Accessibility Tree
The Accessibility Tree is the structured representation of an interface that exposes roles, names, states, and relationships relevant to accessibility. In the agentic web, it also becomes an action map for agents.
Short definition
The Accessibility Tree translates a page into interpretable elements: buttons, links, headings, fields, open or closed states, values, labels, and relationships. It ignores part of the decorative noise and reveals what the interface claims to make available to an assisted user or to a system reading the action structure.
What the concept governs
The concept governs the layer between raw DOM and visual rendering. An agent may use visual rendering to locate elements, HTML to understand the documentary structure, and the Accessibility Tree to identify available actions. When these layers diverge, the agent receives contradictory signals.
What this concept is not
The Accessibility Tree is not a surface to manipulate only for machines. Its primary function remains human accessibility. Its agentic value comes precisely from the fact that an accessible interface often forces roles, names, and states to be more explicit.
Common failure modes
- button without an accessible name;
- link text that does not describe the destination;
- field without an associated label;
- accordion without open or closed state;
- modal without title or focus management;
- interactive icon ignored or wrongly named;
- important element hidden from the Accessibility Tree.
Governance implication
The governance implication is that accessibility becomes a proof surface for interface intention. If the name, role, or state of an element is missing, the agent must reconstruct the intention elsewhere, usually with more inference.
Reading rule
Use Accessibility Tree when the analysis concerns roles, names, states, and relationships exposed by the interface. Do not confuse this layer with the complete DOM or visual rendering.