Google Just Added an AI Audit to Chrome. Then Said You Probably Do Not Need It.
Chrome Lighthouse 13.3 now checks whether your site has an llms.txt file. Google says it does not need one to rank in AI search. The contradiction is the story, and it tells marketers something useful about the difference between AI search and AI agents.
Search visibility and agent readiness are now two separate engineering problems. They share infrastructure but they answer different questions.
Google has added an llms.txt check to Chrome's Lighthouse tool. The new audit lives inside an emerging category called Agentic Browsing, which evaluates whether sites are structured for machine interaction by AI agents. Lighthouse 13.3 shipped this week with the category enabled by default. Sites that do not serve an llms.txt file get flagged as Not Applicable for that specific check.
The odd part is that Google's own search team has repeatedly said llms.txt is not needed for AI search visibility. The new Lighthouse audit does not contradict that position. It is targeted at a different problem. Search engines crawl and rank. Agents browse, click, fetch and act on a user's behalf. The two are starting to need different things from your website.
The Agentic Browsing category in Lighthouse covers four checks: WebMCP integration, agent accessibility, layout stability and llms.txt. It does not produce a traditional 0 to 100 Lighthouse score. Instead it returns a pass ratio and individual pass or fail flags. Google has labelled the whole category as under development, which means the criteria will move as the agent ecosystem matures.
Why it matters
Most Australian businesses still think about AI search as a single thing. It is not. There is AI Mode and AI Overviews inside Google Search, where the ranking signals look a lot like classic SEO with added emphasis on citation-worthy content. Then there are agentic browsing experiences, where a chatbot or an AI assistant navigates the web on a user's behalf to complete a task. The second category needs structural signals like llms.txt, WebMCP endpoints and clean machine-readable content layouts.
The Lighthouse audit is Google's way of telling developers and SEOs that a second technical track is opening up. You do not need llms.txt to rank. You might need it to be useful to an agent that is trying to book a service, compare a product or pull a price from your site.
Buying decisions for AU businesses are starting to flow through both surfaces. The gap between sites built for human visitors and sites built for AI agents is now visible inside Chrome's own tooling.
Lighthouse's new Agentic Browsing category audits WebMCP integration, agent accessibility, layout stability and llms.txt
What to do about it
Run Lighthouse 13.3 against your site and look at the Agentic Browsing report. Not for the score. For the gaps. Identify which of the four checks fail and which apply to your business model.
If you sell products or services that benefit from being included in agent comparisons, add llms.txt. There is no penalty for having it. There is no cost to creating it. The downside risk is zero.
Do not treat llms.txt as an AI search ranking play. Google has been explicit. It is an agent readiness signal, not a search signal.
Keep your SEO programme focused on the same things that already work. Useful content, technical health, citation-worthy data, AU market specificity. The fundamentals have not changed because Lighthouse added a new audit.
Track which AI surfaces actually send traffic to your site. AI Mode, ChatGPT, Perplexity, Claude, Gemini and agent referrals each leave a footprint in your analytics if you are looking for them.
The businesses that win in agentic browsing will be the ones who treat AI agents as a distinct audience with distinct technical needs. The rest will keep optimising for last year's web.