BridgeToAgent
Get the kit — $49
All posts
Audit9 min read

Lighthouse Agentic Browsing FAQ — every common question, answered

A direct-answer FAQ for Chrome's Lighthouse Agentic Browsing category. What it measures, how to run it, why a score reads "0", what passes vs. what informs, the Chrome-version timeline, the relationship to SEO Lighthouse, and the questions teams ask after their first failing report.

BridgeToAgentEditorial team

Lighthouse Agentic Browsing FAQ

If you've run Chrome's Lighthouse Agentic Browsing category and want a direct, technically accurate answer to one specific question — not a walkthrough, not a sales pitch — this is the lookup post. Each question below is a real query searchers run after seeing the audit for the first time. The answers are short, source-anchored where they can be, and point at deeper references when a single paragraph isn't enough.

The post sits between four pillars: the audit-by-audit walkthrough that explains what each check does, the every audit, every fix reference that ships the remediation recipe per audit, the score-reading guide that unpacks what the per-audit detail panel means, and the kit audit-mapping post that draws the honest line between what a $49 kit closes and what stays in your CMS. If your question is broader than a single Q below, those four are the right next reads.

What it is

What is Lighthouse Agentic Browsing?

Lighthouse Agentic Browsing is a Chrome Lighthouse category that sits alongside Performance, Accessibility, Best Practices, and SEO. It scores how easy your site is for an AI agent to read, understand, and act on. The category contains nine audits — three file-presence checks, two file-correctness checks, two discovery checks, one content-structure check, and one emerging-spec check.

The category shipped behind a flag in late 2025 as Lighthouse 12.5. It moved into the Lighthouse default config on May 7, 2026 with Lighthouse 13.3.0. The authoritative spec lives at developer.chrome.com/docs/lighthouse/agentic-browsing/scoring.

What does Lighthouse Agentic Browsing measure?

Nine audits. The full per-audit reference is in Lighthouse Agentic Browsing — every audit, every fix, but the short list:

  • File presence — is /llms.txt, /agents.json, and /agent-instructions.md reachable at your domain root?
  • File correctness — does /llms.txt parse against the llmstxt.org reference parser, and does every action in /agents.json have typed parameters?
  • Discovery — does your homepage <head> contain three <link rel="alternate"> tags pointing at the kit files, and is /sitemap.xml referenced from /robots.txt?
  • Content structure — do your homepage and primary page templates emit Schema.org JSON-LD blocks at minimum density?
  • Emerging — do interactive elements have WebMCP per-element annotations? (intentionally low-weight; nobody passes this in 2026)

What's the difference between Lighthouse Agentic Browsing and regular Lighthouse?

Regular Lighthouse measures site quality for human visitors using a browser — page load speed, accessibility for assistive tech, security best practices, search-engine-readiness. Agentic Browsing measures site quality for AI agents reading and transacting on the site. The category is additive: a site can score 95 on Performance and 15 on Agentic Browsing on the same run. Both categories appear in the same report. Optimizing one does not move the other.

Does Lighthouse Agentic Browsing replace SEO Lighthouse?

No. They check different things. SEO Lighthouse audits meta description, document title, link text, viewport, and hreflang. Agentic Browsing audits the kit files, typed actions, auto-discovery tags, and Schema.org density. They overlap on sitemap discoverability and Schema, but the spec sets and the failure modes diverge. Most sites need both.

How to run it

How do I run Lighthouse Agentic Browsing?

Three surfaces.

The fastest is PageSpeed Insights — enter your URL and PSI returns the audit alongside Performance, Accessibility, Best Practices, and SEO. PSI inherits the Lighthouse default config within two weeks of each Lighthouse release, so any PSI run since late May 2026 includes Agentic Browsing.

The most flexible is Chrome DevTools on Chrome 150 or later — open DevTools, switch to the Lighthouse panel, and run an analysis. The category is on by default.

The CI-friendly is the CLI: npx lighthouse@latest <url> --only-categories=agentic-browsing returns the JSON output a build pipeline can gate against. The CLI deep-dive post covers the JSON shape.

Is Lighthouse Agentic Browsing in Chrome 130 / 140 / stable?

The category exists in Chrome 130 onward but stayed behind the experimental Categories toggle in the DevTools Lighthouse panel through Chrome 149. Chrome 150 ships it on by default. PageSpeed Insights inherits the default config within two weeks of each Lighthouse release, so PSI runs from late May 2026 onward include the audit regardless of which Chrome the user is running.

What if my Lighthouse Agentic Browsing report is empty?

You're on a Lighthouse build older than 12.5, or the category toggle is off in the DevTools panel. Update Chrome to 150 or later, re-open the Lighthouse panel. If the category still doesn't appear, you may be on a Lighthouse-CI container pinned to an older version or a third-party wrapper that hasn't refreshed. Fall back to npx lighthouse@latest <url> --only-categories=agentic-browsing which always pulls the newest spec.

Reading the score

Why is my Lighthouse Agentic Browsing score 0?

Two common causes. The kit files are absent — no llms.txt, no agents.json, no agent-instructions.md. Most pre-2025 sites have none, which fails six audits outright. Or you're running a Lighthouse build that emits a synthetic single-number score where file-presence dominates; missing files drag the weighted total close to zero.

The score-reading guide unpacks what each per-audit detail means once the score isn't zero anymore.

Why doesn't Lighthouse Agentic Browsing show a single score?

By design. The Chrome team's documentation: "the Agentic Browsing category does not have a weighted average score from 0 to 100. Because the standards for the agentic web are still emerging, the current focus is to gather data and provide actionable signals rather than a definitive ranking."

You get per-audit pass, fail, and warning markers. Some third-party Lighthouse wrappers compute a synthetic weighted score for display convenience, but the official category emits per-audit signals only.

Does Lighthouse Agentic Browsing affect Google Search rankings?

Not directly as of mid-2026. Google hasn't announced that Agentic Browsing signals feed the traditional web-search ranker. What is changing is the downstream surfaces — AI Overviews, ChatGPT search, Perplexity, Gemini AI Mode — that read structured data and kit files when deciding whose content to cite. The audit is a signal volume change for agent surfaces, not a Google web-rank lever.

What the audits actually check

What is llms.txt and why does Lighthouse check for it?

llms.txt is a plain-text Markdown file at your domain root that lists the canonical URLs an AI agent should read when summarizing or citing your site. It functions as a curated reading list — the model fetches the file first, then picks the right page from the list instead of guessing from your navigation. Lighthouse checks two things: presence (file reachable at /llms.txt with a text/plain content type) and well-formedness (parses cleanly against the llmstxt.org reference parser).

The llms.txt vs robots.txt comparison covers how the file relates to existing crawler-control conventions.

What is agents.json and why does Lighthouse check for it?

agents.json is a typed action manifest at your domain root that tells an agent what it can do on your site — search, request a quote, contact, browse a category, view a policy. Lighthouse checks presence (valid JSON at /agents.json) and typing (every declared action has a proper JSON Schema parameter block). The agents.json vs WebMCP comparison covers the three-layer mental model.

What is the difference between agents.json and llms.txt?

llms.txt is for reading — it tells an agent which URLs to pull when answering a user question about your site. agents.json is for acting — it tells an agent which endpoints to call when a user wants to do something. Different files, different audiences, different shapes. Lighthouse checks them as separate audits. Sites that ship one without the other still fail the other.

Why is my agents.json failing the audit if the file exists?

Almost always because of agents-json-actions-typed. The presence audit accepts any valid JSON at /agents.json. The typing audit requires every action's parameters block to be a JSON Schema object with explicit type declarations. Common failure modes: parameters set to null, type field missing, type value not one of string, number, boolean, array, object. The actions-typed deep-dive walks the seven most-common typing mistakes.

Can I pass Lighthouse Agentic Browsing without WebMCP?

Yes. The webmcp-annotations audit is intentionally low-weight because the WebMCP spec is still moving and almost no major CMS or theme ships annotated output natively. Most sites in 2026 fail this audit; Lighthouse treats it as informational rather than load-bearing. Closing the other eight audits is a clean passing read.

Improving the score

How do I improve my Lighthouse Agentic Browsing score?

In order of return on effort:

  1. Upload llms.txt, agents.json, and agent-instructions.md to your domain root. Closes three audits in roughly thirty minutes if you hand-write them.
  2. Validate llms.txt against the llmstxt.org reference parser. Broken syntax is the second-most-common failure.
  3. Type every action in agents.json with proper JSON Schema. No null parameters.
  4. Add three <link rel="alternate"> tags to your homepage <head>.
  5. Add a Sitemap: line to robots.txt.
  6. Work on Schema.org density on your homepage and primary templates (this is the audit with the longest tail; see the schema-org-density deep-dive).

The kit-to-audit mapping post draws the line between what a $49 install closes outright and what sits in your CMS.

What's the lowest-effort way to improve my score?

Add three <link rel="alternate"> tags to your homepage <head> and add a Sitemap: line to robots.txt. Closes two audits in ten minutes without uploading any new files. The auto-discovery-links post has the exact snippet for every major CMS.

How long does it take to pass Lighthouse Agentic Browsing?

Six of nine audits in an afternoon if the files are hand-written. The two open-ended audits — Schema density and WebMCP — range from a plugin install to a content audit (Schema) or are best deferred entirely until the spec stabilizes (WebMCP). The 11-site before-and-after cohort reports a modal lift from a starting score around 17 to a passing read around 72 after the kit install lands.

Where do I get llms.txt and agents.json?

Three paths. Hand-write per the spec (llmstxt.org for the reading list, the agents.json reference for the action manifest). Use a generator — the BridgeToAgent kit crawls your live site, extracts your URL inventory and action surface, validates against the reference parsers, and ships the three files plus the auto-discovery snippet. Or use a free generator and validate the output yourself — free generators often emit malformed files that pass the presence audit and fail the well-formed audit.

Coverage and edge cases

What Lighthouse version added Agentic Browsing?

Lighthouse 12.5 in late 2025 as an experimental category. Lighthouse 13.3.0 (May 7, 2026) moved it into the default config. The audit names and shapes have been stable since 12.5; what changed in 13.3.0 was visibility, not the spec.

Does PageSpeed Insights include Agentic Browsing?

Yes, since late May 2026. PSI wraps a server-side Lighthouse run and inherits the default config within two weeks of each Lighthouse release. Lighthouse 13.3.0 shipped May 7; PSI was returning Agentic Browsing audits by May 21.

Does Lighthouse Agentic Browsing work for non-English sites?

Yes. None of the nine audits inspect language. The kit files are language-agnostic, Schema.org JSON-LD has no language dependency, and the audit returns the same pass/fail signals across language locales.

Which audits does Lighthouse Agentic Browsing check?

The nine in the default config: llms-txt-present, llms-txt-well-formed, agents-json-present, agents-json-actions-typed, agent-runbook-present, auto-discovery-links, sitemap-discoverable, schema-org-density, webmcp-annotations. The per-audit fix recipe for every one of them is in Lighthouse Agentic Browsing — every audit, every fix.

What this FAQ doesn't cover

Three honest scope notes.

This FAQ doesn't replace the per-audit fix reference. If a specific audit is red, the every audit, every fix post walks the cause and the remediation per audit — that's the lookup, this is the framing.

This FAQ doesn't compare Lighthouse to other scoring frameworks. The scoring-frameworks reference and the Cloudflare-vs-Lighthouse pillar cover that surface.

This FAQ doesn't cover platform-specific install steps. The install matrix lists all 13 supported CMSs with the per-platform paste path, and the four platform checklists (Shopify, WordPress, Webflow, Squarespace and Wix) walk the install on each.

The audit is stable. The remediation is documented. The only step left is running it.

Related

Keep reading

All posts →