What Lighthouse 13.3’s Agentic Browsing Checks Mean for AEO
Less than the name suggests. Lighthouse 13.3 added an experimental Agentic Browsing category that checks whether AI agents can operate a page on a user's behalf. It looks at whether the accessibility tree is well formed, whether controls have names, whether the layout stays stable, whether WebMCP tools are valid, and whether an llms.txt file is present. It reports a pass ratio, such as 3 of 4, rather than a score out of 100, and it needs Chrome 150 or later. It does not measure whether ChatGPT, Gemini or Google's AI features mention or cite you, which is the core of AEO. Treat it as a usability check for agents. Several of its fixes, especially accessibility and layout stability, also help people and SEO.
What does the Agentic Browsing category check?
Chrome's documentation and early write-ups describe four groups of checks:
| Check | What it tests | Where it comes from |
|---|---|---|
| Accessibility tree | Interactive elements have programmatic names and labels; interactive content is not hidden from the tree | A subset of Lighthouse's existing accessibility audits |
| Layout stability | Cumulative Layout Shift, since agents act on element positions | Existing performance data |
| WebMCP | Whether annotated forms and registered tools match the expected schema; lists every tool on the page | New audit, needs the WebMCP origin trial |
| llms.txt | Whether a file exists at the domain root and follows the proposal | New audit; a missing file counts as not applicable |
Chrome says the audits are deterministic so they can run in automated testing. Results can still vary, for example if WebMCP tools are registered by JavaScript after Lighthouse takes its snapshot, or if large DOM changes alter the accessibility tree.
How is it scored?
Unlike Lighthouse's performance, accessibility, best practices and SEO categories, Agentic Browsing has no weighted score from 0 to 100. It shows how many checks pass, plus pass or fail status and warnings for individual audits. That makes it hard to compare with other scores and easy to over-read: a site with no llms.txt and no WebMCP can still pass the checks that apply to it.
Chrome labels the category and WebMCP support as experimental and based on proposed standards. DebugBear noted in June 2026 that PageSpeed Insights and Chrome DevTools still used an older Lighthouse version, so the category first appeared in the command-line tool and some third-party testers.
What does Chrome's agent-ready website testing mean for SEO?
Very little directly. Google Search does not rank pages on Lighthouse category results, and Google's guidance on AI features says no special files or markup are needed to appear in AI Overviews or AI Mode, and that Search does not use llms.txt. The Agentic Browsing category comes from the Chrome team and is about agents acting in a browser, not about how Google Search indexes or ranks pages.
The overlap is indirect but real:
- Layout stability is also Cumulative Layout Shift, one of the Core Web Vitals that Google uses as a ranking signal.
- Accessible names and labels help screen reader users and make forms easier to complete, which helps conversion.
- Clean, semantic HTML that agents can parse is also easier for search crawlers to understand.
So fixing what the category flags rarely hurts and often helps, but a better pass ratio is not an SEO result in itself.
What does it mean for AEO?
AEO is about being chosen and cited in AI answers. That depends on whether AI crawlers can reach and read your pages, what your content says, and what other sites say about you. The Agentic Browsing category tests none of those. It becomes relevant at the next step: when an agent has chosen your business and tries to book, buy or submit a form on the user's behalf.
Two of its checks deserve caution. llms.txt is a proposal with little evidence that AI systems rely on it; Google says Search ignores it. WebMCP is a proposed standard in an origin trial. Adding either to pass a check is fine if cheap, but neither is a proven route to AI visibility. How WebMCP works is covered in what WebMCP is and whether it matters for AEO.
What should you do with the results?
- Fix accessibility failures first. Unlabelled buttons, fields without labels and interactive elements hidden from assistive technology affect people now.
- Reduce layout shift. Give images dimensions, reserve space for embeds and stop late-loading banners from pushing content down.
- Decide on llms.txt on cost. If your platform generates one, keep it accurate. Do not expect it to change AI answers.
- Consider WebMCP only for task-heavy sites, such as bookings and applications, as covered in how to structure a site so agents can complete bookings.
- Measure AI visibility separately, by checking what AI assistants say about you, as covered in how to audit a website for AI visibility gaps.
Other readiness checkers test similar ground with different emphasis; Cloudflare's is covered in the Cloudflare Agent Readiness score review, and a marketing-focused grader in the HubSpot AEO Grader review.
Who should pay attention now?
- Sites where visitors complete tasks, such as bookings, applications, quotes and checkouts. Agents acting for users will hit the same forms, so the accessibility and stability checks matter most here.
- Agencies, who will be asked about the new category by clients. Explain that it is a usability check for agents, not a visibility score, before it turns up in a report as “AEO readiness”.
- Content and brochure sites can treat it as a routine accessibility review and move on.
How do you run it?
Install the latest Lighthouse command-line tool and run it against a page with Chrome 150 or later. For WebMCP audits, register for the WebMCP origin trial. Some online testers already use the newer Lighthouse version. Run the same pages each time, such as the home page, a product or service page, and a form, and keep the reports so you can compare after changes.
We have not measured whether passing these checks changes how often agents complete tasks on real sites. This page is based on Chrome's Lighthouse agentic browsing documentation and published write-ups of the category.
What might this look like in practice?
Consider an illustrative online learning company, Brightpath Courses, whose marketing lead read that Lighthouse now tests for AI agents and asked the team to “get AEO-ready”. The report showed 2 of 4 checks passing: the enrolment form had three unlabelled fields, and a cookie banner caused a large layout shift. The developer labelled the fields and reserved space for the banner, which also cut form abandonment on mobile. The team left WebMCP for later. Separately, the content team tracked how AI assistants described Brightpath's courses, since the Lighthouse result said nothing about that.
What do people ask about Lighthouse's agentic checks?
Does a good Agentic Browsing result improve Google rankings?
Not directly. Some fixes, such as reducing layout shift, help Core Web Vitals.
Is llms.txt required to pass?
No. A missing llms.txt counts as not applicable, though a server error on that path is flagged.
Why don't I see the category in PageSpeed Insights?
Some Google tools ran an older Lighthouse version after 13.3 was released. Use the command-line tool or a tester on the newer version.
Is the category stable?
No. Chrome marks it experimental and based on proposed standards, so checks may change.
Published by The Kovmere Team · updated Sep 29, 2026 · More resources