Accessibility tree
The accessibility tree is the browser's structured representation of a page built for assistive technology: a hierarchy of elements with roles, names, and states, stripped of purely visual detail. AI agents widely adopted it as their perception format because it is compact, semantic, and already maintained by every browser.
How it works
The browser derives the accessibility tree from the DOM, ARIA attributes, computed styles, and browser rules. It exposes roles, names, and states while omitting much of the visual page. Agents can use that smaller semantic representation to identify interactive elements, although incomplete markup and browser differences make it lossy. An rrweb recording preserves much of the DOM state used to build the tree, so replay can provide a useful basis for later inspection. It cannot guarantee the exact historical accessibility tree by itself. Reconstruction also depends on browser behavior, loaded resources, computed styles, and the fidelity of the recording. Record the accessibility-tree observation directly when exact model input matters.
Where you encounter it
Agent framework documentation, browser devtools accessibility panels, and the perception-format debate around computer use.