AI Agents Rely on Deteriorating Web Accessibility Tree Quality

Image credit: Search Engine Journal

AI agents are increasingly relying on the web’s accessibility tree to interpret websites, even as the quality of this structural model has worsened for the first time in six years, according to recent accessibility data.

This deterioration poses a significant challenge for web developers and the future of artificial intelligence interaction with online content, especially as automated bots now account for the majority of internet traffic.

From May 30 to June 5, 2026, automated bots were responsible for 57.2 percent of all HTTP requests to HTML content, surpassing human traffic, Cloudflare Radar reported. This shift occurred more than one year earlier than initially forecast, according to Cloudflare, a web infrastructure company.

AI agents primarily read websites using the accessibility tree, a simplified structural model of a webpage, rather than processing visual layouts, industry experts said. Browsers construct this tree from the Document Object Model (DOM), providing semantic information such as roles, names, states, and descriptions of web elements.

This semantic information is essential for non-visual software and AI agents to understand and interact with web content, according to the World Wide Web Consortium (W3C) and MDN Web Docs.

AI agents favor the accessibility tree over raw HTML or vision-based models due to its efficiency and reliability. Processing compact text from the accessibility tree is less costly than the extensive image processing required for vision models, and explicit element roles offer more reliable interpretation than guessing based on pixels, sources familiar with the technology said.

Even OpenAI, which develops vision-first artificial intelligence models, has recommended that websites enhance their accessibility, stating that accessibility and readability present the same problem for machines. Matthew Prince, chief executive officer of Cloudflare, has also highlighted the growing dominance of bot traffic.

The declining quality of the accessibility tree, upon which these AI agents depend, represents a significant issue for the future of web interaction, according to data from various accessibility monitoring organizations.

Source: Search Engine Journal