Guide

Lighthouse Agentic Browsing Audit Guide

Chrome Lighthouse now includes experimental agentic browsing checks. Learn what llms.txt, WebMCP, and agent readiness audits mean.

Updated May 17, 2026

Chrome Lighthouse’s agentic browsing category matters because it brings agent-readiness checks into a mainstream developer audit workflow. The current checks are experimental, but they signal where the web is heading: AI agents need machine-readable summaries, discoverable tools, and deterministic interaction surfaces.

What changed#

Chrome for Developers now documents Lighthouse audits for agentic browsing. The documented checks include an llms.txt audit and an informational audit for registered WebMCP tools. Chrome says the category is experimental and based on proposed standards, so it should be treated as an early diagnostic layer rather than a final score.

Primary sources:

What the audits check#

AuditWhat it meansPractical AEO response
llms.txtChecks whether the site root provides a machine-readable summaryPublish a concise llms.txt with key links
Registered WebMCP toolsLists tools the page exposes to browser agentsAdd clear tool names and descriptions for actions
Agentic browsing categoryShows pass/fail and informational readiness signalsUse it as a technical checklist, not a ranking score

Chrome’s llms.txt documentation makes an important distinction: a missing file can be marked Not Applicable because the file is optional, but server errors are flagged. That means a broken llms.txt is worse than no file.

Why this matters for AEO#

The AEO readiness audit already checks discovery, read layer, execution layer, and trust. Lighthouse adds a developer-friendly way to catch some of those signals inside a familiar tooling environment.

It does not replace a full AEO audit. It helps developers validate specific technical pieces:

  • Is the site root serving llms.txt cleanly?
  • Are page actions visible as WebMCP tools?
  • Are experimental agent-readiness signals visible during browser testing?

How to prepare llms.txt#

For agentic browsing, llms.txt should be short and useful. It is not a sitemap clone.

A practical file should include:

  • site purpose
  • core concepts
  • best starting pages
  • key tools or services
  • machine-readable resources
  • update date

The llms.txt guide explains how to structure it. The main rule is simple: help an agent understand what to read first.

How to prepare WebMCP tools#

The registered WebMCP tools audit is informational. If no tools are registered, the list is empty. That is not automatically a failure, but it reveals whether the page exposes actions in a browser-agent-friendly way.

Good first tools include:

  • request audit
  • book appointment
  • request quote
  • add to cart
  • check availability
  • submit support ticket
  • compare products

The WebMCP guide covers when to use page-level tools versus server-side MCP.

Suggested workflow#

StepActionOutput
1Run Lighthouse on homepage and important templatesBaseline agentic browsing signals
2Fix llms.txt response errorsStable root machine-readable summary
3Add WebMCP to one low-risk action pageVisible registered tool
4Re-run LighthouseConfirm detection
5Add server-side logs for agent actionsMeasurement layer

This workflow connects technical validation to the execution layer. Passing an audit is less important than proving an agent can complete a useful task.

What not to do#

Do not add a fake llms.txt file just to pass a check. Do not register generic WebMCP tools like submit_form without a meaningful description. Do not expose high-risk actions because the audit exists.

The best agentic browsing implementation is boring: clear content, stable actions, explicit permissions, and good logs.

FAQ#

Is the Lighthouse agentic browsing score a ranking factor?#

No official source says that. Treat it as an experimental readiness signal, not a Google ranking factor.

Is llms.txt required?#

Chrome’s documentation says it is optional at the moment. A missing file can be Not Applicable, while server errors can be flagged.

Should I add WebMCP to every form?#

No. Start with high-value, low-risk actions where agent execution would help users.

How is this different from a normal SEO audit?#

Traditional SEO audits focus on crawlability, speed, metadata, and content. Agentic browsing audits focus on whether agents can understand and act on the site.

What should be measured after implementation?#

Track agent-triggered actions, form completions, failures, invalid requests, and the pages agents read before taking action.