Perceive
Can an agent build a clean model of the page from the accessibility tree, the landmarks and a stable layout, or is it guessing.
Free tool · AI agent readiness
Can an AI agent actually use your site?
Agents do not see your design. They read the accessibility tree, the DOM and a screenshot, then try to act. This free check pulls the real accessibility tree your browser builds and scores whether an agent can perceive, navigate and cite your page, against Google's own agent guidance. It takes about a minute.
How the check works
Load it like an agent does, read the tree, score the gaps.
No dashboard to learn. A real browser does the work, then it gets out of your way.
01 · Load
It loads your page
A real headless browser opens your URL, the same way an agent would, and waits for it to settle. No guessing from source code alone.
02 · Read
It reads the accessibility tree
It pulls the semantic map an agent relies on: the roles, names and states of every control, plus layout stability, semantic markup and the signals that say a thing is clickable.
03 · Score
It scores perceive, act, extract
You get a score for whether an agent can see the page, operate it, and cite it, with the exact elements at fault and the fix for each.
Why it matters
Your next visitor might not have eyes.
Agents are starting to browse, compare and book on behalf of real people. They do not read a beautiful page, they read a machine-readable one. A button an agent cannot name, a form field with no label, a control built from a styled div: each one is a task an agent cannot finish on your site.
Act
Can it operate the page: named controls, labelled fields, semantic buttons and links it can actually click and fill.
Extract
Can it read and cite you: a clear heading outline, descriptive links, structured data and an answerable summary.
FAQ
Fair questions about the check.
How do AI agents read my site?
Not the way you do. An agent works from a screenshot, the raw HTML, and the accessibility tree: the semantic map of roles, names and states that browsers build and screen readers use. Google calls that tree the agent's primary data model. This tool reads it from your live page.
What does the score actually measure?
A 0 to 100 score across three things an agent tries to do: perceive the page, act on it, and extract and cite it. It is built from Google's agent-friendly rules and the Lighthouse agentic-browsing checks, plus a content layer for citability that most audits skip.
Is it really free?
Yes. The score and the headline issues run with no email. You give an email to unlock the full breakdown, every check and the fix for each, which I also send to your inbox as a report you can hand to a developer.
Is this just an accessibility audit?
They overlap, and that is the point: the same accessibility tree that serves a screen reader serves an agent. But this check also weighs the things specific to agents and to being cited in AI answers, like layout stability, actionability signals and structured data, not only WCAG conformance.
When the check turns something up
Being seen by the robots was always important but now they outnumber us it's more important than ever.*
This free check finds where an agent gets stuck. Fixing any problems identified across the whole site makes you legible to the agents and answer engines your customers are starting to use. Because if the robots aren't surfacing you, the humans that use them won't find you either.
* Bots now outnumber humans online. Source: Forbes ›