Free tool, no account needed
AI agents can now shop, book and fill in forms for people. Check if they can use your website.
An AI agent is an assistant that acts on websites for a person. It can search, click, fill in forms and shop.
Agents look at a page in two ways. They look at a screenshot, and they read the page's built-in label map, the same one screen readers use. Google's Chrome team calls that label map the agent's primary data model.
Which mix each product uses is mostly not public. But good names on buttons, links and fields, and a page that doesn't jump around, help both views.
We read the HTML your server sends. We visit the page once without a name, and three times with a name tag for the fetchers of ChatGPT, Claude and Perplexity. We don't click, log in, submit forms or run your scripts.
Real agents come from their own addresses, and browser agents run JavaScript. So a blocked test visit is a sign, not proof, and a page built in the browser can look different to them.
Our page checks follow the rules of Google's Lighthouse Agentic Browsing category. Google calls that category experimental and gives it no score, just checks that pass or fail. We do the same.
Lighthouse: how Agentic Browsing is scoredMost sites don't, yet. Lighthouse treats llms.txt as optional, and Google says Google Search doesn't use it.
WebMCP is a draft from a W3C community group, not a standard. Chrome is testing it and ChatGPT's own browser supports it. It mostly helps apps where agents take real actions, like booking or ordering.
MCP servers and API catalogs only matter if you offer an API. The standards are still moving: the ARD discovery file changed its name in 2026.
No. Agent readiness means an agent can use your page. AI visibility means ChatGPT or Google AI actually cites or recommends it.
A page can be easy for agents and still never be picked. Check how ready your page is for AI answers
A crawler collects pages ahead of time, to build a search index or training data. An agent visits right now, for one person, to do a task like finding a price or filling in a form.
There is no evidence that it does. Google calls its own Agentic Browsing check informational, with no score. Fix the issues because they help agents and people who use screen readers, not for rankings.
Usually a firewall or CDN rule, a human check, or a login. Cloudflare, for example, now blocks agent bots on pages with ads by default for new sites. Check your bot settings first.
No. The agent makers we checked tell their agents to stop at human checks and ask the person to finish them. A CAPTCHA can be the right choice for your forms.
Yes. One check sends about 10 requests (plus a few redirects if your site uses them). We only read what your server sends back. We never click, never log in and never submit anything, and each request carries a name tag that links to this page.
Lighthouse checks the page after JavaScript in a real browser, but it's a developer tool. Cloudflare's checker lists agent protocols in detail. We explain the basics in plain words and also test robots.txt and firewall access.
Because nobody knows yet how much each check matters. Google left the score out of its own agent checks for that reason. A made-up score would look precise, but it wouldn't be.
The GEO check scores one page on 22 points that matter for AI answers, from structure to facts.
Run a GEO check