Is Your Website Ready for AI Agents? Google's New Lighthouse Audit, Explained

What changed

AI agents are no longer just reading websites — they're starting to use them. Google describes AI agents as autonomous systems that can perform tasks on behalf of people, such as booking a reservation, and its guide to optimizing for generative AI features now points site owners toward agent-friendly website practices.

To make that measurable, Chrome added a new Agentic Browsing category to Lighthouse, the auditing tool built into Chrome DevTools. It's available from Chrome 150 and runs a set of deterministic checks on how easily an agent can understand and operate your page. This guide covers what those checks look at, how to run them, and which fixes are worth your time — most of which also make your site better for people.

How AI agents "see" a web page

According to Chrome's guidance on agent-friendly websites, browser agents combine three views of your page: screenshots, the HTML, and the accessibility tree — the structured version of the page that browsers build for assistive technology like screen readers. Agents use the DOM and accessibility tree to get a clean list of the interactive elements on the page, then cross-check it against the visual rendering to understand layout and grouping.

That has a practical consequence: if your "Book now" button is a styled <div> with no accessible name, it may look like a button to a person but not register as one to an agent.

What the Agentic Browsing category checks

Area	What Lighthouse checks	Why it matters to agents
Accessibility for agents	A subset of the accessibility audits — e.g. every interactive element has a programmatic name, roles and parent-child relationships are valid, and interactive content isn't hidden from the accessibility tree	The accessibility tree is the agent's primary map of what it can click and fill in
Layout stability	Cumulative Layout Shift (CLS)	Elements that move after the page loads cause misclicks
WebMCP	Registered WebMCP tools, forms missing declarative WebMCP, and schema validity	Lets a site describe its actions to agents directly instead of making them work it out visually
llms.txt	Whether /llms.txt can be fetched from your domain root without a server error	Without a machine-readable summary, agents may spend more time crawling to understand the site's structure

How the score works

Unlike Lighthouse's Performance or SEO categories, Agentic Browsing doesn't produce a weighted 0–100 score. It shows a pass ratio — how many of the checks your page passes — along with pass/fail results and warnings for individual audits. Chrome describes the category as informational and unbenchmarked for now, because the standards for the agentic web are still emerging.

How to run it

1. Open the page you want to test in Chrome 150 or later.
2. Open DevTools and go to the Lighthouse panel.
3. In the categories list, tick Agentic Browsing — it's off by default, so it won't appear in a standard report unless you select it.
4. Run the report and open the Agentic Browsing section.

Test your most important task pages, not just the homepage: a booking form, a contact form, a product page with an add-to-cart button. Those are the pages an agent is most likely to be sent to do something on.

What to fix first

1. Give every button, link, and form field a name

This is where most sites fail, and it's the same fix that helps screen-reader users. Chrome's documentation puts it plainly: missing labels can block both users with visual disabilities and agents from completing a task. The common offenders:

• Clickable <div> and <span> elements. Use a real <button> for actions and <a> for navigation — agents recognize those as interactive.
• Icon-only buttons — a magnifying glass, a cart icon, a hamburger menu — with no text or aria-label.
• Form fields without connected labels. Link each <label> to its input with the for attribute so the field's purpose is explicit.
• Image links with no alt text. When a link contains only an image, the image's alt text is what names the link. Our alt text guide covers how to write it.

2. Stop the layout from jumping

Agents that work from screenshots get confused when elements move. The usual causes of layout shift are images without set dimensions, ads, and content injected after the page loads. Reserve space for them by setting width and height on images and fixed-size containers for embeds. Keeping key controls in consistent places across similar pages — the add-to-cart button in the same spot on every product page, for example — helps too.

3. Make interactive elements visible and obviously clickable

Chrome's guidance recommends that elements an agent needs to use are visible and larger than 8 square pixels, so they aren't filtered out of the model's visual analysis, and that clickable elements use cursor: pointer in CSS as a signal that they're actionable.

4. Add an llms.txt file — and make sure it actually loads

Lighthouse treats llms.txt as optional: if your site doesn't have one, the audit is marked not applicable rather than failed. It flags a problem when requesting the file returns a server error. If you add one, keep it at your domain root and check that it loads. Our llms.txt guide covers the format, and the free llms.txt generator drafts a starter file from your homepage for you to review and edit.

5. Consider WebMCP later — unless you build web apps

WebMCP is a proposed open web standard that lets a site expose its actions to browser agents as structured "tools": a declarative API for standard actions defined directly in HTML forms, and an imperative API in JavaScript for more complex interactions. It's in early preview, with an origin trial that started in Chrome 149. If your site is mostly content plus a contact or booking form, fixes 1–3 will matter far more today. If you build a web app or a multi-step checkout, it's worth tracking now.

What this is not

• Not a Google Search ranking factor. Lighthouse is a diagnostic tool, Chrome calls this category informational and unbenchmarked, and Google hasn't said Agentic Browsing results feed into Search rankings. Google's guidance for AI features in Search is still that standard SEO fundamentals apply — see our GEO vs SEO guide.
• Not a guarantee an agent will succeed. The checks are deterministic tests of specific signals. A perfect pass ratio doesn't mean every agent will complete every task on your site; it means you've removed known obstacles.
• Not the same as your AI crawler settings. Whether AI crawlers such as GPTBot or ClaudeBot can fetch your pages is controlled by robots.txt — a separate question from whether an agent can operate a page once it's there.

Why this is worth doing anyway

Almost everything on this list — real buttons, labeled form fields, alt text, a stable layout — is long-standing accessibility and usability advice. Agent readiness gives it a new reason, but the work pays off today for every visitor using a screen reader, a keyboard, or a slow connection. It's no-regret work, whether or not agents end up sending you customers.

Start with the foundation

Peak Visibility's free checker covers several of these signals on your live page — image alt text, whether an llms.txt file exists, and whether your robots.txt blocks AI crawlers like GPTBot or ClaudeBot — so you can clear the basics before running a full Lighthouse audit.

FAQ

Does the Agentic Browsing score affect my Google rankings?

Google hasn't said it does. Chrome describes the category as informational and unbenchmarked, and Google's guidance for appearing in AI features on Search remains standard SEO fundamentals: be crawlable, indexable, and genuinely useful.

Will Lighthouse fail my site if I don't have an llms.txt file?

No. Lighthouse treats llms.txt as optional — a missing file makes the audit not applicable. It flags an issue when fetching the file returns a server error.

Where do I find the Agentic Browsing category?

In the Lighthouse panel of Chrome DevTools, starting with Chrome 150. It's off by default, so tick it in the categories list before running the report.

Do I need to implement WebMCP now?

For most small business and content sites, no. WebMCP is a proposed standard in early preview. Accessible names, labeled forms, and a stable layout fix the problems that affect the most agents and people today.

Is this the same as the Lighthouse Accessibility score?

It overlaps. Agentic Browsing uses a subset of the accessibility audits that matter most for machine interaction, then adds layout stability, WebMCP, and llms.txt checks. A strong Accessibility score is a good head start.