipping option or quantity. Check the final order against what the buyer actually chose. Start with a test purchase: one shirt, a specified color and size, and delivery to a known destination. Ask the assistant to carry those choices through checkout. Before confirmation, the buyer should be able to recognize the selected variant, quantity, delivery address and final total. Then make a supported change, such as correcting the delivery address. Check the shipping options and total again before approval. The purpose is to establish that the purchase review reflects the current order after a change. Include a step that returns control to the buyer. Where a payment challenge appears, the person should understand what they need to complete and when they can continue. Use the store’s approved test process to verify the handoff without placing unintended customer orders. After completion, inspect the order confirmation and the merchant’s record. Both should describe the same item and amount, and the customer should have an order reference for support. If the buyer stops before approval, verify that the result is equally clear. This is also an opportunity to review custom checkout steps. A delivery instruction, required acknowledgment or other store-specific requirement should remain understandable when an assistant helps with the flow. Document any step the chosen agent cannot complete so the team can set accurate expectations. The [AI shopping pilot guide](https://lantern.is/resources/scope-ai-shopping-pilot-before-buying-software) can help bound that review to a small product set and a supported assistant before a broader rollout. ## Common questions ### Does Shopify require new merchant configuration for these tools? Shopify says the checkout tools require no merchant configuration. Retailers should still verify their own checkout and the assistant they intend to use. ### Can an agent submit an order without buyer confirmation? The announced tools submit checkout after buyer confirmation. Payment authentication or other required input can return control to the person. ### Does this mean every AI assistant can now check out on Shopify? The announcement establishes Shopify’s browser checkout support. Confirm that the assistant you plan to use supports the tools and can complete your store’s journey. ## Sources [Shopify: WebMCP support for checkout](https://shopify.dev/changelog/posts/webmcp-support-for-checkout) --- # Amazon gives sellers ongoing AI workflows with approval controls Source: https://lantern.is/resources/amazon-seller-assistant-workflows-approval-controls Author: Lantern Briefing Published: 2026-09-26 Amazon’s September Seller Assistant update brings ongoing inventory and pricing workflows. Start by checking scope, approvals and the resulting changes. Amazon [announced new Seller Assistant workflows on September 23, 2026](https://www.aboutamazon.com/news/innovation-at-amazon/seller-assistant-plugin-amazon-quick-claude), allowing sellers to monitor their businesses and prepare responses to changing conditions even while logged out. The update also introduces a selling partner plugin for Amazon Quick and Anthropic’s Claude. Amazon describes the plugin as a beta for sellers in its U.S. stores. For workflows, sellers can choose recommendations or actions; the company says actions are reviewed and approved before execution and recorded in an audit trail. Those controls deserve as much attention as the proposed time savings. A practical first use is a small group of products with a recurring operational problem. Consider a seller who regularly checks whether a promoted backpack has enough inventory. Begin with a workflow that flags the affected item and proposes a response. Before allowing a change, check the exact variant, current stock and any inbound shipment the team already knows about. The proposal should make the commercial choice clear. Reducing promotion, changing a listing and placing a replenishment order carry different costs. Assign a person who can judge the proposed response against supplier lead times and the campaign plan. Keep the initial product group narrow enough that the owner can inspect each result. For a pricing workflow, write down the products it covers and the conditions under which a proposed change is acceptable. Review a sample proposal against the current offer, fees and margin requirements. Then check that the recorded result matches the approved change. An attractive recommendation is useful only if the right item was changed in the intended way. Evaluate the pilot on completed work: whether a real stock or pricing issue was caught, how much review it required, and whether the final listing or inventory decision was correct. A seller-side workflow improves operations when the team can verify those outcomes. Product recommendations shown to shoppers remain a separate question. The [product-page guide](/resources/ai-legible-pdps) offers checks for the customer-facing information affected by listing changes. ## Common questions ### Where is the selling partner plugin available? Amazon’s announcement describes a U.S. seller beta and integrations with Amazon Quick and Claude. Confirm availability in the account you plan to use. ### Can a seller start with recommendations only? Amazon says sellers can choose recommendation-only workflows. Its action workflows include review and approval before execution. ### What should the first workflow cover? Choose a recurring task affecting a small, identifiable product group. Keep a record of the issue, proposed response, approval and final result so the owner can judge its usefulness. ## Sources - [Amazon: Seller Assistant workflows and selling partner plugin, September 23, 2026](https://www.aboutamazon.com/news/innovation-at-amazon/seller-assistant-plugin-amazon-quick-claude) --- # AudioEye study links website accessibility to AI task completion Source: https://lantern.is/resources/ai-shopping-adoption-and-website-accessibility Author: Lantern Briefing Published: 2026-09-25 AudioEye’s controlled tests found large differences in AI task completion between site versions. Retailers can use the finding to examine specific shopping obstacles. Website accessibility can affect whether an AI assistant completes a task, according to a study AudioEye released on September 24, 2026. The company ran 1,560 agent tests across six websites using six commercial AI models. On the site with the most accessibility issues, reported completion was 31% on the inaccessible version and 96% on the accessible version. The [AudioEye release](https://www.prnewswire.com/news-releases/audioeye-study-finds-up-to-68-drop-in-ai-agent-task-completion-on-inaccessible-websites-302888585.html) describes a controlled experiment from a company that sells accessibility services. The results concern the tested sites and tasks. They do not establish a conversion increase that another retailer can apply to its own sales forecast. Consumer research provides a reason to examine those journeys. In a separate September 24 release, [NielsenIQ reported](https://nielseniq.com/global/en/news-center/2026/majority-of-u-s-consumers-now-use-ai-to-shop-niq-finds/) that 51% of U.S. consumers had used an AI tool to support shopping in the previous month. Its monthly research uses approximately 500 respondents. Shopping support includes research and recommendations, so that figure should not be presented as a share of completed purchases. A useful storefront review begins with one task a customer needs to complete. For a clothing product, that might be finding the fabric composition, selecting an available size and confirming the selected item in the cart. Each step can fail for a different reason and needs a specific description. Check the information first. If fabric details or measurements appear only inside an image, provide equivalent readable text. Name sizes and colors clearly so the shopper does not have to infer a choice from position or appearance alone. Make an unavailable variant distinguishable before the buyer attempts to add it. Then check the controls. A size selector should communicate the current selection, and a failed add-to-cart attempt should explain what needs attention. Ask a person to complete the journey using a keyboard and the accessibility checks appropriate to the site. An AI agent’s success cannot substitute for evaluating whether people can use the page. If an assistant stops, preserve the page, selected variant and visible error so the team can reproduce the obstacle. After a repair, repeat the same task and verify the cart’s contents. Reaching the cart is useful evidence about that step; revenue and recommendation visibility require separate evidence. The [guide to product pages AI can interpret](https://lantern.is/resources/ai-legible-pdps) covers the product facts to make explicit. Combine those content checks with accessibility work so the page both explains the offer and supports the customer’s next step. ## Common questions ### Can retailers expect AudioEye’s completion rates on their own sites? No. The reported percentages describe a particular experiment. Check relevant tasks on your own storefront before predicting an improvement. ### Does NIQ’s 51% figure mean half of purchases happen through AI? No. NIQ measured reported use of AI to support shopping, including activity before a purchase. ### Does successful AI task completion prove a website is accessible? No. Agent task completion is one observation. Human accessibility evaluation remains necessary to understand whether people can use the site. ## Sources [AudioEye: Website accessibility and AI task completion study](https://www.prnewswire.com/news-releases/audioeye-study-finds-up-to-68-drop-in-ai-agent-task-completion-on-inaccessible-websites-302888585.html) [NIQ: U.S. consumer use of AI shopping tools](https://nielseniq.com/global/en/news-center/2026/majority-of-u-s-consumers-now-use-ai-to-shop-niq-finds/) --- # Coach and Kate Spade add checkout in Gemini and Google Search Source: https://lantern.is/resources/coach-kate-spade-gemini-google-checkout Author: Lantern Briefing Published: 2026-09-24 U.S. shoppers can buy a single Coach or Kate Spade item through Google Pay in Gemini and Search. The launch makes product and order clarity central to the experience. Coach and Kate Spade products became purchasable inside Gemini and Google Search, including AI Mode, in a U.S. launch announced on September 16, 2026. Shoppers can complete a single-item checkout through Google Pay without being redirected to the brand’s website. [Tapestry’s announcement](https://www.businesswire.com/news/home/20260916065964/en/) says the experience uses the Universal Commerce Protocol, or UCP, and requires explicit consumer approval for each transaction. The launch establishes a specific purchase route for the two brands. It does not show how much revenue that route generates. For other retailers evaluating checkout inside an AI conversation, the useful question is whether the product remains clear as the shopper moves from a recommendation to an order. The familiar product page may no longer be the place where the buyer reviews every detail. Suppose a shopper chooses a handbag from an AI recommendation. A recommendation might show the right style while leaving the size or color unresolved. Before approval, the buyer should be able to identify the exact variant, its dimensions and the final charge. A photograph of a larger bag should not stand in for the smaller version selected for purchase. Separate the product’s description from promises that depend on the order. Materials and dimensions are product facts. Availability, shipping charges and delivery estimates depend on the selected item and destination. Review how the purchase experience presents both, especially after the shopper changes a choice. Returns and support also deserve attention before a pilot expands. Ask where the buyer sees the applicable return terms, how the merchant is identified and where an order confirmation leads for help. These are evaluation questions for a retailer’s own implementation; the launch announcement does not document every post-purchase step. Test a change of mind as well as a straightforward purchase. Have the shopper switch variants or decline the final order, then check that the resulting state is clear. An approval step is most useful when a person can understand and correct the choices it contains. Retailers planning their own trial can use the [AI shopping pilot guide](https://lantern.is/resources/scope-ai-shopping-pilot-before-buying-software) to choose a limited product set and define the buyer’s route through confirmation and support. The goal is a purchase the customer can recognize and manage from beginning to end. ## Common questions ### Where did Coach and Kate Spade launch this checkout? The September launch was live in the U.S. through Gemini and Google Search, including AI Mode. ### Can the assistant place the order without the customer’s approval? Tapestry says each transaction requires explicit consumer approval. The announced experience supports single-item checkout through Google Pay. ### What should another retailer verify before a similar launch? Check variant accuracy, the final total, delivery information, return terms and the path to merchant support. Test changes and cancellations within the purchase flow. ## Sources [Tapestry: Coach and Kate Spade checkout in Gemini and Google Search](https://www.businesswire.com/news/home/20260916065964/en/) --- # Consumers expect AI to handle 15% of purchases within five years Source: https://lantern.is/resources/consumer-expectations-for-ai-purchases-rise-to-15 Author: Lantern Briefing Published: 2026-09-23 Global Payments reports rising expectations for AI-assisted purchases. The survey also points to a need for clear payment approval. Consumers expect AI agents to handle 15% of their purchases within five years, according to research Global Payments released on September 23, 2026. The comparable expectation was 9% in its earlier survey. That is a rise in what respondents anticipate, with actual purchase behavior still to be established. The [Global Payments release](https://investors.globalpayments.com/news-events/press-releases/detail/516/consumers-expect-ai-to-make-15-of-their-purchases-within) also says 45% of Americans have used or would consider using an AI agent for purchases. Combining those two groups makes the figure a measure of experience or openness. It cannot be read as a 45% adoption rate. The research used two survey waves: 8,000 respondents in August and September 2025 and 8,027 in May 2026. Both covered seven markets: Australia, Brazil, China, France, Singapore, the U.K. and the U.S. The release also reports that 33% of Americans want to approve every payment. For a retailer, a repeat purchase provides a concrete way to examine that preference. Imagine a customer asking an assistant to reorder the same laundry detergent. “The same” could refer to brand, fragrance, pack size and quantity. A substitute that preserves only the brand may change the product the buyer intended to receive. A useful approval screen would show the exact item, the number of packs and the final charge, including delivery. It should also identify whether the order creates any recurring commitment. If the requested size is unavailable, present the replacement and revised total before asking for confirmation. That example gives a pilot team a manageable set of decisions. Which changes can the assistant propose? Which require the buyer to review the order again? What happens when the buyer declines a substitution or cannot complete payment? Write those expectations in plain language before the first customer uses the service. Keep the purchase record understandable after checkout. The confirmation should identify the merchant and ordered variant, then give the buyer a direct path to order support. A smooth recommendation is only useful if the customer can recognize and manage the resulting order. The survey can help a team choose questions for an [AI shopping pilot](https://lantern.is/resources/scope-ai-shopping-pilot-before-buying-software). Budget decisions should then follow the pilot’s completed orders, customer feedback and support issues, with expectations from the survey kept in context. ## Common questions ### Are AI agents already making 15% of purchases? That is not what this research establishes. The figure describes consumers’ expectations for purchases within five years. ### Does the 45% U.S. figure measure active users? It combines people who have used an agent with people who would consider doing so. The combined figure should not be presented as current use. ### What should a retailer test before allowing repeat purchases? Check exact-product matching, substitutions, final totals and recurring-order disclosures. Make it clear when the buyer must approve a changed order. ## Sources [Global Payments: Consumer expectations for AI purchases](https://investors.globalpayments.com/news-events/press-releases/detail/516/consumers-expect-ai-to-make-15-of-their-purchases-within) --- # Six banks propose principles for buyer control in AI shopping Source: https://lantern.is/resources/six-banks-publish-trust-principles-for-ai-agents-in-commerce Author: Lantern Briefing Published: 2026-09-22 Six banks published voluntary principles for AI-assisted commerce. Retailers can use them to ask clearer questions about approval, choice and order support. Six banks published principles for trusted AI-assisted commerce on September 22, 2026. ASB, Bank of America, Capital One, Commonwealth Bank of Australia, ING and NatWest set out five areas: transparency, safety, privacy and data use, choice, and interoperability. The [joint announcement](https://newsroom.bankofamerica.com/content/newsroom/press-releases/2026/09/global-banks-collaborate-on-principles-for-trusted-agentic-comme.html) presents the principles as a basis for industry discussion. They are voluntary proposals. The announcement does not create a new payment rule or guarantee a particular outcome when a shopper disputes an order. For a retailer choosing an AI shopping provider, the principles can inform a practical demonstration. Ask the provider to show what the buyer sees before approval and what happens when the proposed purchase changes. General assurances about trust are easier to assess when attached to a specific order. Suppose a shopper asks for a pair of shoes in a specific size and color. The assistant finds a different color at a higher price because the requested option is unavailable. A clear experience would show the replacement, revised total and delivery details, then ask the buyer to approve that proposal. The same demonstration should include a refusal. If the buyer declines the replacement, can they return to the original search or complete a purchase themselves? Ask how the experience preserves that choice and how it communicates that no order was placed. Next, inspect the confirmation and support route. The buyer and merchant should be able to identify the product that was approved, the amount charged and the order reference. Ask the provider how a support team can understand an issue without requiring the customer to reconstruct a long conversation. Privacy questions should be equally concrete. Ask what customer information the service needs for this purchase, which parties receive it and how the buyer is told. A retailer should understand the proposed use before asking customers to trust the experience. The principles provide questions to bring into an [AI shopping pilot](https://lantern.is/resources/scope-ai-shopping-pilot-before-buying-software). The provider still needs to demonstrate its own behavior, and the merchant still needs clear terms for payment, fulfillment and support. ## Common questions ### Are the banks’ principles mandatory payment rules? No. The announcement describes voluntary principles intended to support industry discussion. ### What are the five areas covered? Transparency, safety, privacy and data use, choice, and interoperability. ### What is the most useful approval scenario to ask a provider to demonstrate? Use an order that changes after the shopper’s initial request. Check whether the buyer sees the substitute and revised total, can decline them and receives a clear result. ## Sources [Bank of America: Joint principles for trusted AI-assisted commerce](https://newsroom.bankofamerica.com/content/newsroom/press-releases/2026/09/global-banks-collaborate-on-principles-for-trusted-agentic-comme.html) --- # Amazon blocks Meta’s Muse shopping assistant Source: https://lantern.is/resources/amazon-meta-muse-shopping-access Author: Lantern Briefing Published: 2026-09-21 Amazon confirmed blocking Muse in September. Retailers evaluating an assistant should check which stores it can access and where the buyer takes over. Amazon confirmed that it was blocking Meta’s Muse shopping assistant in a dispute over access to its marketplace, GeekWire reported on September 20, 2026. The case illustrates a practical limit for AI shopping: an assistant’s ability to discuss a product does not establish that it can access the store or complete the purchase. In [GeekWire’s reporting](https://www.geekwire.com/2026/amazon-blocks-metas-muse-ai-assistant-in-new-standoff-over-agentic-shopping/), Amazon described Muse’s access as unauthorized and said it had asked Meta to remove Amazon from the service. Those are Amazon’s stated positions. They should not be treated as an independent finding about every use of the assistant. For a retailer assessing an AI shopping channel, store coverage deserves a direct check. Ask the provider which destinations it supports, what the shopper can do there and which steps require a normal browser visit. Broad claims about shopping “anywhere” are less useful than a working journey to a specific product. Use a read-only review first. Ask the assistant to find an exact product and follow the supplied link. Check the merchant, model, variant and availability on the destination page. A recommendation can name the right product while linking to an unavailable offer or a different seller. Then examine the handoff the provider actually supports. Can the shopper carry the chosen variant into a cart? Does the assistant stop at the product page? If access fails, does it explain that clearly and let the person continue? Record the stopping point so the team knows which part of the journey it has verified. Catalog coverage and payment capability should be evaluated separately. An assistant that can retrieve public product information may still be unable to operate a store’s checkout. A retailer should describe its own available purchase route precisely, including any return to the merchant’s site. For brands selling through both a marketplace and their own store, keep each destination’s offer clear. Prices, stock and bundles can differ. A shopper who arrives at the direct store needs to see what is available there without assuming the marketplace offer carries across. The [AI referral landing-page guide](https://lantern.is/resources/ai-referred-shoppers-product-landing-page) covers the information to preserve when the shopper reaches the store. Evaluate supported routes as they work today, and respect a marketplace’s access restrictions when testing them. ## Common questions ### Why did Amazon say it blocked Muse? Amazon described Muse’s access as unauthorized, according to GeekWire’s September 20 report. That is the company’s stated reason. ### Does a product recommendation prove an assistant can buy it? No. Product discovery, access to the merchant and checkout are separate capabilities. Verify the steps that the service supports. ### What should a retailer do when an assistant cannot reach a store? Make the limitation clear and offer a supported path for the shopper to continue. Do not promise an end-to-end purchase that the service cannot complete. ## Sources [GeekWire: Amazon confirms blocking Meta’s Muse shopping assistant](https://www.geekwire.com/2026/amazon-blocks-metas-muse-ai-assistant-in-new-standoff-over-agentic-shopping/) --- # Five practical tests for AI product discovery Source: https://lantern.is/resources/practical-ecommerces-five-test-guide-for-ai-discovery Author: Lantern Briefing Published: 2026-09-20 Use identification, proof, verification, evidence and shopping checks to find product-data gaps before an AI assistant sends a buyer elsewhere. A product can be easy to find and still be difficult to buy with an AI assistant. [Practical Ecommerce's five-test guide](https://www.practicalecommerce.com/test-your-products-for-ai-discovery) separates the problem into identification, proof, verification, supplying evidence and shopping. The useful outcome is a list of facts a merchant can correct and test again. Identification asks whether the assistant understands the product and its intended use. Proof asks what supports a claim about it. Verification checks the answer against the product's actual specifications and terms. Supplying evidence asks whether a reader can reach the supporting source. Shopping checks whether the recommendation leads to the right purchasable item. Record the assistant's answer and links at each step so a plausible explanation does not substitute for a working purchase path. Consider an illustrative test for a 750 ml insulated bottle. Ask for a bottle that fits a cup holder with a 7 cm opening. If the product page gives capacity but no base diameter, the assistant cannot verify the fit from that page. An answer that says it fits has identified the product but has not established suitability. Add the measured base diameter to the page and relevant product data, then repeat the question. The evidence check goes further: open the cited page and confirm it supports the particular claim. A link to a general collection page may identify the brand while saying nothing about the bottle's dimensions. The shopping check then follows the exact variant through to the cart, checking its price, availability and delivery terms. A correct recommendation for an unavailable size still leaves the buyer without a usable answer. Use the same prompt, assistant mode and location for the follow-up, and keep several observations rather than treating one improved answer as proof of a lasting change. A merchant's own page is one possible source among several, so record any external page the assistant used as well. The [product-page checklist](/resources/ai-legible-pdps) explains how to keep visible facts, structured data and feed information consistent. The practical implication is to fix the first unsupported step in the buyer's task. “The assistant named us” and “the buyer could verify and purchase the right item” are different results, and the five checks make that difference observable. ## Common questions ### What should a merchant record for each test? Keep the exact prompt, assistant and mode, date, location where relevant, answer, cited URLs and product variant. Record whether each check passed and the specific evidence for that decision. ### Does a missing recommendation mean the product page is unreadable? No. It may reflect the request, availability, competing products or the sources the assistant selected. Inspect access and product facts, then compare several equivalent observations before attributing the omission to one cause. ### How can a team tell whether a fix helped? First verify that the corrected fact is visible and consistent across the page, markup and feed. Then repeat the relevant buyer task under comparable conditions and report the observed results with their sample size. ### Is appearing in an answer enough to pass? No. The answer also needs to support the buyer's constraints and lead to the correct offer. A recommendation with an unsupported fit claim, wrong variant or unavailable item should remain a failed task. --- # Pinterest announces Visual Search Ads beta Source: https://lantern.is/resources/pinterest-visual-search-ads-beta Author: Lantern Briefing Published: 2026-09-17 Pinterest’s Visual Search Ads beta adds placements around visual discovery. Retailers should confirm access and test how well the advertised product matches the landing page. Pinterest announced Visual Search Ads on September 17, 2026, adding an advertising format for shoppers exploring products visually. The company said a beta would reach eligible advertisers across its advertising markets in the following weeks. Retailers considering the format should confirm access in their own ad account before scheduling a campaign. The [Pinterest announcement](https://newsroom.pinterest.com/en-gb/news/new-pinterest-visual-search-performance-ads/) describes placements in search results and alongside opened Pins. Pinterest also reported more than 80 billion monthly searches, with commercial intent in more than half. Those figures describe activity across the platform. They do not establish how much traffic an individual advertiser will receive from the new format. A useful test starts with the relationship between the image and the product being sold. Suppose a shopper is exploring compact outdoor dining furniture. An image of a table surrounded by four chairs can create an expectation of a complete set. If the offer is for the table alone, both the ad and destination need to make that clear. The landing page should preserve the visible choice. Match the advertised color, material and configuration to the selected product option. If a visual contains accessories or several products, identify what the price includes. A shopper should not have to reconstruct the offer after clicking. Visual resemblance also leaves practical questions unanswered. Two chairs can look similar while differing in dimensions, weight capacity or suitability for uncovered outdoor use. Put verified specifications and care instructions where a shopper can find them. Avoid extending a manufacturer’s weather-resistance claim beyond what it actually covers. Before expanding spend, choose a product group with reliable stock and a stable landing page. Agree on the spending limit and the purchase outcome the campaign is meant to support. Keep enough of the existing campaign consistent to understand what changed, and review completed orders alongside clicks. A visually appealing placement can earn attention without producing the intended purchase. Inspect the journey on a phone as well. Follow the ad to the product, confirm the selected option, and check whether delivery and return information remain easy to reach. The same checks apply to [product pages reached from AI recommendations](https://lantern.is/resources/ai-referred-shoppers-product-landing-page): the destination needs to answer the buying question that brought the shopper there. ## Common questions ### Are Pinterest Visual Search Ads available to every advertiser? Pinterest announced a beta for eligible advertisers, with rollout planned over the weeks following September 17. Confirm account access before committing to a launch date. ### Do Pinterest’s search-volume figures predict campaign results? No. Platform-wide search activity does not establish an advertiser’s reach, order volume or acquisition cost. ### Which products are useful candidates for an initial test? Choose products with clear images, accurate variant selection and dependable availability. Make the contents of the offer and the final destination unambiguous. ## Sources [Pinterest: Visual Search Ads announcement](https://newsroom.pinterest.com/en-gb/news/new-pinterest-visual-search-performance-ads/) --- # Google expands Merchant Center AI reporting to five markets Source: https://lantern.is/resources/google-merchant-center-ai-performance-insights Author: Lantern Briefing Published: 2026-09-16 Google’s September 16 update separates available performance reporting from U.S. betas and checkout features still rolling out. Google made AI performance insights in Merchant Center available across Australia, Canada, India, New Zealand and the U.S. on September 16, 2026. [Its announcement](https://blog.google/products-and-platforms/products/shopping/google-shopping-updates-holiday-shopping/) says the report compares how a retailer's brand and products appear across surfaces including AI Mode and AI Overviews. That gives a merchandising team a specific place to examine product discovery. Establish a baseline for a relevant category, then choose a small set of products whose descriptions need a clearer explanation of use or fit. Retain the original feed values and record the date of each change so later observations have a meaningful reference point. The announcement also describes U.S. betas and gradual checkout rollouts. A team should confirm access in its own account before making a feature part of a launch commitment. Google also describes conversational attributes that add context to product listings. In [Google's test with lululemon](https://blog.google/products-and-platforms/products/shopping/google-shopping-updates-holiday-shopping/), submitted attributes appeared in 50% of relevant AI Mode recommendations. That reported result concerns whether supplied information appeared in answers. It provides a reason to examine the descriptions an assistant uses, with commercial impact assessed from the retailer's own orders. For a useful feed test, choose a customer question that has a verifiable answer. A bag's laptop compartment dimensions or a garment's care requirements are better test material than a broad claim that the item is ideal for everyone. Compare the information supplied with what the shopping experience actually presents. Repeat the same questions across multiple checks before attributing an answer change to the feed edit. Lantern's [study of repeated AI answers](https://lantern.is/resources/sampling-on-the-bubble) explains why a single observation can give a misleading impression of movement. The immediate work is to establish the baseline and make one documented improvement. A record of what changed gives the next report something specific to explain. ## Common questions ### Where are Merchant Center AI performance insights available? Google's September 16 announcement names five markets: Australia, Canada, India, New Zealand and the U.S. Account access should be confirmed in Merchant Center. ### Does the lululemon result measure sales growth? It measures how often submitted conversational attributes appeared in relevant recommendations during Google's test. Order results need a separate measurement. ### Which product information should a team test? Use a specific, substantiated detail that answers a common customer question. Keep the original value and the change date so later observations can be compared. ## Sources - [Google: Merchant Center reporting, feed attributes and rollout details](https://blog.google/products-and-platforms/products/shopping/google-shopping-updates-holiday-shopping/) --- # Mastercard: teens show more interest in AI shopping than parents Source: https://lantern.is/resources/mastercard-teen-parent-ai-shopping-survey Author: Lantern Briefing Published: 2026-09-15 A parent-and-teen survey across Europe and Israel finds greater teen interest in AI shopping. Product comparisons still need specific, verifiable answers. Teenagers in a Mastercard survey showed more interest in AI shopping assistants than their parents. In research published on September 8, 2026, 27% of teens said they were likely to use a fully AI-powered shopping assistant, compared with 16% of parents. The finding describes interest in a possible shopping experience. [Mastercard’s report](https://www.mastercard.com/news/europe/en/newsroom/press-releases/en/2026/mastercard-report-predicts-that-one-in-10-people-will-routinely-use-ai-agents-to-shop-and-pay-by-2030/) also found that 18% of teens used AI weekly to find prices or discounts, compared with 10% of parents. Opinium surveyed 13,000 parent-and-teen pairs, or 26,000 people, across 13 markets in Europe and Israel in June and July 2026. The teens were ages 13 to 18. Those boundaries matter. The results cannot stand in for U.S. adoption, and willingness to use an assistant does not establish permission to make a purchase. For a retailer, the more immediate opportunity is to make price and product comparisons easier for the people already researching them. Consider a shopper looking for headphones within their budget that work with a particular game console. An answer is useful only if it gets the connection type, microphone support and included accessories right. A lower advertised price can be misleading if the buyer also needs an adapter that is sold separately. Put those compatibility facts in readable product copy beside the relevant variant. A generic statement that headphones “work with gaming devices” leaves the buyer to resolve the exact case. Where support differs by console or connection, describe the difference explicitly and link the manufacturer’s compatibility guidance. Price comparisons need an equally clear unit. A discounted single item and a full accessory bundle are different offers. Show what is included, the current total for the selected option and any required add-ons. That helps a teen researching a purchase and the parent reviewing it make the same comparison. Use customer-support questions to identify the details most often misunderstood. Then check whether a shopper can answer those questions from the page alone. If you also test an AI assistant, retain the exact question and product link so an incorrect answer can be traced to a specific claim. The [guide to interpreting individual AI answers](https://lantern.is/resources/why-one-ai-answer-is-never-enough) explains why one response cannot establish how a brand performs everywhere. ## Common questions ### Did this study measure U.S. teenagers? No. The survey covered parent-and-teen pairs in 13 markets across Europe and Israel. ### Does interest in AI shopping mean teens authorize autonomous purchases? No. Stated interest does not establish a completed transaction or permission to spend. Keep product research and purchase approval distinct. ### What should a retailer improve in response? Make compatibility, bundle contents, selected-variant prices and required accessories clear. Those facts help families compare an offer accurately. ## Sources [Mastercard: Parent-and-teen research on AI shopping](https://www.mastercard.com/news/europe/en/newsroom/press-releases/en/2026/mastercard-report-predicts-that-one-in-10-people-will-routinely-use-ai-agents-to-shop-and-pay-by-2030/) --- # Visa survey finds AI assistant use ahead of payment trust Source: https://lantern.is/resources/visa-ai-assistant-payment-trust-survey Author: Lantern Briefing Published: 2026-09-13 Visa’s U.S. survey separates using an AI assistant from trusting one to pay. A checkout pilot should make the buyer’s approval clear. Using an AI assistant and trusting it with a payment are separate decisions. A Visa survey released on September 9, 2026, found that 72% of U.S. respondents had used an AI assistant. Only 23% said they trusted generative AI platforms to manage payments. [Visa’s research release](https://investor.visa.com/news/news-details/2026/New-Visa-Research-Finds-Consumer-Trust-is-Accelerating-the-Path-to-Agentic-Commerce/default.aspx) also reported 61% trust in Visa for that payment role. Visa sponsored the research. The comparison measures stated trust in named providers; it does not demonstrate that adding a payment brand to a checkout increases sales. The Harris Poll surveyed 2,065 U.S. adults from May 26 to May 28, 2026. The payment-trust questions used roughly half of that sample per brand. The assistant-use figure includes uses beyond shopping, so it should not be presented as the percentage of Americans who buy through AI. For a retailer evaluating an AI checkout, the practical issue is what a buyer is being asked to approve. “Continue with assistant” leaves too much unanswered. A useful purchase review identifies the merchant, exact item, quantity, delivery destination and final total before the buyer confirms. Suppose a customer orders a two-pack of replacement filters. A buyer should be able to see the compatible appliance model and whether the order is a single purchase or a recurring delivery. If the requested pack is unavailable, substituting another product or changing the quantity deserves a fresh review. An earlier instruction to find filters should not be treated as approval for an unspecified purchase. Have someone unfamiliar with the pilot read the approval screen and explain what will happen next. Can they identify the total charge and the product they will receive? Can they stop, correct a choice or return to the store’s normal checkout? That exercise can reveal confusing language before the team debates broader adoption forecasts. The order confirmation needs to carry the same clarity. Give the buyer an order reference and an obvious route to the merchant for delivery, cancellation or return questions. Our [guide to scoping an AI shopping pilot](https://lantern.is/resources/scope-ai-shopping-pilot-before-buying-software) helps define a bounded purchase journey before a wider rollout. ## Common questions ### Does the 72% figure mean most Americans shop through AI? No. Visa measured whether respondents had used an AI assistant, which includes activities outside shopping. ### Does Visa’s higher trust score prove a conversion benefit? No. The survey reports attitudes. A retailer needs evidence from its own checkout to assess effects on completed orders. ### What should a buyer approve before an AI-assisted purchase? The exact item, merchant, quantity, delivery details and final total should be clear. Recurring purchases and substitutions also need explicit treatment. ## Sources [Visa: Consumer trust and AI-assisted commerce survey](https://investor.visa.com/news/news-details/2026/New-Visa-Research-Finds-Consumer-Trust-is-Accelerating-the-Path-to-Agentic-Commerce/default.aspx) --- # Amazon Ads opens a U.S. pilot for advertising in ChatGPT Source: https://lantern.is/resources/amazon-ads-chatgpt-advertising-pilot Author: Lantern Briefing Published: 2026-09-11 Selected U.S. advertisers can extend Amazon Ads campaigns to ChatGPT in a pilot. Paid placements and organic AI recommendations remain separate. Amazon Ads announced a U.S. pilot for advertising in ChatGPT on September 10, 2026. The program allows selected advertisers to extend Amazon Ads campaigns to ChatGPT. For an eligible brand, it creates another paid placement to evaluate within a defined campaign. The [Amazon Ads announcement](https://advertising.amazon.com/library/news/amazon-ads-chat-gpt-advertising-integration) names Delta Vacations as a participant. It does not establish general availability for every advertiser or a performance benchmark that other brands can expect. Confirm eligibility, supported creative and available reporting with the campaign team before allocating budget. Paid placements also need to be distinguished from the assistant’s answer. [OpenAI’s published advertising principles](https://openai.com/index/our-approach-to-advertising-and-expanding-access/) say ads are separately labeled and do not influence ChatGPT’s answers. Buying an ad should therefore not be described as buying a product recommendation in the organic response. Write a pilot brief around one buying need. A travel-accessory retailer could test a compact charger compatible with a named device. Specify the product and variant, the claim the creative will make, the destination page, the spending limit and the purchase event the team will review. The landing page must support that claim. If the creative describes charging two devices at once, show the relevant ports, supported output and any required cable. If the pictured cable is sold separately, make that visible. The shopper should be able to resolve the buying question without searching through unrelated products. Ask which placement and conversion details the available reports actually provide. A campaign total that combines several destinations cannot, by itself, establish results for ChatGPT alone. Where reporting cannot separate the placement, keep that limitation in the campaign readout and avoid attaching a precise return to it. Review ad delivery, landing-page behavior and completed orders as different stages. A recorded impression establishes that an ad was served; it does not establish that the assistant endorsed the product. A click shows interest, while an order provides a different piece of evidence about the purchase journey. For the creative itself, use the [guide to writing ChatGPT ads for a specific buying need](https://lantern.is/resources/write-chatgpt-ads-specific-buying-need). The pilot should give the team a clear decision about the product, message and destination it tested. ## Common questions ### Can every Amazon Ads customer advertise in ChatGPT? The September announcement describes a pilot for selected U.S. advertisers. Confirm access before building a launch plan. ### Does a ChatGPT ad change the organic answer? OpenAI’s published principles say advertising does not influence ChatGPT’s answers. Ads are presented separately and labeled. ### What should a retailer agree before starting a pilot? Define the product, supported claim, landing page, spending limit and desired purchase outcome. Check whether reporting can identify results for the placement. ## Sources [Amazon Ads: ChatGPT advertising integration](https://advertising.amazon.com/library/news/amazon-ads-chat-gpt-advertising-integration) [OpenAI: Advertising principles for ChatGPT](https://openai.com/index/our-approach-to-advertising-and-expanding-access/) --- # UCP adds a payments council as AI checkout expands Source: https://lantern.is/resources/ucp-payments-council-ai-checkout Author: Lantern Briefing Published: 2026-09-10 UCP’s September payments council adds industry participation. Retailers still need to verify which checkout capabilities their chosen service supports. The Universal Commerce Protocol, or UCP, [announced its Payments Technical Council on September 2, 2026](https://ucp.dev/documentation/announcements/#september-2-2026-payments-technical-council-payments-tc-formed). Its eight inaugural members are Adyen, Ant International, Coinbase, Global Payments, Google, PayPal, Shopify and Stripe. The development adds a dedicated payments group to a standard used to connect shopping services with merchants. UCP’s [announcement history](https://ucp.dev/documentation/announcements/) distinguishes several groups. Amazon, Meta and Microsoft representatives joined the Tech Council in April. Stripe joined the Governing Council later that month, alongside permanent members Google and Shopify. A company’s participation identifies its role in the standards effort; it does not establish that every shopping service it operates offers the same checkout. For a retailer considering UCP, the next useful question is which parts of a purchase its chosen service can complete. The [UCP specification](https://ucp.dev/specification/overview/) describes capabilities including cart, checkout and order updates. A transaction uses capabilities supported by both the shopping service and the merchant. A broad statement that a service “supports UCP” therefore needs a more specific explanation before it becomes a launch requirement. Consider a shopper ordering a pair of shoes for delivery before a trip. A useful demonstration should show the selected size, stock status, final price and delivery option before purchase approval. Follow the resulting order into the retailer’s support process. The customer should have a clear way to find confirmation and request help. Make that demonstration part of the provider conversation. Ask which markets and payment options are available to your business, which purchase steps stay inside the assistant, and when the buyer is sent to your site. Record any unsupported step so merchandising, payments and customer service can plan around the actual experience. The council announcement gives teams a reason to follow the standard’s development. A launch decision still needs evidence from the buying journey the retailer intends to offer. The [AI shopping primer](/resources/agentic-commerce-primer) explains the distinction between a recommendation, a prepared cart and a completed purchase. ## Common questions ### Did Amazon, Meta and Microsoft join UCP’s Governing Council? The April announcement names their representatives as additions to the Tech Council. It separately records Stripe joining the Governing Council. ### Does UCP support mean every checkout feature is available? Available capabilities depend on what the shopping service and merchant both support. Confirm the specific purchase steps, payment options and markets with the provider. ### What should a retailer ask to see before a pilot? Follow one relevant product through variant selection, final pricing, delivery, approval and order confirmation. Include a customer-service lookup so the team can handle the purchase afterward. ## Sources - [UCP: council announcements and dates](https://ucp.dev/documentation/announcements/) - [UCP: capabilities and compatibility](https://ucp.dev/specification/overview/) --- # Instacart’s Clementine turns grocery requests into carts Source: https://lantern.is/resources/instacart-clementine-ai-grocery-carts Author: Lantern Briefing Published: 2026-09-09 Instacart’s September 9 launch makes store availability and pack size practical checks for brands reviewing AI-generated grocery carts. Instacart introduced Clementine on September 9, 2026, describing an assistant that turns a conversation, grocery list or recipe into a shopping cart. [The company says it is available across the U.S. and Canada on the Instacart Marketplace](https://company.instacart.com/pressreleases/meet-clementine-instacart-s-ai-shopping-assistant-that-takes-what-s-for-dinner-off-your-plate), using availability at the customer's chosen store. The same announcement distinguishes Clementine from Cart Assistant, Instacart's service for grocers' own websites and apps. Food Bazaar, Heritage Grocers Group and Woodman's were named as live Cart Assistant customers. A grocery brand evaluating distribution needs to identify which shopping experience it is actually reviewing. A recipe-to-cart journey creates a useful product-data test. For an illustrative request for 600 g of dry pasta sold in 500 g packs, one pack would be too little and two would meet the requested amount. Check whether the cart contains the right product and whether the assistant explains the extra quantity. The test connects the requested amount with the units the store actually sells. Use the same request at two stores that carry the brand, and keep the store location with each result. A missing item may reflect local availability or a different assortment. The brand needs that context before treating absence from a cart as evidence of a broader discovery problem. The product record should make comparisons possible. Check the pack size, unit of measure and variant against the actual packaging. If a multipack appears beside a single unit, confirm that the assistant's explanation gives the buyer enough information to compare the amounts they will receive. These checks create an actionable record for the retailer or catalog team: the request, selected store, intended item and observed cart. Fixing a specific mismatch is more useful than responding to a general claim that the brand is missing from AI shopping. ## Common questions ### Are Clementine and Cart Assistant the same service? Clementine serves shoppers on the Instacart Marketplace. Cart Assistant is the related offering for a grocer's own branded website or app. ### Why should a brand check the same request at different stores? Availability and assortment can differ by store. Retaining the location with each result helps the team distinguish a local stock issue from a recurring product-information problem. ### What should a grocery brand verify first? Check whether the cart contains the right variant and quantity for the request. Retain the store location and product details so any mismatch can be reproduced. ## Sources - [Instacart: Clementine launch and Cart Assistant availability, September 9, 2026](https://company.instacart.com/pressreleases/meet-clementine-instacart-s-ai-shopping-assistant-that-takes-what-s-for-dinner-off-your-plate) --- # Bain: 24% of online holiday shoppers plan to start with AI in 2026 Source: https://lantern.is/resources/bain-holiday-2026-ai-shopping-forecast Author: Lantern Briefing Published: 2026-09-08 Bain’s holiday survey points to more AI-assisted product discovery. Gift pages need clear answers about suitability, price and delivery. Bain & Company expects more online holiday shoppers to begin with AI in 2026. In its September 3 forecast, 24% planned to start their shopping with an AI tool, compared with 17% in 2025. For retailers preparing gift pages, the finding makes the quality of the first product comparison worth attention. [Bain’s forecast](https://www.prnewswire.com/news-releases/us-holiday-retail-sales-set-to-outpace-last-years-seasonal-growth-performance-to-exceed-1-trillion-for-the-first-timebain--company-forecasts-302868926.html) draws on Consumer Lab research involving more than 1,100 consumers. The AI figure describes shopping intentions. It does not mean that AI will complete 24% of holiday orders or generate that share of retail revenue. The same release forecasts 4.5% growth in U.S. retail sales during November and December, with higher prices accounting for more than half of that growth. That distinction matters when setting expectations: a larger dollar total does not imply an equivalent increase in the number of products sold. A useful holiday page should answer the constraints in a real gift request. Take a gift request such as: “What is a good gift under $75 for someone who runs before sunrise?” A shopper comparing a reflective vest with a rechargeable light needs more than a seasonal headline. They need the current price, available sizes, battery or charging details where relevant, and a clear description of what the item does. Build the answer from verified product facts. If a light includes a cable but requires a separate charger, state that. If a vest comes in multiple fits, link the size guide beside the selection. Avoid broad claims such as “perfect for every runner” when the information needed to judge suitability is available. Delivery information needs similar care. Show how a customer can check the arrival estimate for their destination and chosen shipping service. A universal holiday cutoff can mislead shoppers when availability or location changes the promise. Keep gift-return terms easy to find, especially where they differ from the standard policy. Start with products that already receive gift-related questions or holiday traffic. Improve the missing answers on those pages, then follow a product link from the recommendation through to the selected variant. The [AI referral landing-page guide](https://lantern.is/resources/ai-referred-shoppers-product-landing-page) explains how to preserve the buying context when a shopper reaches the store. ## Common questions ### Does Bain’s 24% figure represent completed AI purchases? No. It describes the share of online holiday shoppers who planned to begin with AI. An intention to use a tool does not establish a completed order. ### Which gift-page information should a retailer review first? Check current price, suitability, available variants, included accessories, delivery estimates and gift-return terms. Prioritize the details customers need to rule a product in or out. ### Should retailers rebuild their entire holiday catalog around this survey? Use it to inform focused improvements. Your own product questions, traffic and orders should determine which pages receive the most work. ## Sources [Bain & Company: 2026 U.S. holiday retail forecast](https://www.prnewswire.com/news-releases/us-holiday-retail-sales-set-to-outpace-last-years-seasonal-growth-performance-to-exceed-1-trillion-for-the-first-timebain--company-forecasts-302868926.html) --- # Walmart reports 35% higher order values among Sparky users Source: https://lantern.is/resources/walmart-sparky-order-values-complete-basket Author: Lantern Briefing Published: 2026-09-07 Walmart’s reported Sparky order values raise a practical question for retailers: can an AI assistant help shoppers assemble a useful, complete basket? Walmart said customers using Sparky, its AI shopping assistant, had average order values about 35% higher than customers who did not use it. CEO John Furner reiterated the comparison on the company's [May 21, 2026 earnings call](https://corporate.walmart.com/content/dam/corporate/documents/newsroom/2026/05/21/walmart-releases-q1-fy27-earnings/q1-fy27-earnings-call-transcript.pdf). That is a comparison between customer groups. It does not, by itself, establish that Sparky caused the difference or that another retailer should expect the same result. For merchants considering an assistant, the useful question is whether it helps customers assemble a basket that meets their needs. [Walmart describes Sparky](https://corporate.walmart.com/about/technology) as an assistant in its app that helps people compare products, build lists and plan for occasions, using product information and reviews. Those tasks offer a concrete starting point for a retail pilot: helping someone buy the items needed for a particular activity. Consider an illustrative request to set a dinner table for eight people. The customer already owns eight dinner plates and needs glasses, napkins and a tablecloth for a specified table size. A useful assistant would respect what the customer owns, check pack quantities and help them choose a tablecloth that fits. Recommending another dinnerware set would increase the basket while ignoring the request. The catalog needs to answer those practical questions. A glassware listing should distinguish a single glass from a set and state how many pieces are included. A tablecloth should give its actual dimensions and shape. Product images can show the setting, but the text should make clear which photographed accessories are sold separately. Fulfillment also matters to a basket assembled around an occasion. Ask whether all the selected items can reach the customer's location before the event. If an item is unavailable, a replacement should preserve the important requirement, such as enough glasses for eight guests. The shopper should see any changed price, quantity or delivery promise before accepting it. A retailer can use this dinner-table task to review an assistant before a wider rollout. Give the same stated needs to a merchandiser and check the proposed basket against the catalog. Look for duplicate items, incorrect pack sizes, products that do not fit and delivery dates that miss the occasion. Keep the shopper's budget visible throughout the review. Judge the eventual orders alongside returns, cancellations and support questions. A larger basket is valuable when the customer receives useful products and the order remains profitable. Walmart's reported result is a reason to investigate that possibility, not a revenue forecast for an untested assistant. ## Common questions ### Did Walmart say Sparky increased every customer's spending by 35%? No. Walmart reported average order values about 35% higher among Sparky users than nonusers. The statement does not establish an individual customer's increase or isolate the assistant's effect. ### What should a retailer check before offering basket recommendations? Check quantities, compatibility, the items the shopper already owns, total cost and delivery timing. A complete basket should satisfy the stated task without unnecessary additions. ### Which result matters beyond order value? Whether the customer receives the right items and keeps them. Review cancellations, returns and support issues alongside sales and the cost of serving the order. ## Sources - [Walmart: May 21, 2026 earnings call transcript, page 4](https://corporate.walmart.com/content/dam/corporate/documents/newsroom/2026/05/21/walmart-releases-q1-fy27-earnings/q1-fy27-earnings-call-transcript.pdf) - [Walmart: Current AI shopping capabilities](https://corporate.walmart.com/about/technology) --- # What AI-referred shoppers need from a product landing page Source: https://lantern.is/resources/ai-referred-shoppers-product-landing-page Author: Lantern Briefing Published: 2026-09-06 Review the journey from an AI recommendation to the exact product, variant and available offer. Includes checks for stale links, missing details and substitutions. A shopper arriving from an AI recommendation needs to verify the product, choose the right version and understand the offer. The landing page should make those decisions possible even when the shopper has never visited the retailer before. The channel deserves attention. In an [April 16, 2026 report](https://business.adobe.com/blog/ai-traffic-surge-retail-sites-not-machine-readable), Adobe said AI-referred traffic to U.S. retail sites grew 393% year over year during January through March. For March, Adobe reported a purchase conversion rate 42% higher for AI traffic than for non-AI traffic in its data. Those findings describe Adobe's observed traffic. They do not establish that readable pages caused the difference or that every retailer will see similar results. Use your own incoming journeys to identify what a visitor can actually buy. ## Follow the link a shopper receives Where you have an example of an AI recommendation, save the product link and the specific product claim for review. Open that link directly on a phone and a desktop, including while signed out. Check the destination after any country selector or redirect. Use this illustrative example: an assistant recommends a laptop bag for a particular laptop model and links to the retailer. The page opens on the smaller bag size. The shopper now needs to know whether the recommended fit applies to the selected variant. A general 'fits laptops' description cannot settle that question. Show the usable compartment dimensions, any verified model compatibility and the selected size. Keep that information close to the variant selector. If compatibility has not been established, say what dimensions the customer should compare rather than inventing a fit claim. ## Resolve the mismatch at the point of arrival | What the visitor encounters | What to improve | | --- | --- | | The link opens a different size or color. | Preserve the linked variant where possible and make the selected version clear. | | The recommendation names an older model. | Identify that model and its availability. Label any replacement as a different product and explain relevant differences. | | The product cannot ship to the visitor's location. | Show regional availability and delivery limits before asking the visitor to complete checkout. | | The recommended variant is sold out. | Show its status and distinguish available alternatives by size, features and price. | | The assistant makes a claim the page cannot support. | Verify the claim against product records. Add accurate facts and limitations to the page without adopting an unsupported claim. | Avoid sending every unavailable product link to the homepage. A visitor who was comparing a specific bag should be able to understand what happened to it and assess the nearest relevant alternatives. ## Make the answer available on the page Put decision-critical details in readable text, supported by useful images. A size diagram helps, but the measurements also need labels and units that a shopper can find and compare. Keep the selected variant's price, stock status and specifications consistent. [Google's guidance for AI features in Search](https://developers.google.com/search/docs/appearance/ai-features) recommends making important content available as text and keeping structured data consistent with the visible page. It also says there is no special AI schema or additional technical requirement for AI Overviews and AI Mode. That guidance applies to Google's Search features; other services have their own access and product-data requirements. ## Assign the fix to the team that controls it For each failed arrival, record the incoming URL, the final page, the selected variant and the unanswered customer question. Send incorrect product facts to the catalog owner, broken routing to the web team and unclear delivery terms to the team responsible for the offer. Open the same link again after the fix. The practical acceptance check is whether a new visitor can confirm the item, understand any limitation and reach the intended product or a clearly labeled alternative. ## Sources - [Adobe: April 2026 U.S. retail AI traffic report](https://business.adobe.com/blog/ai-traffic-surge-retail-sites-not-machine-readable) - [Google Search Central: AI features and your website](https://developers.google.com/search/docs/appearance/ai-features) --- # AWS adds controls for retailers running their own shopping assistants Source: https://lantern.is/resources/aws-shopping-assistant-retailer-controls Author: Lantern Briefing Published: 2026-09-06 AWS’s September 26 update adds configuration history, catalog controls and cart features, giving retailers specific operating capabilities to evaluate. AWS added features to its Agentic Shopping Assistant on September 26, 2026, including a portal for managing catalog information, store policies and the instructions that shape answers. [The update](https://aws.amazon.com/blogs/industries/introducing-seven-new-features-of-the-agentic-shopping-assistant-on-aws/) includes version history and a restore control for configuration changes. For a retailer, these controls address a recurring operating question: who can correct an assistant's answer when the underlying offer changes? An internal owner needs to know where the answer came from, how to update the relevant information and how to verify the result. A return-policy change makes a useful demonstration. Ask the implementation team to show the answer before the change, update the applicable policy, and repeat the question. Then establish which version the assistant used and whether restoring an earlier configuration also restores the intended behavior. This separates the presence of a restore button from evidence that the team's recovery process works. AWS also describes cart management, memory across visits and a Shopify connection through the Universal Commerce Protocol. It says the solution runs in the retailer's AWS account and that customers pay for the underlying services. Those statements describe the operating model; a budget still needs staffing, integration and ongoing service usage. Cart behavior deserves a separate acceptance test. Start with an item already in the cart, ask for a different size, then inspect the final line items and quantities. Revisit the conversation to check how a previous preference affects the next recommendation. Each result should be tied to the configuration used so the team can reproduce a problem. Retailers evaluating the offering can use these demonstrations to assign responsibility before rollout. The merchandising owner should be able to maintain the commercial information, while the operating team should be able to trace and correct an error. Those capabilities make the assistant's ongoing upkeep assessable. ## Common questions ### What changed in the September 26 AWS update? AWS announced controls for managing the assistant, alongside additional shopping and cart features. The portal includes configuration history and a restore function. ### Does paying only for AWS services cover the full operating cost? A retailer also needs to account for implementation and staff time. Estimate those costs with actual usage before comparing the offering with another service. ### What should a retailer ask to see in a demonstration? Follow a policy change from its source to the answer, and test a cart with an existing item. Keep the results associated with the configuration that produced them. ## Sources - [AWS: Shopping assistant features and operating model, September 26, 2026](https://aws.amazon.com/blogs/industries/introducing-seven-new-features-of-the-agentic-shopping-assistant-on-aws/) --- # How to scope an AI shopping pilot before buying software Source: https://lantern.is/resources/scope-ai-shopping-pilot-before-buying-software Author: Lantern Briefing Published: 2026-09-05 Define the shopping task, required evidence, ownership and cost before choosing AI commerce software. Includes a worked pilot brief and vendor demo questions. Before choosing AI shopping software, write down the customer problem, the change you want to make and the evidence that would justify continuing. A useful pilot gives the team a specific decision to make at the end. ## Choose the problem your team can act on Start with a concrete failure. A product page may omit a compatibility fact. An assistant on your store may recommend the wrong accessory. A checkout service may fail to carry the customer's chosen variant into the order. Each problem has a different owner and requires a different kind of work. Ask who can make the change. A tool that reports an incorrect answer in an external assistant still needs a team to investigate the source and update information it controls. A store assistant needs access to trustworthy catalog data. A checkout pilot needs an order owner who can verify what was authorized and fulfilled. For work aimed at Google's AI Overviews and AI Mode, [Google says existing SEO practices remain relevant and no special AI markup is required](https://developers.google.com/search/docs/appearance/ai-features). Ask a proposed supplier to identify the specific problem and deliverable before accepting a requirement to add a new file or change the site. ## Write a pilot brief that a supplier can demonstrate The following is an illustrative brief for a retailer testing product compatibility assistance on its own store. The scope is deliberately small enough for its merchandiser to inspect each recommendation. Choose a scope that your team can actually review. | Brief field | Worked example | | --- | --- | | Customer task | Find a replacement filter for a specified vacuum model. | | Starting problem | The shopper cannot tell which current filters fit the vacuum they own. | | Pilot scope | One filter category on the store, covering current products and a selection of older vacuum models. | | Required information | Verified compatibility records, product identifiers, pack quantities, prices, stock and product-page links. | | Expected result | Name a compatible, available filter and show the supporting product information. Ask for the model when it is missing. | | Unknown or unsupported fit | State that compatibility cannot be confirmed and offer the existing support route. | | Permitted actions | Recommend and link to products. The shopper makes the selection and completes the store's normal checkout. | | Owner and review | A merchandising owner checks compatibility; the web owner handles the customer journey. Agree a review date before starting. | | Evidence for the decision | Examples of correct recommendations, unresolved questions, completed purchases and compatibility-related support or returns. | Keep the brief in the same document as the demo notes. When a supplier proposes an additional capability, record which customer problem it solves, what information it needs and who would maintain it. ## Make the demo answer your difficult cases Give each supplier the same approved product records and shopping tasks. Include a missing model number, an unsupported model, an unavailable filter and a multipack that could be mistaken for a single item. Ask the supplier to show the actual customer response and destination page. An acceptable response may be a clarifying question or an admission that compatibility is unknown. A confident recommendation without supporting product information should prompt investigation before a customer sees it. Then change one source record, such as an item's availability, and ask the supplier to demonstrate how the change reaches the customer experience. Agree who corrects errors and how the team can pause the feature while a problem is resolved. ## Include the work your team will have to do Request a cost estimate covering setup, ongoing usage, catalog maintenance and staff review. Ask what your team can export if it stops using the service, including any corrected product records and its own review notes. Confirm who controls access to the store and how that access ends. At the review date, compare the evidence with the original customer task. Continue when the experience helps shoppers find the right item and the team can operate it at an acceptable cost. Narrow or stop the pilot when unresolved product errors or ongoing work outweigh that benefit. ## Source - [Google Search Central: AI features and your website](https://developers.google.com/search/docs/appearance/ai-features) --- # How to write ChatGPT ads for a specific buying need Source: https://lantern.is/resources/write-chatgpt-ads-specific-buying-need Author: Lantern Briefing Published: 2026-09-05 Build ChatGPT ads around a concrete customer need, support each claim on the landing page, and use distinct creative angles. Includes a worked retail example. A useful ChatGPT ad should help someone decide whether a product meets a specific need. Start with the customer's constraint, put the relevant fact in the headline, and send the click to a page that proves it. [OpenAI's creative guidance](https://help.openai.com/en/articles/20001212-create-ads-for-chatgpt-ads) favors specific benefits, distinct creative variations and relevant destinations. Its [advertiser documentation](https://help.openai.com/en/articles/20001207-ads-in-chatgpt-the-basics) describes ads below ChatGPT responses and allows context hints about relevant topics. Those hints do not guarantee placement in a particular conversation. ## Start with a constraint you can answer Consider an illustrative furniture campaign for someone who needs a bookcase narrower than 80 cm. Assume the retailer has verified that one model is 78 cm wide when assembled, with adjustable shelves. The ad can answer that width constraint directly. Headline: 78 cm-wide bookcase for narrow spaces. Description: Adjustable shelves for books and storage. See full dimensions, finishes and delivery options. The width matters because it helps the customer rule the product in or out. A headline such as 'Transform your home with beautiful storage' leaves that decision unresolved. Before using the example, replace every detail with verified facts about the exact product being advertised. ## Make the landing page prove the claim Send this ad to the specified bookcase. The customer should see the same model and finish pictured in the creative, with dimensions labeled as assembled dimensions. A packaging measurement would answer a different question. | Customer question | Evidence to show on the page | | --- | --- | | Will it fit the available space? | Assembled width, height and depth, with units and a clear dimension diagram. | | Will my books fit on the shelves? | Usable shelf dimensions and the product's verified shelf load limits. | | What will I receive? | The selected finish, included parts, assembly information and any required wall anchoring. | | Can I get it when I need it? | Availability, delivery options and costs for the customer's location. | Ask a colleague to open the ad's destination on a phone and choose the pictured finish. If the page switches to a different size, hides the dimensions in an unrelated download or sends them to a broad furniture category, fix that path before spending on the ad. ## Give each creative a different reason to matter Build another variation only when there is another supported buying need. A verified shelf-height range could support a message about fitting tall books. A documented finish could support a message about matching existing furniture. Replacing 'beautiful' with 'stylish' does not test a different product benefit. For each variation, keep a short record of the intended customer need, the exact claim and the page section that supports it. That record gives the merchandiser and campaign owner a shared review brief. It also makes a product change, such as a revised size or finish, easier to carry through to the ads. ## Keep paid placement separate from recommendations [OpenAI states that ads are labeled and separate from its answers](https://openai.com/index/our-approach-to-advertising-and-expanding-access/). An ad impression is evidence of paid delivery. It does not show that the assistant independently recommended the product. Review what happens after the click: whether the visitor reaches the advertised item, can select it and can complete the intended next step. If people arrive but cannot find the size they came for, changing the headline again will leave the underlying problem in place. ## Sources - [OpenAI: Create Ads for ChatGPT Ads](https://help.openai.com/en/articles/20001212-create-ads-for-chatgpt-ads) - [OpenAI: Ads in ChatGPT, The Basics](https://help.openai.com/en/articles/20001207-ads-in-chatgpt-the-basics) - [OpenAI: Our approach to advertising](https://openai.com/index/our-approach-to-advertising-and-expanding-access/) --- # What repeated AI answers reveal about brand visibility Source: https://lantern.is/resources/sampling-on-the-bubble Author: Eric Taurone Published: 2026-09-04 Explore 91 recorded answers to see how brand appearances vary and why useful visibility checks need more than a single response. A brand can appear in several answers to the same shopping question and disappear from another. Repetition helps show that variation, but the design of the test determines what the observations can tell you. ## What the historical example shows This historical chart from Lantern’s July 2026 repeated-question experiment shows one brand and one question in an OpenAI example. Each column represents a collection morning. Across 91 displayed observations, the brand appears in 73 answers and is absent from 18. The observations include ranks from 1 to 18. One column contains ranks 2 and 18, showing how far the reported position can vary within a collection morning. This describes the recorded example; it does not establish that the prompt was the only influence on the answer or that all underlying conditions stayed fixed. Dots show recorded ranks and crosses show completed answers that omitted the brand. Eighteen columns contain five observations each; column 10 contains one. The four missing observations in that column have unknown outcomes. ## The observations behind the chart The table lists the observations shown in the chart. Dates appear only where the chart provides a label. | Column | Original date label | Ranks when present | Recorded absences | Observations shown | | --- | --- | --- | --- | --- | | 1 | Jul 4 | 1, 2, 7, 1, 3 | 0 | 5 | | 2 | Not labeled | 2, 1, 1, 2, 1 | 0 | 5 | | 3 | Not labeled | 2, 1 | 3 | 5 | | 4 | Not labeled | 6, 1, 1, 1, 1 | 0 | 5 | | 5 | Jul 9 | 1, 1, 11 | 2 | 5 | | 6 | Not labeled | 1, 1, 3, 1, 1 | 0 | 5 | | 7 | Not labeled | 3, 4, 1, 2 | 1 | 5 | | 8 | Not labeled | 5, 15, 5 | 2 | 5 | | 9 | Jul 14 | 9, 2, 13 | 2 | 5 | | 10 | Not labeled | 1 | 0 | 1 | | 11 | Not labeled | 12, 7, 2, 9, 7 | 0 | 5 | | 12 | Not labeled | 13, 7, 2, 18, 2 | 0 | 5 | | 13 | Jul 18 | 12, 12, 7, 2 | 1 | 5 | | 14 | Not labeled | 14, 12, 3, 2, 7 | 0 | 5 | | 15 | Not labeled | 9, 2, 1 | 2 | 5 | | 16 | Not labeled | 13, 10, 12, 14 | 1 | 5 | | 17 | Jul 22 | 10, 11, 16, 10, 7 | 0 | 5 | | 18 | Not labeled | 16, 14, 10 | 2 | 5 | | 19 | Not labeled | 9, 10, 1 | 2 | 5 | ## What this example cannot establish The chart does not disclose the exact prompt, model identifier, search settings or full collection log. It is an anonymous historical illustration, not a reproducible provider benchmark. It cannot establish how ChatGPT's consumer interface, another model, a different query or a different month will behave. It also cannot tell us which part of a change came from generation, retrieved information, product availability or another condition. A repeated prompt controls the wording. It does not control every input to the assistant. Answers collected close together can share conditions, such as a source being unavailable throughout a session. Spread repeat checks across time as well as within a sitting so one temporary condition does not define your view of the brand. ## A practical sampling plan 1. Choose a fixed set of buyer questions and write down the decision the test should inform. Keep exploratory questions separate from the set used for trend comparisons. 2. Record the assistant, model when exposed, search mode, locale, date, time and conversation setup. Save the full answer and cited links so claims can be checked. 3. Collect some repeated observations within sessions and some across days. This separates variation within a sitting from variation over time without assuming a universal best interval. 4. Distinguish answers that omit the brand from requests that fail or produce no answer. Keep gaps in the available evidence visible. 5. Compare equivalent prompts and settings. If either changes, report the changed cohort separately before making a trend claim. 6. Choose the next sample size based on the decision and the observed variability. A small diagnostic sample can reveal a problem; a reliable estimate of a small change generally requires more evidence. For routine monitoring, choose a schedule the team can sustain and review results more closely around launches, product changes and unexplained shifts. For a shorter explanation of interpreting repeat checks, read [Why one AI answer is never enough](/resources/why-one-ai-answer-is-never-enough). ## Common questions ### How often should a brand check its AI visibility? There is no universal interval established by this example. Choose a schedule around the decision, collect observations across time, and adjust when the data or business context justify it. ### Are five questions in one sitting five independent observations? Not necessarily. They share some conditions, and a count of five alone does not establish independence. Keep session and date information so the analysis can account for that relationship. ### Does the missing data in column 10 count as four omissions? No. The chart shows only one observation for that column and does not establish what happened to the other expected records. Missing observations and completed answers without the brand are different states. ### Can this chart establish the best provider for brand discovery? No. It is one historical, anonymous example from one provider. A comparison would need comparable prompts, modes, periods, sample sizes and a documented method across the providers being compared. --- # Mastercard adds 22 startups to its AI commerce program Source: https://lantern.is/resources/mastercard-ai-commerce-cohort-order-verification Author: Lantern Briefing Published: 2026-09-04 Mastercard’s September cohort spans payments, identity and merchant connections. Retailers should test how buyer approval connects to a fulfilled order. Mastercard’s [September 2026 Start Path update](https://www.mastercard.com/us/en/innovation/partner-with-us/start-path/news.html) names 22 companies in its inaugural Agentic Commerce & Services program. The cohort covers work such as payment credentials, transaction verification and connections between merchants and AI services. For retailers, those capabilities meet at a practical task: accepting an order that can be traced back to the buyer’s decision. Mastercard’s [Agent Pay description](https://www.mastercard.com/us/en/business/artificial-intelligence/mastercard-agent-pay.html) emphasizes registered agents and purchases backed by authenticated intent and explicit consent. Those are useful questions to bring to a payment-provider discussion. A retailer needs to understand what evidence reaches its own systems and which team can retrieve it when a customer asks about an order. Consider an illustrative purchase of a coffee machine. The buyer approves a particular model and a final total. In a provider demonstration, ask to follow that approval into the payment record, the retailer’s order and the customer’s confirmation. Check that all four refer to the same item and amount. Then introduce a realistic change: the selected model becomes unavailable, or a different delivery option changes the total. Ask the provider to show how the customer is informed and what approval is required before the purchase proceeds. Keep the expected behavior in the pilot brief so the team can distinguish a supported flow from a feature still being developed. Customer service needs a usable record too. Give the support team the confirmation available to the buyer and ask it to locate the order, identify the product and explain the next step for a return or refund. Establish which provider handles each part of that request. The applicable payment and merchant terms should govern the actual process. Use the cohort announcement as a starting point for investigating a specific capability. Participation in a startup program alone does not establish that a service supports your market, payment provider or order system. Ask for evidence of the exact connection the business needs, together with a contact responsible for resolving failed transactions. The useful outcome of a pilot is a purchase the buyer can recognize and the retailer can fulfill and support. The [AI shopping primer](/resources/agentic-commerce-primer) places those checks within the wider discovery-to-purchase journey. ## Common questions ### What did Mastercard announce? Its September Start Path update lists 22 participants in a new program covering AI commerce and services. ### Does joining the cohort make a service ready for every retailer? A retailer still needs to verify availability and support for its payment provider, market and order system with the service itself. ### What should a merchant verify in an AI-assisted purchase? Check that buyer approval, payment, order and confirmation agree on the item and amount. Include a change to the offer and a support lookup in the demonstration. ## Sources - [Mastercard: Start Path’s September 2026 cohort](https://www.mastercard.com/us/en/innovation/partner-with-us/start-path/news.html) - [Mastercard: Agent Pay and buyer consent](https://www.mastercard.com/us/en/business/artificial-intelligence/mastercard-agent-pay.html) --- # Why one AI answer is never enough to measure brand visibility Source: https://lantern.is/resources/why-one-ai-answer-is-never-enough Author: Eric Taurone Published: 2026-09-02 A single recommendation is one observation. Learn how to compare repeated AI answers without confusing an omission, a missing result or a changed sample. An AI assistant can name a brand in one answer and leave it out of the next, even when you submit the same prompt. That makes a single answer useful evidence of what happened once. It takes a set of comparable answers to describe how often it happens. The prompt is only part of the situation. The assistant's model, search mode, conversation history, location, retrieved sources and available products can also affect an answer. Keeping the wording unchanged does not establish that all those conditions stayed fixed. Start with a question a real buyer would ask. Keep the exact wording and record the assistant, mode, date and relevant context alongside the full response. Use a fresh conversation when that is part of the test design. Separate a completed answer that omits the brand from a failed request or a run that produced no answer. Then report a numerator and denominator. In an illustrative sample, eight brand appearances in 20 valid answers is a 40% observed appearance rate. Twelve in a comparable sample of 20 is 60%, a difference of 20 percentage points. Those figures describe the samples; they do not by themselves prove that a site change caused an improvement or that the underlying rate is now exactly 60%. Repeated questions in one sitting can reveal variation within that sitting. Observations across days can reveal variation over time. Both are useful, but they answer different questions, and closely related observations may provide less independent information than their count suggests. There is no universal number of repetitions that makes every comparison reliable. The [sampling example](/resources/sampling-on-the-bubble) shows this distinction using a historical set of recorded answers. Its practical lesson is to retain the observations and their context, then make the strength of the conclusion match the evidence. A useful update to a colleague sounds like this: “The brand appeared in 12 of 20 answers this week, compared with eight of 20 under the same test setup last week. We will check whether the pattern persists.” That is a claim someone else can inspect and a sensible basis for the next decision. ## Common questions ### My brand appears in half the answers. Did visibility drop? That result alone cannot tell you. Compare it with an earlier sample collected under equivalent conditions, and keep the sample sizes and uncertainty in view. ### Does repeating the same question teach us anything? Yes. It can show how variable answers are within the conditions you tested. It does not automatically establish how answers vary across other days, users or assistant modes. ### Can a brand be permanently in or out of an assistant's answers? An observed streak does not establish permanence. Sources, products, models and user context can change, so describe the period and conditions in which the streak occurred. ### Should a failed request count as a missing brand? No. A completed answer that leaves the brand out is an observed omission; a failed request provides no answer to evaluate. Record the failure separately so it does not silently become a zero. --- # How ChatGPT finds information for brand recommendations Source: https://lantern.is/resources/how-chatgpt-decides-which-brands-to-recommend Author: Andrew Lissimore Published: 2026-07-02 Andrew Lissimore explains how learned knowledge, live retrieval and product feeds shape shopping answers, and what a merchant can actually verify. Not long after ChatGPT launched, I asked it which headphones to buy. I had spent years building Headphones.com, and the model never brought us up. In that early experience there was no live web search to inspect. A familiar way of measuring discovery had stopped being enough. The useful distinction for an operator is between what a model has learned and the information an assistant retrieves while answering. It is a simplified way to organize the work, not a complete description of a recommendation system. Shopping answers can also use structured product data and the buyer's current request and context. OpenAI documents [merchant product feeds](https://developers.openai.com/commerce/specs) as an input for its shopping experiences. Audio taught me why the public record matters long before AI existed. Head-Fi threads, subreddit discussions and reviewers helped buyers decide who was credible. At Headphones.com, reputation accumulated through the products we sold, the advice we gave and the policies we stood behind. That material can be useful to an assistant when it is available in the sources the system uses. We cannot inspect a model and establish that it absorbed every thread, review or article. Retrieval creates a different, more immediate check: what information can the assistant reach now? Some answers use live search, but that behavior depends on the product and mode. Perplexity, for example, documents [controls that let a user turn web sources on or off](https://www.perplexity.ai/enterprise/videos/an-introduction-to-queries). A result collected with one setting should not stand in for every experience. The traffic from these systems is observable even though a recommendation's causes are not always visible. Adobe reported a [1,200% increase in traffic from generative AI sources to US retail websites between July 2024 and February 2025](https://blog.adobe.com/en/publish/2025/03/17/adobe-analytics-traffic-to-us-retail-websites-from-generative-ai-sources-jumps-1200-percent). That is a historical growth comparison over a specific window, not a forecast for an individual store or a measure of every AI-assisted purchase. For a merchant, access is an actionable starting point. Inspect the page a shopper or search system would receive, and check the actual crawler rules. OpenAI distinguishes [OAI-SearchBot for search from GPTBot for model-training purposes](https://developers.openai.com/api/docs/bots). Cloudflare likewise documents [controls by bot purpose](https://developers.cloudflare.com/bots/additional-configurations/block-ai-bots/). Review the intended search access separately from the training policy instead of assuming one blanket setting describes both. The accessible page then has to answer the buyer's question. Consider an illustrative listing for closed-back wired headphones. A shopper trying to avoid disturbing colleagues needs facts about the design, connection and intended use, plus evidence for any claim about sound leakage. A memorable model name alone does not establish suitability. Clear specifications help a reader assess the product, but they do not guarantee that an assistant will choose it. Check consistency across the visible page, its structured data and any feed supplied to a shopping platform. A correct description paired with an outdated price or unavailable variant can still produce a poor buying experience. Follow the exact item through the purchasing path instead of stopping at a screenshot of a recommendation. Press can contribute useful independent evidence, too. A published story may be retrievable if the assistant's search system can access and select it. Publication does not guarantee inclusion in a future training set or a place in an answer. I would judge coverage first by what it establishes for a reader and whether it reaches the people the business serves. To see where you stand, choose a small set of real buying questions with budgets and use cases. One in audio might be “closed-back headphones under $500 for an office.” Ask under recorded conditions, keep the complete answers and cited pages, and repeat across comparable sessions and days. Record both appearances and omissions. The [sampling guide](/resources/sampling-on-the-bubble) explains why one good answer is not a stable visibility result. Lantern grew out of wanting a way to inspect that changing picture. The useful work is to see how answers describe the brand, identify evidence gaps and test specific improvements. A recommendation tool can help prioritize the investigation; an observed change still needs to be checked against the same measurement conditions. I expect assistants to take on more of the buying process. The timing and the winning interfaces remain uncertain. What a merchant can do now is concrete: make the product understandable, keep the offer accurate, build a credible public record and retain evidence of what the assistants actually answered. ## Common questions ### Do recommendations come from training data or live information? Both can contribute, and some shopping experiences also use structured product feeds. The inputs vary with the assistant, feature and request, so the training-versus-retrieval distinction is a useful starting point rather than an exhaustive mechanism. ### Does blocking GPTBot remove a site from ChatGPT search? OpenAI documents separate controls for training and search crawlers. Review the named user agent and current documentation rather than assuming that a training restriction automatically prevents search discovery. ### Why can a brand rank on Google and still be absent from an AI answer? Different systems can use different sources, context and selection processes. Access, product facts and cited evidence are useful things to inspect, but an omission alone does not identify its cause. ### How long does a product-data fix take to improve visibility? There is no dependable universal timetable or guarantee. Verify that the corrected information is available first, then compare repeated observations under consistent conditions before attributing a change to the fix. --- # How to read Google Search Console's AI performance reports Source: https://lantern.is/resources/google-search-console-ai-reports Author: Eric Taurone Published: 2026-06-05 Understand what Google's AI reports measure, how Search differs from Discover, and why page impressions are different from brand mentions or sales. Google Search Console's AI performance reports help site owners inspect how their pages appear in Google's AI experiences. Google introduced the reports on June 3, 2026, and [announced global availability on August 31](https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports). The first distinction is the surface being measured. The [Search report](https://support.google.com/webmasters/answer/16984139?hl=en) covers AI Overviews and AI Mode in Google Search. [Discover has a separate report](https://support.google.com/webmasters/answer/16983858?hl=en). Neither is a general report of every answer in the Gemini app or of recommendations from other assistants. An impression is a visibility event defined by the report. It is not, by itself, evidence that the answer recommended the brand, that someone visited the page, or that a purchase followed. Read the current metric definitions before combining the report with web analytics: the two systems observe different stages of a journey. A useful first review compares the same pages over equivalent periods, with the same report and filters. Keep the period, surface, page and available dimensions with each observation. Note any changes to page availability, indexing, product assortment or measurement coverage before treating a difference as an editorial effect. For an illustrative example, a product page might receive 120 reported impressions in one period and 180 in the next. That is an observed increase of 60 impressions, or 50% relative to the first period. It does not establish 60 additional shoppers, 60 brand recommendations or any number of orders. Those require separate evidence, and the counts alone do not identify what caused the change. Use the report to decide where to investigate. If a page appears more often, inspect whether it answers the relevant buyer question accurately and whether the associated visit and purchase data show a corresponding change. If a report is absent, check eligibility and available data before concluding that the site has no AI visibility. Google's [guidance for AI features](https://developers.google.com/search/docs/appearance/ai-features) points site owners to established search fundamentals. Keep pages accessible, use useful internal links and make structured data agree with visible content. Treat the reporting as evidence about a defined Google surface, then use separate, clearly labeled measurements for other assistants. ## Common questions ### Does this include the Gemini app? The Search report describes AI Overviews and AI Mode in Google Search. It should not be relabeled as a measurement of the Gemini app, ChatGPT, Claude or every Google AI product. ### Why might the report be missing after a global rollout? Availability of the feature does not mean every property has enough eligible impressions to display a report. Check Google's current help documentation and the property's available data before interpreting an absent report. ### Can these reports prove that AI generated revenue? An impression count cannot establish an order or its cause. Compare it with separately defined referral, visit and purchase measurements, keeping the attribution method and its gaps visible. ### Should a page be rewritten whenever impressions fall? First check whether the comparison uses equivalent periods, pages and filters, and whether there were availability or measurement changes. Inspect the page and repeat the observation before attributing the movement to its wording. --- # AI shopping and agentic commerce, explained Source: https://lantern.is/resources/agentic-commerce-primer Author: Lantern Briefing Published: 2026-05-22 Understand the difference between an AI recommendation, a prepared cart and an authorized purchase, with a practical product-data checklist for merchants. AI shopping uses an assistant to help a buyer research, compare or obtain products. Agentic commerce describes experiences in which software also carries out parts of the shopping task on the buyer's behalf. The practical question is which actions the particular service can take and what the buyer must approve. A recommendation, a prepared cart and a completed purchase are different outcomes. Measuring them together can make a merchant appear successful while the buyer is still unable to get the right item. ## Follow one buying task Consider an illustrative request: “Find a dishwasher-safe bottle that holds at least 700 ml, fits a 7 cm cup holder and arrives before Friday.” | Stage | What the assistant must establish | Merchant information that helps | | --- | --- | --- | | Discovery | Which products might satisfy the request | Product type, capacity and intended use | | Comparison | Whether each candidate meets the constraints | Base diameter, care instructions and supporting evidence | | Offer selection | Which exact variant can be purchased | Variant, current price, availability and seller | | Cart preparation | Whether the right quantity and variant were selected | Pack size, item identifiers and cart contents | | Purchase | Whether the buyer authorized the actual terms | Delivery estimate, total cost, payment and confirmation | An assistant that names a bottle but cannot verify its base diameter has not completed the comparison. One that creates the right cart but has no authorization to pay has not completed the purchase. ## Where the information can come from A model can draw on learned information. An assistant may also retrieve pages, use product feeds or call connected services. The available inputs depend on the product, mode and task. OpenAI documents [structured merchant feeds](https://developers.openai.com/commerce/specs) for product information. Google documents [product data attributes](https://support.google.com/merchants/answer/7052112?hl=en) for Merchant Center. These are provider-specific interfaces, not a universal file that gives every assistant the same access. A third-party review may help explain a tradeoff; the merchant's current offer is a different source of evidence. For a time-sensitive fact such as stock or delivery, check the current source and timestamp instead of relying on an old descriptive passage. ## What changes for the merchant The product page still needs to serve a person who wants to inspect the recommendation. Clear dimensions, material, compatibility, care instructions and return terms help that person decide whether the item fits the task. Keep visible content, structured data and feeds consistent. A page that says “in stock” while the selected variant is unavailable creates a practical failure regardless of whether the assistant mentioned the brand. Check access by purpose. A training crawler, search crawler and user-initiated visit can be different mechanisms. OpenAI's [crawler documentation](https://developers.openai.com/api/docs/bots) explains its named agents; use the equivalent current documentation for each service you intend to support. ## What SEO, AEO and GEO can usefully mean SEO concerns discovery through search. AEO and GEO are commonly used for improving how content can answer questions or appear in generated responses. The labels do not establish a separate guaranteed ranking system. For Google's AI search features, [the existing search fundamentals apply](https://developers.google.com/search/docs/appearance/ai-features). Google does not require a special AI text file or additional AI-specific schema. A useful answer, accessible page and accurate supporting information matter more than adopting a label. ## Test the buyer outcome Run a small set of real purchase questions and preserve the answers, cited pages and selected variants. Repeat under recorded conditions; do not use one favorable screenshot as the whole result. For the bottle example, score each stage separately: identified, constraints verified, offer current, cart correct, and purchase confirmed if that stage was actually tested with authorization. Stop the test at the stage you intend to evaluate. A discovery study does not need to place an order. Report the limits clearly. “The assistant identified a suitable bottle in four of six valid answers” is an inspectable observation. It does not establish how often all buyers will see the product or how many sales the recommendation caused. The [product-page guide](/resources/ai-legible-pdps) gives a worked data example. The [benchmark guide](/resources/q1-2026-benchmark) explains how to assess whether a comparison reflects your buyers and category. --- # Make product pages useful to shoppers and AI assistants Source: https://lantern.is/resources/ai-legible-pdps Author: Lantern Briefing Published: 2026-05-22 A worked product-page example covers visible specifications, valid structured data, provider-specific feeds and tests for variant, price and availability consistency. A useful product page lets a buyer verify whether the item meets a specific need. The same explicit facts can help an assistant interpret the product. Start with accurate, visible information, then keep structured data and feeds consistent with it. There is no universal attribute count or placement rule that guarantees an AI citation. Required fields depend on the product category, destination and program. ## Start with a buyer constraint In this illustrative example, a shopper needs a bottle that fits a cup holder with a 7 cm opening. The product holds 750 ml, but its base is 7.2 cm wide. Capacity alone cannot answer the fit question. | Weak description | Useful replacement | | --- | --- | | “The perfect bottle for every journey” | “750 ml bottle with a 7.2 cm base diameter” | | “Easy to clean” | The actual care instructions, including any lid/body differences | | “Available now” | Availability for the selected size and color | | “Fast delivery” | A delivery estimate for the buyer's destination and order time | For this example, the correct fit answer is that a 7.2 cm base is wider than the stated 7 cm opening. Do not imply compatibility simply because the product belongs to the right category. ## Add structured data that matches the page Google documents [Product structured data](https://developers.google.com/search/docs/appearance/structured-data/product) and its supported search appearances. The following hypothetical example illustrates consistent product facts. Replace every example value and URL with real, visible information, and check the requirements of the appearance you are implementing. ```json { "@context": "https://schema.org", "@type": "Product", "name": "Example Bottle, 750 ml", "image": "https://example.com/images/bottle-750.jpg", "description": "A 750 ml bottle with a 7.2 cm base diameter.", "brand": { "@type": "Brand", "name": "Example Maker" }, "sku": "EXAMPLE-BOTTLE-750", "additionalProperty": [ { "@type": "PropertyValue", "name": "Capacity", "value": "750 ml" }, { "@type": "PropertyValue", "name": "Base diameter", "value": "7.2 cm" } ], "offers": { "@type": "Offer", "url": "https://example.com/products/bottle-750", "price": "22.00", "priceCurrency": "USD", "availability": "https://schema.org/InStock" } } ``` This is an example, not evidence that a particular product exists or qualifies for every search feature. Add genuine identifiers and applicable properties where required. For subscriptions, make recurring charges, billing frequency and cancellation terms clear rather than presenting one charge as the full cost. Include ratings only when they reflect real eligible reviews and the visible page. Never copy an example rating or review count into production. Google's [structured-data policies](https://developers.google.com/search/docs/appearance/structured-data/sd-policies) require the markup to represent the page accurately. ## Use the feed's actual specification Google Merchant Center publishes a [product data specification](https://support.google.com/merchants/answer/7052112?hl=en). Review the requirements for your category and destination rather than targeting a generic number of populated fields. | Documented attribute | Useful check | | --- | --- | | id | Stable identity for the item | | title and description | Accurate product and variant information | | link and image_link | Working destinations for the correct product | | price and availability | Consistency with the current offer | | item_group_id | Appropriate grouping of related variants | | material | Relevant, accurate material information where applicable | | additional_image_link | Additional product images when useful | Use documented feed attributes within their stated scope. A similarly named field in another service is not interchangeable. ## Answer real questions in visible content A concise question such as “Will this fit a 7 cm cup holder?” can be useful when the answer states the measured limitation. Questions should resolve buying uncertainty rather than repeat the sales pitch. Keep any FAQ markup consistent with visible questions and answers. Do not add arbitrary numbers of FAQs or hide important specifications inside markup. Google's AI-feature guidance [does not require additional AI-specific schema](https://developers.google.com/search/docs/appearance/ai-features). ## Verify the complete path Check the page as rendered to a visitor and through the relevant testing tools. Validate the structured data, inspect feed diagnostics and compare the selected variant with the cart. For the bottle example, test capacity, base diameter, the selected color, price, stock status and delivery estimate separately. Keep the request, answer and cited URL with the result. If the assistant claims the bottle fits, record the unsupported claim and the source it used before changing the page. --- # Understand your brand’s presence in AI answers Source: https://lantern.is/resources/brand-ai-health-scoring Author: Lantern Briefing Published: 2026-05-22 Read brand mentions, recommendation position and cited sources together to identify discovery gaps and choose the next useful improvement. AI visibility results help you understand whether buyers encounter your brand, how it is presented and which sources support the answer. Read those signals together, then open the underlying responses to see what a buyer would actually learn. ## Look at presence, prominence and evidence | Signal | What it helps you understand | What to check in the answer | | --- | --- | --- | | Presence | Whether your brand appears for the questions you track | Is the brand recommended, compared with another option or mentioned in passing? | | Position | How prominently your brand appears in a recommendation | Does the answer lead with your product, place it among alternatives or mention it as a caveat? | | Citations | Which linked sources the assistant presents as evidence | Does the cited page support the claim about the correct product? | | Recommendation quality | Whether the suggestion fits the buyer’s request | Are budget, compatibility, availability and other constraints addressed accurately? | Frequent mentions are useful only in context. An assistant may name your brand while recommending a competitor. It may also recommend your product using an outdated specification. The full answer shows which problem needs attention. ## Start with the buyer question Separate questions that already name your brand from questions that ask for a product or solution. “What is Example Maker’s return policy?” tests information about a known brand. “Which carry-on fits under this airline’s seat?” tests discovery and suitability. Strong results on the first question do not establish coverage of the second. Focus on questions that represent actual purchase decisions. Include the constraints customers raise in support conversations, product reviews and sales discussions. The [prompt-set guide](/resources/prompt-set-playbook) provides examples for different buyer needs. ## Turn the result into a specific action | What you observe | What to inspect | A useful next step | | --- | --- | --- | | Your brand rarely appears for a relevant need | The alternatives recommended and the evidence used to support them | Check whether your pages clearly answer that need with verifiable product facts | | Your product appears with an incorrect claim | The passage, cited page and exact product variant | Correct information you control and request a factual correction on independent pages | | Your brand appears but does not fit the request | The buyer’s budget, dimensions, compatibility or other constraints | Clarify suitability and limitations so buyers can choose the right product | | A competitor is supported by stronger evidence | The comparison, review or specification the answer relies on | Develop useful evidence for unanswered buyer questions | | Buyers reach the site but struggle to purchase | The landing page, selected variant, availability and checkout | Fix the purchase obstacle and verify the complete journey | ## Read a recommendation as a buyer would Consider an illustrative request for a bottle that fits a 7 cm cup holder. An answer that names your bottle but omits its base diameter leaves the buyer unable to verify fit. An answer that cites the correct measurement and links to an available variant is more useful. The action is to confirm the measurement and make it clear on the product page and relevant product data. Then check whether subsequent answers represent that fact accurately. A more prominent mention alone would not resolve the buyer’s question. ## Compare changes in context Use the same buyer questions, assistants and relevant settings when comparing periods. Check whether enough recent answers are available to understand a pattern. A missing result is a reason to investigate the available evidence before drawing a conclusion. Read examples from both periods and note product launches, stock changes or page edits that could affect the result. One changed answer is a useful lead; repeated observations give you a better basis for deciding where to invest. The [weekly review guide](/resources/industry-changes-diff) explains what to look for. ## Choose a buyer outcome Define the improvement in concrete terms: an accurate specification, a suitable recommendation, a working product link or a completed purchase. Check that the correction is live, then review later answers and relevant business outcomes. Use the [product-page guide](/resources/ai-legible-pdps) to improve missing facts and the [citation guide](/resources/citation-score-methodology) to act on source evidence. Keep visits and orders alongside visibility when assessing business impact; each answers a different question. --- # Turn AI citation evidence into useful content improvements Source: https://lantern.is/resources/citation-score-methodology Author: Lantern Briefing Published: 2026-05-22 Use the pages cited in AI answers to find missing evidence, correct product facts and prioritize content that helps buyers make a decision. Citations show the pages an AI answer presents as support. They give you a practical starting point for understanding what information buyers encounter and where a correction or a better explanation could help. Open the cited page before acting. A familiar domain or a prominent link does not tell you whether the source supports the specific claim. ## Check the claim behind the link - Identify the sentence or recommendation the citation supports. - Confirm that the page describes the right product, variant and market. - Check whether specifications, prices, availability and policies are current. - Look for the evidence behind the claim: a published specification, a documented test or a relevant firsthand account. A link to a collection page might establish that a brand sells bottles without supporting a claim that a particular bottle fits a 7 cm cup holder. In that case, the missing evidence is the product’s measured base diameter. ## Choose the action by who controls the source | Source | What to look for | Useful action | | --- | --- | --- | | Your product page | Missing facts, conflicting specifications or an outdated offer | Publish the verified information where buyers can find it | | Your guide or comparison | An unanswered question or a conclusion without supporting evidence | Add the relevant evidence and explain the tradeoffs | | A retailer or marketplace listing | Wrong variant, pack size, availability or policy | Use the listing’s correction process with the exact item and supporting facts | | An independent review or publication | A factual error or a useful question your site does not answer | Request a specific factual correction, or create evidence that addresses the gap | | A community discussion | Recurring buyer confusion or reports that need context | Use the questions to improve your documentation and respond where appropriate under the community’s rules | ## Prioritize decisions that matter to buyers Start with information that could change a purchase decision. Incorrect compatibility, dimensions or return terms deserve attention even if the source appears only occasionally. A frequently cited page that already answers the question accurately may need no change. Separate a factual disagreement from an independent opinion. You can provide evidence that a specification is wrong; a reviewer remains responsible for their assessment of the product. ## Make the improvement useful on its own Write the answer a buyer needs and support it with information they can verify. Use clear headings, relevant measurements and comparable criteria. Where a result comes from your own test, state the conditions and limitations needed to understand it. Google’s [people-first content guidance](https://developers.google.com/search/docs/fundamentals/creating-helpful-content) emphasizes useful, trustworthy information. Its [guidance for AI features](https://developers.google.com/search/docs/appearance/ai-features) also directs site owners toward established search fundamentals. A useful page should serve the reader even when they arrive without an AI referral. ## Verify the page and review later answers After making a change, open the public page and confirm the corrected fact is visible. Check related product data for conflicting information. Keep the original answer and source so a later review has a clear reference. Review comparable buyer questions over time. Look for accurate use of the information and a better answer to the buyer’s need. A page update does not guarantee that an assistant will cite it. Use the [source-review guide](/resources/what-gets-cited-2026) to organize what you find, and the [product comparison guide](/resources/comparison-content-citations) when the missing evidence concerns a choice between alternatives. --- # Write product comparisons buyers can verify Source: https://lantern.is/resources/comparison-content-citations Author: Lantern Briefing Published: 2026-05-22 Build fair comparison pages with dated sources, consistent criteria and explicit tradeoffs. A worked example shows how to answer a buyer's constraint. A useful comparison page helps a buyer choose between real alternatives using the same criteria. Clear structure also makes individual facts easier to extract and check. It does not guarantee that an AI assistant will cite the page. ## Start with the decision Choose a real buyer question, such as which bottle fits a narrow cup holder while carrying enough water for a commute. The criteria should follow that question: base diameter and capacity come before a general claim that one brand is “better.” State who the comparison is for, which exact products and variants it covers, and when material facts were checked. If your company sells one of the products, make that relationship clear. ## Use the same criteria on both sides Here is a hypothetical example. These values are invented solely to demonstrate the comparison method. | Criterion | Bottle A | Bottle B | | --- | --- | --- | | Capacity | 750 ml | 600 ml | | Base diameter | 7.2 cm | 6.5 cm | | Fit for a 7 cm opening | Wider than the opening | Narrower than the opening | | Meets a 700 ml minimum | Yes | No | | Meets both stated constraints | No | No | The useful conclusion is that neither product meets both requirements. Recommending the product you sell anyway would defeat the purpose of the comparison. For a real page, place a source beside each material specification or group of facts. If a measurement is your own, describe how you took it. If it comes from the manufacturer, name the source and keep the checked date. ## Separate facts, tests and judgment A published dimension is a specification. A measured fit in a particular holder is a test result. “More comfortable to carry” is a judgment that needs a stated basis. Do not use one category as a substitute for another. If you have not tested durability, describe the material and warranty accurately rather than assigning a durability score. If you cannot verify a competitor's feature, mark it unknown and avoid converting the gap into a disadvantage. Prices and availability need particular care. Specify the relevant region, variant, purchase terms and date. Compare a subscription offer with another subscription offer, or explain why the terms differ. ## Give a conditional recommendation Useful conclusions name the tradeoff: “Choose A if the extra capacity matters and your holder accommodates its base; choose B if the narrower base matters and 600 ml is enough.” In the original request above, recommend looking for another option because both constraints are mandatory. A short answer near the beginning can help a reader orient themselves. The evidence table and testing detail should remain available below it. Keep the same meaning in headings, summaries and metadata so an extracted sentence does not reverse the actual conclusion. ## Review before publication Check that each cited source supports the particular claim, that the products and variants are still available, and that the comparison applies equivalent criteria. Ask someone unfamiliar with the draft to make a choice from it and identify any missing information. After publication, use buyer questions and product changes to maintain the page. If you also review assistant citations, [inspect the sources in the answer](/resources/what-gets-cited-2026) and the claims they support. A useful comparison remains useful even when no assistant cites it. --- # A 30-day plan for accurate brand information beyond your website Source: https://lantern.is/resources/external-signals-30-day-plan Author: Lantern Briefing Published: 2026-05-22 Audit profiles, correct conflicting facts and pursue useful independent coverage. Separate tasks you control from editorial decisions and AI visibility outcomes. A month is enough to audit and improve a brand's public information. It is not a dependable deadline for earning a Wikipedia article, changing a model's training data or increasing AI recommendations. Use the plan below to correct facts buyers encounter and make legitimate evidence easier to find. The schedule describes work a team can perform; it does not promise a particular discovery result. ## Week 1: inventory the public record Choose the profiles and third-party pages that actual customers use in your category. Record the URL, owner or contact route, business name, product description, relevant contact information and last checked date. Compare them with the current facts on your own site. Look for outdated product names, obsolete policies, broken links and confusion with another business. Keep screenshots or passages showing the discrepancy so a correction request is specific. Prioritize errors that could change a purchase decision. An incorrect return policy or discontinued product matters more than a cosmetic difference in a short profile. ## Week 2: correct what you can verify Update profiles your business is authorized to manage. For independent pages, use the publisher's correction process and provide evidence for the factual change. A request to correct an outdated specification is different from asking a reviewer to adopt your preferred opinion. Keep an evidence ledger: | Field | Example of what to record | | --- | --- | | Page | Exact public URL | | Problem | Outdated warranty length | | Evidence | Current published policy and effective date | | Action | Profile edit or correction request | | Outcome | Corrected, declined, pending or no response | | Checked | Date the public page was re-read | The example issue is illustrative. Preserve the actual source and response for each real case. ## Week 3: develop something worth covering Choose a buyer question your team can answer with firsthand evidence: a compatibility test, a transparent comparison, a repair guide or a documented customer problem. Publish a clear method and limitations with any test result. Approach relevant publishers with that evidence when outreach is appropriate. Editorial interest, timing and conclusions remain theirs. Do not make positive coverage, rankings or AI citations a condition of an ordinary factual contribution. Customer reviews should reflect real experiences. Follow the platform's rules and keep any permitted incentives or commercial relationships transparent rather than manufacturing a favorable consensus. ## Week 4: verify and measure the work Revisit the edited pages and correction requests. Report what actually changed: outdated facts corrected, working links restored, evidence published and responses received. Then examine comparable assistant answers if that is part of the program. Record the full responses and sources. A change in mentions after the work is an observation to investigate, not automatic proof that a specific profile edit caused it. ## Treat Wikipedia as an independent editorial project A business cannot claim a Wikipedia article as if it were a profile. Eligibility depends on qualifying coverage under Wikipedia's [organization notability guideline](https://en.wikipedia.org/wiki/Wikipedia:Notability_(organizations_and_companies)); a count of press placements alone does not establish it. People working for or with the subject must also follow the [conflict-of-interest policy](https://en.wikipedia.org/wiki/Wikipedia:Conflict_of_interest). Review the applicable disclosure and proposal processes before participating. A brand's preferred wording, existence or marketing schedule does not determine whether an article will be accepted. For many businesses, accurate information on relevant customer-facing platforms and useful evidence on their own sites will be the more immediate work. ## Keep the outcome honest End the month with a short record of completed corrections, unresolved gaps and the next useful piece of evidence to create. Avoid universal claims that an inactive channel is ignored after a fixed number of days or that one platform is always the strongest AI signal. --- # How to use weekly AI visibility changes Source: https://lantern.is/resources/industry-changes-diff Author: Lantern Briefing Published: 2026-05-22 Review changing mentions, recommendations and citations in context, then choose a specific action supported by the underlying answers. A weekly AI visibility review should tell your team what changed, why it matters to buyers and what to inspect next. Start with the answers behind the result, then decide whether the evidence calls for a correction, a closer look or continued observation. ## Check that the periods are comparable Review the same buyer questions, assistants and relevant settings across both periods. Note newly added prompts, products, markets or assistant experiences. Check whether the available answers cover both periods well enough to make a useful comparison. A gap in available answers needs investigation before it can support a conclusion about your brand. Keep launches, stock changes, promotions and significant site edits alongside the review. ## Read the changes that affect a buyer decision | Observed change | What to inspect | Possible next action | | --- | --- | --- | | Your brand appears less often | The affected questions and the alternatives now recommended | Check whether your product still meets the request and has clear supporting information | | A competitor becomes more prominent | The reasons given and the cited evidence | Look for an unanswered buyer need or a meaningful difference in the offer | | An answer introduces an inaccurate claim | The exact passage, product variant and source | Correct information you control or request a factual correction from the publisher | | A new source is cited | The page and the claim it supports | Assess whether it supplies useful evidence or repeats an error | | An answer improves after a page change | The corrected information and other changes during the period | Verify the improvement across further comparable answers | ## Use the context to interpret a change Suppose an assistant stops recommending a product after the relevant size goes out of stock. The immediate question is whether the stock information and available alternatives are accurate. Rewriting the product description would not resolve the missing size. Alternatively, an answer may recommend the item while misstating a compatibility requirement. That specific error can deserve attention even when the overall visibility result barely changes. These examples illustrate how to choose the next check. Read representative answers from both periods before attributing a change to a particular cause. ## Separate urgent corrections from trends to watch Prioritize inaccurate information that could lead a buyer to the wrong product or terms. Verify the source, fix the facts where you can, and check the public result. For broader changes, look for a pattern across relevant questions and dates. A single unusual answer may warrant another check before a substantial content project. Consider the importance of the affected buyer need as well as how consistently the change appears. ## Write a weekly note the team can use - What changed: identify the buyer question, assistant and comparison periods. - Evidence: include representative answer passages and source links. - Context: note changes in products, availability, questions or coverage. - Next action: name the specific check or correction and the person responsible. - Follow-up: set a review date and describe what a useful improvement would look like. Keep completed work separate from its later effects. Confirm that a corrected page or listing is live, then examine whether subsequent answers use the information accurately. The [guide to interpreting AI visibility](/resources/brand-ai-health-scoring) explains how to read mentions, position and citations together. Use the [revenue guide](/resources/visibility-vs-favorability) when assessing visits and purchases alongside discovery.