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 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 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 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