Google Added an 'Agentic Browsing' Score to PageSpeed Insights — What It Means for Your Website

In May 2026, Google's Lighthouse 13.3 added a new category called Agentic Browsing. It now appears in PageSpeed Insights, and it grades a question no tool asked before: can an AI agent actually use your website?

For twenty years, PageSpeed Insights measured your site for one audience: people. How fast does it load, how stable is it, is it accessible, is it structured for search. On May 7, 2026, that changed. Lighthouse 13.3 — the engine behind PageSpeed Insights — shipped a brand-new category called Agentic Browsing, and within a couple of weeks it was live for anyone who runs a report at pagespeed.web.dev.

It answers a sharper question than "is this page fast?" It asks: can an AI agent read this page, understand its controls, trust its layout, and complete a task without guessing?

Why Google added it now

Because a meaningful and growing share of the traffic hitting public web servers in 2026 isn't human. Look at who's browsing on a user's behalf: OpenAI's Operator, Anthropic's Computer Use, Google's own Project Mariner, Perplexity, and ChatGPT's browse mode. A person hands one of these agents a task — "compare these three vendors and book the one that can start next week" — and the agent executes it across multiple sites. If your site is unusable to that agent, it simply routes around you to a competitor it can use.

The tooling finally caught up to that reality. Agent-readiness is now a visible, shareable line item on the exact screen every developer, agency, and client already opens.

What it actually checks

Unlike the familiar 0–100 categories, Agentic Browsing shows a pass/fail ratio — like 2/3 or 3/3. In PageSpeed Insights, three default checks drive that ratio:

• Accessibility tree — An agent-centric read of your accessibility tree to confirm elements are named and structured so a machine can use them. Agents navigate the accessibility tree far more efficiently than raw HTML or a screenshot. Unlabeled buttons and unnamed inputs read as "unknown" to a bot.
• Cumulative Layout Shift (CLS) — Confirms the page stays still while an agent reads it. Agents that take screenshots get confused when your layout keeps shifting as images and fonts load in.
• llms.txt — The presence and validity of an llms.txt file at your domain root: a plain-text summary of your site and links to your key pages, written for AI systems.

The broader Lighthouse category includes more checks (such as WebMCP tool annotations — declarative APIs an agent can call), but those three are what most site owners will see first.

What it does NOT do — read this part

It's easy to turn this into fear-marketing, so let's be precise:

• It does not affect your Google Search rankings. It's a separate section that doesn't feed your Performance, Accessibility, Best Practices, or SEO scores.
• It's experimental / "under development." Google has been clear it won't behave perfectly yet, and not every audit applies to every site.
• You won't fail just because you haven't bolted on AI features. A simple, clean site can score well.

So why care? Because this is the first time Google has put agent-readiness on the same screen as speed and SEO — and it's a leading indicator. The businesses that get ahead of it are the ones AI agents will actually be able to research, trust, and transact with as agentic browsing goes mainstream.

Where most sites stand today

In practice, the majority of sites we test score 2/3 or lower, and the reason is almost always the same: no llms.txt file. Accessibility trees are frequently messy, and layout shift is nearly universal on image-heavy pages. A site can have a perfect 100 SEO score and still fall short here, because these checks measure something SEO never did.

What to do about it

First, see where you stand. You can run your own URL through Google's free tool, or use our free Agent-Readiness Check — it wraps the same Google Lighthouse engine and translates the result into plain English.

Then fix the gaps in order. We wrote a step-by-step guide: How to Score 3/3 on PageSpeed's Agentic Browsing Check. If you'd rather have it done for you — full-site diagnosis, competitor benchmark, and a prioritized fix plan — that's exactly what our Agent-Readiness Audit delivers.

Sources: Google Lighthouse 13.3 release; DebugBear, "Lighthouse Has A New Agentic Browsing Category"; Google PageSpeed Insights documentation.