Presented by AudioEye
Developers trust AI to write accessible code. A new AudioEye study shows it can't.
Ask an AI coding assistant to build a checkout flow, and it will hand one back before your coffee cools.
That speed is why nine in 10 developers use at least one AI tool at work, and 74% have adopted a dedicated AI coding tool, according to a JetBrains survey of 10,000 developers.
The promise of vibe coding is simple: code done right, at a pace no person could match. Most teams take that at face value. But move through that same checkout experience with a keyboard instead of a mouse, or a screen reader instead of a screen, and the cracks show immediately. AI writes code fast. It just doesn't write code that works for everyone.
The web AI learned from was already broken
Most AI coding tools were trained on an internet that is largely inaccessible to people with disabilities. WebAIM has tracked the accessibility of the top million homepages since 2019. In that span, the average number of accessibility issues per page has never meaningfully improved. In WebAIM’s latest report, the average number of issues rose 10% from the previous year, and AI-assisted code was named a likely contributor to the increase.
AI coding tools learned from an inaccessible web, so they write inaccessible code by default, even when instructed to use accessibility best practices. Still, 81% of developers believe their AI-generated code already meets accessibility standards. We wanted to know if that was true. So we tested it.
AI can’t build accessible pages even when asked
We gave five AI tools (OpenAI, Anthropic, Google, xAI, and Lovable) the same brief: build three websites that meet the latest Web Content Accessibility Guidelines (WCAG) at the required AA standard. Each tool returned sites they claimed were accessible, but they were riddled with issues when we tested them.
All 15 sites failed WCAG Level A, the standard's most basic tier, averaging 55 issues per page. AudioEye's 2026 Digital Accessibility Index puts the typical site at 62 issues per page, which means AI-produced pages are essentially as inaccessible as the average website today.
91% of issues found were medium or high severity. These are failures that prevent someone from completing a purchase, submitting a form, or booking an appointment. They’re also the same issues that show up in lawsuits.
AI did not hallucinate these failures. It learned from an inaccessible web, and it now reproduces serious inaccessibility issues into new code as fast as teams can ship it.
The trust gap: Developers still trust AI-generated code
Accessibility issues on a site aren't new. What’s worrying is that developers believe AI is solving them.
81% of teams who use an AI tool are confident that the AI-generated code on their site meets accessibility guidelines. Of that same group, 50% have also noticed more accessibility issues and complaints since adopting AI.
That gap between perception and reality has a cost: 46% of the organizations we surveyed that use AI to write code or generate content said they had received an accessibility complaint, demand letter, or lawsuit in the last 24 months. 71% of those respondents also reported that AI was involved in coding the page that received a complaint.
Accessibility has to scale with AI
Today’s LLMs lack the accessibility expertise to build accessible sites. Additionally, the old way of fixing accessibility issues at the source with consultants and audits won't keep up with the speed of modern code development. The only way to address accessibility at scale is with a dataset tested and verified against millions of use cases. Since LLMs lack this critical accessibility data, our robust dataset enables seamless integration with your developer environment or AI-generated fixes on your front end, resulting in an accessible experience.
AI is building the web faster than anyone could have imagined. If your accessibility solution doesn’t scale at the same pace, your litigation risk will.
Kelly Georgevich is CEO at AudioEye.
Sponsored articles are content produced by a company that is either paying for the post or has a business relationship with VentureBeat, and they’re always clearly marked. For more information, contact sales@venturebeat.com.