Shoppers who click through from an AI assistant buy far more often than other visitors when they arrive on a computer, and far less often when they arrive on a phone. That is the split in a study CHEQ published on October 21, 2026, which followed 14 retailers selling fashion, beauty, outdoor gear, DIY supplies, homeware and electronics in the UK, Europe, Japan and South Africa from August 19 to October 18. On desktop, 3.9% of visits referred by an AI assistant ended in an order, against 1.6% of human visits. On mobile, AI-referred visits converted at 0.6%, against 1.7% for human visits. CHEQ sells software that classifies and filters the traffic reaching a website, including AI agents.
The phone side decides the total. In the study, 86% of AI-referred visitors arrived on a phone, compared with 78% of human visitors, so AI-referred traffic as a whole converted at 1.1%, below the human rate of 1.7%. The 14 stores received about 1.8 million visits from AI assistants over the 60 days.
Most of those phone visits appear to start inside the ChatGPT app, which is also why they are hard to find in a store's reports. CHEQ found that only 1 in 9 AI-referred visits carried an AI referrer. The rest arrived with a utm tag alone, which many analytics setups file as Unassigned. Most of the mobile visits carried a ChatGPT tag with no referrer, the pattern CHEQ reads as taps from inside the app. A channel report built on referrers puts most of these shoppers in Unassigned, where their low mobile conversion rate is averaged in with everything else that lands there.
CHEQ's reading of its own data is that these shoppers arrive ready to buy and drop off at the mobile landing page and checkout. The study counts a visit as converted when that visitor reached order confirmation on the same day, so a shopper who taps a link on a phone and orders later on a laptop would not show up as a mobile sale.
Adobe's U.S. data points the other way on the overall picture. In Adobe's monthly figures, drawn from more than a trillion visits to U.S. retail sites, visits referred by an AI assistant have converted better than other visits every month since September 2025, and 48% more often in August 2026. The two studies measure different things. Adobe compares AI-referred visits with all other traffic to U.S. retailers, including paid search and email. CHEQ compares them with a sample of human visits to 14 stores outside the U.S.
AI agents arrived in much smaller numbers, about 99,000 visits across the 14 stores, and their results depended on the kind of agent. Visits where an AI agent worked inside the shopper's own browser converted at 7.1%, about four times the human rate. Visits from hosted agents such as ChatGPT Agent, which run in their own browser from start to finish, converted at 0.2%. CHEQ treats both agent figures as directional because they come from smaller samples.
With two large studies pointing different ways, the number that settles it for a store heading into the holiday season is its own, split by device. In your analytics, filter sessions whose source contains "gpt", which catches ChatGPT's tagged visits even when they sit in Unassigned, and compare their conversion rate on mobile with the rate for everyone else who shops on a phone.
Common questions
Do shoppers from AI assistants convert better than other visitors?
On desktop they did in CHEQ's study of 14 retailers in the UK, Europe, Japan and South Africa: 3.9% of AI-referred desktop visits ended in an order, against 1.6% of human desktop visits. On mobile the order reversed, at 0.6% against 1.7%. Because 86% of AI-referred visitors came on a phone, AI-referred traffic as a whole converted at 1.1%, below the human rate of 1.7%. Adobe's data on U.S. retail sites shows the opposite overall: AI-referred visits converted 48% more often than other visits in August 2026.
Why does traffic from ChatGPT show up as Unassigned in analytics?
In CHEQ's data, only 1 in 9 AI-referred visits carried an AI referrer. The rest arrived with a utm tag and no referrer, which many analytics setups file as Unassigned. Most of the mobile visits carried a ChatGPT tag, which CHEQ reads as taps from inside the ChatGPT app.
How do AI agents convert compared with human shoppers?
Visits where an AI agent helped inside the shopper's own browser converted at 7.1% in the study, about four times the human rate. Visits from hosted agents such as ChatGPT Agent converted at 0.2%. CHEQ describes both figures as directional because the agent samples were small.
How did CHEQ measure conversion?
It counted a visit as converted when that visitor reached an order confirmation page on the same day, using data from August 19 to October 18, 2026. Human rates came from a random 5% sample of about 42 million visits, and traffic flagged as invalid was excluded from both the human and AI-referred groups.