Free AI Readiness Check

What Google’s Agentic Browsing audit checks

Lighthouse 13 added a fifth category next to Performance, Accessibility, Best Practices and SEO: Agentic Browsing. It measures whether a page can be operated by an AI agent that browses on a user’s behalf — the way Gemini in Chrome, ChatGPT’s agent mode or Claude for Chrome do when they fill a form, compare offers or complete a booking. The category is scored like Google’s own PageSpeed Insights, as a pass count: “2/3” means two of the three audits that applied to your page passed.

The audits fall into four groups. The accessibility tree: agents read your page the same way screen readers do, so a button without a name or a link without text is invisible to them. Layout stability: Cumulative Layout Shift under 0.1 matters because an agent that clicks where a button used to be gets the wrong result. llms.txt: a plain-text summary of your site at /llms.txt, checked for an H1, a short description and a sensible number of links. And WebMCP: the proposed standard that lets a site register tools and form schemas for agents; these audits report “not applicable” until you adopt it, so you lose nothing by not having it yet.

What our AI checks add

Lighthouse looks at the rendered page. We add nine checks over the files and tags assistants actually fetch when they decide what to say about you. robots.txt: whether GPTBot, ClaudeBot, PerplexityBot and Google-Extended are allowed to read the site, and whether training crawlers are treated separately. llms.txt: whether it exists and follows the recommended shape. JSON-LD: whether the page names your organisation or business in structured data — our LocalBusiness schema generator builds it if it is missing. Then the basics assistants quote verbatim: title, meta description, Open Graph tags, the html lang attribute and a canonical URL. If llms.txt is the one you are missing, the llms.txt generator writes one from your sitemap in a minute.

How to read the scores

The five Lighthouse gauges use Google’s thresholds: 90–100 is green, 50–89 orange, below 50 red. Core Web Vitals (LCP, CLS, TBT, FCP, Speed Index) are shown with Lighthouse’s own display values, and the top performance opportunities list what would save the most milliseconds. Mobile runs on a simulated mid-range phone over a throttled 4G connection, which is why mobile Performance is usually lower than desktop — test both.

AI Readiness is our own 0–100 score, marked “Our score” so it is not mistaken for a Lighthouse category. It weights the nine checks above by how often each one breaks an assistant’s ability to find, read or quote a site; a blocked crawler costs more than a missing Open Graph image. It is a readiness measure, not a prediction: it does not estimate rankings, traffic, or whether ChatGPT, Claude or Gemini will mention you. For that you need to ask the assistants themselves.

What to fix first

The fix list at the bottom of the result is ordered by impact. Robots and llms.txt come first because they are one-file fixes that change what every assistant can see. Structured data is next: a single JSON-LD block that names your business, address and offering. Then the markup audits agents trip over — unnamed buttons, links without text, missing form labels — which are usually a few attributes in a template. Performance opportunities come last; they matter for users and for Core Web Vitals, but a slow page is still readable to an agent while an unnamed button is not. Once the files are in place, run the AI visibility checker to see how assistants describe you, and the local SEO audit for the on-page basics that feed both search and AI.

Limits of this check

This runs Google’s open-source Lighthouse in a real headless browser on our servers, plus our own checks. A single run is a snapshot: scores vary a few points between runs because of network noise, and a page behind a login, a cookie wall or aggressive bot protection will score lower than it does for a signed-in user. We check one URL at a time, not a whole site, and we cache results for 24 hours per URL and device, so a fix you ship today shows up tomorrow. We only fetch public https pages, never private networks, and we keep nothing except the result you see.

Frequently asked questions

Is this the same as PageSpeed Insights?

The Lighthouse part is: we run the same open-source Lighthouse Google uses in PageSpeed Insights, in a real headless Chrome on our servers, with the same mobile and desktop presets. PageSpeed Insights also shows field data from real Chrome users (CrUX); we do not. On top of Lighthouse we add our own AI checks and the AI Readiness score, which PageSpeed Insights does not have.

What is Agentic Browsing and why does Lighthouse score it as 2/3?

Agentic Browsing is the fifth Lighthouse category, introduced in Lighthouse 13. It checks whether an AI agent can operate your page: a well-formed accessibility tree, a stable layout, an llms.txt file and optional WebMCP tool declarations. Google scores it as a pass count rather than a percentage because several audits only apply to sites that have adopted WebMCP, so “2/3” means two of the three applicable audits passed.

Does a high AI Readiness score mean ChatGPT will recommend my business?

No. AI Readiness measures whether assistants can find, read and quote your site: open robots.txt, an llms.txt, structured data and clean metadata. Whether they recommend you depends on your reviews, citations and how you compare with alternatives. Our Brand Visibility product asks ChatGPT, Claude and Gemini directly and shows you what they say.

Why do I get a different Performance score each time?

Lighthouse measures a live page load, and network and server conditions change between runs, so a few points of variance is normal. The thresholds are the ones that matter: stay in the green band rather than chasing a specific number. We cache a result for 24 hours per URL and device, so you will see the same numbers until the cache expires.

Is it free, and do you keep my URL?

Yes, it is free and needs no account. We keep the result for 24 hours so the “Copy link” button works and repeat runs do not queue another browser session; after that it expires. We only check public https URLs and never fetch private or internal addresses.