The accessibility tree is a semantic representation of the page generated by the browser. It extracts from the DOM the information assistive technologies need: roles, accessible names, states, relationships and interactive elements.
Why it matters for agents
AI agents can use this tree as a functional map of the interface. Visual rendering indicates where elements are. The DOM indicates how they are structured. The accessibility tree indicates what those elements are meant to do.
When a button has a clear accessible name, when a field is connected to its label, when a modal correctly exposes its role and title, the agent receives a more reliable signal for action.
What the tree exposes
The accessibility tree exposes, among other things:
- an element’s role: button, link, field, menu, dialog;
- its accessible name: the text or label an assisted user will receive;
- its state: enabled, disabled, expanded, collapsed, selected;
- its relationships: field linked to a label, error linked to an input, control linked to a region.
Frequent errors
The most common errors are unnamed buttons, unlabelled fields, undescribed clickable icons, missing states, poorly structured modals and interactive elements hidden from accessibility.
These errors first harm people who rely on assistive technologies. They also harm agents that try to interpret the interface from programmatic signals.
Link with agentic navigability
A clean accessibility tree does not guarantee that a site is agent-ready by itself. But an incomplete or incoherent tree directly weakens agentic navigability.
That is why the agentic readiness audit compares the accessibility tree with visual rendering and hydrated DOM.