Google adds llms.txt testing to Chrome Lighthouse

Google’s new Lighthouse “Agenttic Browsing” test now checks for the presence of the llms.txt file. The new Lighthouse test documents make llms.txt as an availability and signal efficiency for AI agents, not a general crawl directive.

• The audit is part of Chrome’s “Agent Browsing” category, which checks whether sites are designed for machine interaction.
• This document comes less than a week after Google published new guidelines for developing search AI features such as AI Overview and AI Mode, where it said you don’t need llms.txt files in the legend section of its new guide for setting up productive AI features.

What is Lighthouse testing now. Lighthouse’s Agetic Browsing section evaluates “how well your site is optimized for machine usage” using a deterministic test, according to Google’s documentation. Among the checks:

• WebMCP integration.
• The integrity of accessibility medicine.
• Structural stability with CLS.
• Existence of llms.txt file.

Lighthouse checks for “the presence of a machine-readable hash at the root of the domain.” Google also explained why the file is important to agents:

“Without llms.txt, agents can spend more time analyzing the site to understand its high-level structure and main content.”

The audit stage does not produce a standard Lighthouse score (0-100). Instead, Google looks at a partial pass rate and a pass test that is consistent with signs of agent readiness.

Tension. The new Lighthouse documentation doesn’t directly contradict Google’s advice on optimizing your website for productive AI features because these tests focus on AI agents and browser tools, not Google’s Search rankings. However, seeing llms.txt mentioned in Chrome’s readiness test may cause some SEOs to rethink previous doubts about the file.

Improving the Agent engine. The Lighthouse test also aligns with Google Cloud AI engineering director Addy Osmani’s comments in April about Agentic Engine Optimization. Osmani said AI agents with limited context windows may skim long pages or miss important information buried deep in the content. Among his recommendations:

• Pure semantic structure.
• Effective token content.
• Markdown delivery.
• llms.txt layers to find.
• Power signature files such as AGENTS.md.

SEO vs. llms.txt. Here are Google’s recommended Mythbusting search AI productivity don’ts:

• LLMS.txt files and other “special” markups: You don’t need to create new machine-readable files, AI text files, markup, or Markdown to appear in the AI ​​search generated. Note that Google may find, crawl, and index more file types than HTML on a website: this does not mean that the file is treated in a special way.

Here’s what Google’s John Mueller had to say about Google using llms.txt, in response to Lily Ray asking him on Bluesky “Hey @johnmu.com – if you can answer, many people express the irony that Google uses LLMs.txt files, and markup pages, despite the fact that these things are not necessary for search performance. (I’m sure I’ll get this question soon!)”:

The short answer is that it’s not for search. There is more to websites than just SEO :-).

The long and flawed version is that it is worth distinguishing “discovery” (finding a website or pages with a global search engine) vs “operation” (there is probably a more accurate term for this, but in reality: once someone has found the page, to help him do the job he wants to do better).

Maybe that’s the same as CTAs on traditional pages? You don’t “do it” in SEO (to get found), but if you are responsible for the overall website, ensuring a high “discovery rate” (SEO) and a high conversion rate helps to justify your work.

To go back to the developers.google.com site, AI coding has become very popular, and these coding programs (I think) can be efficient and accurate in the code they produce if they can easily read / parse reference materials, such as developer documentation.

In those cases, it can help provide them with a way to understand the context of the documents they are viewing, as well as a simplified version of the reference page (eg, in footnotes). Of course they can read HTML well, so this is a temporary solution, maybe to save some tokens.

For non-engineering sites, I don’t think this makes much sense, even more agency traffic in the future (and if you check your logs, you don’t get much of that yet). Making a markdown version of a shoe’s specs won’t get you more sales (competitors appreciate that).

And (I know, no one reads so far), if you think this is important to prepare when agents are everywhere: your site (all sites) have more important things to do with SEO than to prepare for a future situation that may or may not come. Put needs before dreams.

What Google says agents rely on. Beyond llms.txt, Google’s new Lighthouse section puts more emphasis on accessibility and stability of the interface. The documents state that agents rely on the reachability tree as their “core data model.” Lighthouse specifically tests:

• Program labels are interactive elements.
• Valid accessibility tree structure.
• Whether interactive content is hidden from auxiliary programs.
• Structural stability with CLS.

Google also warns that registered WebMCP tools and major DOM changes may affect search results.

Why do we care. Google says you don’t need llms.txt for Search, but Chrome now checks if the file exists. At the same time, Google tools seem to favor sites that are easy for machines to read and use, especially sites with strong accessibility, stable structures, and clear agent access.

Google’s help document. Lighthouse agetic browsing points

Celebrate deeply.