Google PageSpeed Now Audits Your Website's AI Readiness — Here's What the Score Means

In May 2026, Google quietly added a new category to Lighthouse — the open-source tool that powers Google PageSpeed Insights. The category is called “Agentic Browsing.” It checks whether a website is correctly configured for AI agent interaction.

This is not a small change. Google’s PageSpeed Insights is the testing tool that most website owners and developers use to evaluate a site’s technical health. Adding a new audit category to it is Google’s way of saying: this matters now, not later.

Here is exactly what the audit checks, what each check means in practice, and what the score does — and does not — mean for your business.

The four checks

The Agentic Browsing audit (Lighthouse 13.3.0, released May 7, 2026) runs four checks. Not six, not twelve — four. Some coverage of this audit conflates sub-checks within individual categories with top-level scored items. Here is the accurate breakdown:

1. Accessibility tree is well-formed

This check verifies that the website’s accessibility structure — the underlying semantic HTML that screen readers, automated tools, and AI agents use to understand page structure — is correctly formed.

The specific issue it looks for: aria-hidden attributes applied to elements that contain interactive content like buttons, links, or form fields. When an element is marked aria-hidden, it is declared invisible to accessibility systems. If that element contains a button, the button becomes unreachable to any system that navigates by accessibility tree — including AI agents that interpret page structure without rendering it visually.

The fix is straightforward: use the inert HTML attribute instead of aria-hidden on elements like slide-out menus or modal overlays. inert correctly removes elements from both the accessibility tree and keyboard navigation without breaking the underlying structure.

2. Cumulative Layout Shift (CLS)

CLS measures whether the page’s visual content moves around during loading. If an image loads late and pushes text down, or a font change shifts a headline, or an advertisement appears and bumps the content below it — all of these generate layout shift.

CLS is already part of Google’s Core Web Vitals and is a documented ranking signal. Its inclusion in the Agentic Browsing audit relates to a different concern: an AI agent that reads the page’s layout at one moment and tries to interact with it a moment later may find that the element it identified has moved. A page with high CLS is unreliable for agent interaction, not only for human users.

3. WebMCP tools registered

This check confirms whether the website has registered any tools through the document.modelContext.registerTool() API. If the site has tools registered, they are listed here with their names, descriptions, and input schemas — giving a developer immediate visibility into what the AI infrastructure layer looks like.

Important clarification: this check typically shows as “Not Applicable” in Google’s PageSpeed Insights web interface because Google’s testing lab does not run with WebMCP enabled. This is not a failure — it is expected. The check works in the local Lighthouse CLI (npx lighthouse --preset agentic-browsing), which does run in a WebMCP-capable environment.

The sub-checks within this category — schema validity, form annotation coverage — are informational, not separately scored items.

4. llms.txt follows recommendations

This check confirms whether the site has a llms.txt file at its root URL and whether that file follows the format the W3C Web Machine Learning Community Group recommends: a valid Markdown document beginning with an H1 heading containing the site or business name.

llms.txt was proposed by Jeremy Howard of Answer.AI in September 2024. By June 2026, it had been adopted by approximately 8.7% of the top 1,000 global websites — a significant increase but still a minority. Cloudflare, Azure, GitHub, and many major technology companies have implemented it.

The honest caveat: despite widespread implementation, no major AI platform’s crawlers have been observed actually reading llms.txt files in server log data. The Lighthouse check confirms the file is correctly formatted — not that any AI system is reading it. The value is in having the infrastructure in place when reading systems catch up.

What the score means

A 4/4 score means all four checks passed. The denominator is dynamic: a site without a llms.txt file is scored against three checks and can achieve 3/3 at most. A site that has implemented both WebMCP and llms.txt is eligible for 4/4.

Critical clarification: this score is informational only. It does not affect your PageSpeed performance score (0–100) and it is not a Google Search ranking signal.

Google has explicitly stated that the Agentic Browsing category “does not have a weighted average score from 0 to 100.” The audit gathers data. It does not penalise or reward sites in the traditional ranking sense.

On llms.txt specifically, Google’s Search Central team has confirmed it “won’t harm (nor help) your visibility or rankings in Google Search.”

What a 4/4 means for TVS sites

Every site TVS builds passes this audit at 4/4 — measured against the local Lighthouse CLI, where the WebMCP check is evaluable.

This is not a marketing claim about ranking. It means the AI infrastructure layer is in place: the accessibility tree is clean, layout shift is within the correct threshold, WebMCP tools are registered and correctly specified, and the llms.txt is valid and present.

The practical significance of a 4/4 is that the site is correctly positioned for AI agent interaction. When AI agents that consume WebMCP tools ship to mainstream users — which Google has indicated is their direction with Gemini in Chrome — these sites will have structured tools ready to be called. Sites that have not implemented this will need to retrofit it.

Google does not add audit categories to PageSpeed Insights without intent. They added HTTPS as a ranking signal years after making it auditable. They added Core Web Vitals as a ranking signal after making them auditable. The Agentic Browsing audit is in the early stage of that same trajectory.

How to check your site

Run a local Lighthouse audit with the agentic browsing preset to see your current score:

npx lighthouse https://yoursite.co.za --preset agentic-browsing --output html --output-path ./agentic-report.html

Or run the TVS visibility check to see how your site scores on the full TVS audit criteria, which includes the Agentic Browsing checks alongside the complete technical, content, and trust evaluation.

---

TVS builds websites that pass the full Google Agentic Browsing audit at 4/4. Run a free AI Readiness check to see how your site currently scores, or speak to us about what it would take to fix the gaps.