AI coding tools fail to meet basic web accessibility standards

A study by AudioEye has revealed that popular AI coding tools fail to meet basic web accessibility standards. The research tested tools from OpenAI, Anthropic, Google, xAI, and Lovable, tasking them with building news, e-commerce, and financial services websites intended to comply with Web Content Accessibility Guidelines (WCAG) 2.2 Level AA.

All 15 generated sites failed to meet even Level A, the most fundamental level of accessibility compliance. Testing identified 306 distinct accessibility issues that appeared over 59,000 times across the sites. This lack of compliance is attributed to the fact that these AI models learned to code from the existing web, which is already largely inaccessible.

The findings come at a time of rapid AI adoption in software engineering. Data from DX indicates that AI-authored code accounted for 51.9% of total code in mid-2026, a significant increase from 24% just two quarters prior. Despite these failures, an AudioEye survey found that 81% of developers and business leaders expressed confidence that the AI-generated code on their sites meets accessibility standards.