AudioEye study finds AI-built websites failed basic accessibility checks

AudioEye says it asked five AI tools—OpenAI, Anthropic, Google, xAI and Lovable—to build websites to the WCAG 2.2 AA standard.[1] For each tool, the assignment covered a news and media site, an online store and a financial services site, producing 15 sites in total.[1][2] AudioEye tested areas including keyboard focus, dialog handling, form-error announcements and screen-reader output.[1] This was a test of the sites produced for those assignments; the evidence does not establish that every site built with these tools has the same results.

The 15 resulting sites all failed WCAG Level A, which VentureBeat described as the standard’s most basic tier.[2] AudioEye identified 306 distinct accessibility issues across the sites, with those issues appearing more than 59,000 times across the pages.[1] The company said 91% of the issues were medium or high severity.[1] AudioEye rated issues high, medium or low based on their impact on users and legal risk.[1]

AudioEye said 33% of identified issues blocked a task outright, and 13 of the 15 sites had at least one issue that completely blocked a task.[3] The company said medium- and high-severity issues made tasks harder or impossible for people with disabilities and AI agents.[1] One example involved navigation: 12 of the 15 sites included a skip link, but AudioEye said none had its target wired.[3] The results point teams toward testing whether users can complete key tasks, not just checking whether an interface element appears on a page.

In AudioEye’s business survey, 81% of respondents said they were confident that AI-generated code on their site met accessibility standards.[1] At the same time, 73% reported that accessibility issues and complaints had increased since adopting AI.[1] Almost half, 46%, said they had received a demand letter or lawsuit; among those respondents, 71% said AI tools had been used on the pages receiving the complaint.[1] AudioEye also said its AI-built sites averaged 55 accessibility issues per page, compared with 62 for the typical site in its 2026 Digital Accessibility Index.[1] Its conclusion was that asking an AI tool to build an accessible site made little difference compared with the typical website.[1]