AudioEye has published research showing that website accessibility is directly linked to AI agent performance. The company sent 1,560 AI agents to complete everyday tasks on six websites, once with accessibility fixes and once without. On the site with the most accessibility issues, agents completed only 31% of assigned tasks, compared with 96% on an accessible version of the same site. Task completion dropped by about two-thirds for every model tested. Agents also used 43% more tokens on average on inaccessible sites, and up to six times more on the worst site. AI agents read a webpage using its accessibility tree; when that map is incomplete, tasks fail. In one example, numbers that appeared only inside an image without a text description could not be read by any of the six models across 60 attempts. Results were independently rechecked using WebJudge and agreed 95% of the time. AudioEye argues that accessibility investment made for compliance also prepares sites for agentic use.
AudioEye has published research showing that website accessibility is directly linked to AI agent performance. The company sent 1,560 AI agents to complete everyday tasks on six websites, once with accessibility fixes and once without. On the site with the most accessibility issues, agents completed only 31% of assigned tasks, compared with 96% on an accessible version of the same site. Task completion dropped by about two-thirds for every model tested. Agents also used 43% more tokens on average on inaccessible sites, and up to six times more on the worst site. AI agents read a webpage using its accessibility tree; when that map is incomplete, tasks fail. In one example, numbers that appeared only inside an image without a text description could not be read by any of the six models across 60 attempts. Results were independently rechecked using WebJudge and agreed 95% of the time. AudioEye argues that accessibility investment made for compliance also prepares sites for agentic use.