The Accessibility Tree Is How AI Agents Read Your Site

What Is the Accessibility Tree?

The accessibility tree is the structured representation a browser builds from your HTML — every element's role, name, state, and relationships. It is the agent's primary data model: how an AI agent understands what is on your page and what it can do.^[1]

Originally built for screen readers, the tree is now the interface AI agents read first. Google's agentic browsing category includes audits that check its integrity, because if the tree is broken, the agent is effectively blind.

How Do You Fix a Malformed Accessibility Tree?

The fixes are semantic-HTML fundamentals, and they double as human accessibility wins:

• Use native elements. A real <button> or <a> carries its role automatically; a <div onclick> does not.
• Name every control. Give icon-only buttons an aria-label; associate inputs with <label>.
• Keep roles correct. Do not override a native role with a conflicting ARIA role; keep parent-child structures valid (lists contain list items, etc.).
• Expose interactive content rather than hiding it behind purely visual markup or rendering it only after interaction.

A clean tree makes your site legible to agents and to the assistive technology humans use — one fix, two audiences.

Frequently Asked Questions

Is the agent accessibility audit the same as the Lighthouse Accessibility score?

It overlaps but is distinct. The agentic browsing category uses a subset of accessibility checks that AI agents specifically depend on — names, tree integrity, and visibility of interactive content — scored as part of the agent pass-ratio.

Does fixing the accessibility tree help human users too?

Yes. The same semantic HTML and ARIA labeling that agents need also powers screen readers and assistive technology, so the work improves human accessibility at the same time.