Does Google Use llms.txt? It Depends Which Google You Ask

Does Google use llms.txt? For search ranking and AI Overviews, no. Google's own guidance lists llms.txt as a tactic you do not need. But Chrome's Lighthouse tool now ships an audit that checks whether your site has one.

Same company, two answers. This week LinkedIn has been arguing about it as if that is a contradiction. It is not. The two answers come from two different teams documenting two different things: getting cited in AI search, and being navigable by AI agents. Once you separate those, the guidance lines up cleanly. Here is the full picture.

What llms.txt actually is

llms.txt is a proposed file that sits at the root of your domain, at yoursite.com/llms.txt. It is a Markdown file that gives an AI model a curated, plain-text map of your most important content, so the model does not have to wade through your full HTML, navigation, and scripts to understand what your site is about.

The idea was proposed in 2024 by Jeremy Howard, co-founder of Answer.AI. The mental model: it is to language models roughly what robots.txt and sitemap.xml are to crawlers, a small file at a known location that helps a machine understand your site faster.

The important caveat, and the source of most of the confusion: llms.txt is a community proposal, not a standard any major AI company has formally committed to honoring. Whether a given engine reads it is up to that engine. Right now, most of the big ones do not.

One more distinction matters, because the rest of this argument keeps tripping over it. An llms.txt file is not the same thing as a Markdown version of a page. llms.txt is a single index file at your domain root. A Markdown page is just a plain-text rendering of one specific page, served at its own URL. Some sites, including Google's own developer documentation, publish Markdown versions of individual pages. That is a different thing from adopting llms.txt, and conflating the two is what makes the "Google does it itself" argument sound stronger than it is.

What Google Search says: skip it

Google's position on llms.txt for Search is not vague. In its 2026 AI search guidance for site owners, Google includes a "mythbusting" section listing tactics that do not help with its generative AI features. llms.txt is named directly. Google's Gary Illyes has confirmed publicly that Google does not support llms.txt and has no plans to.

The reasoning is straightforward. Googlebot already renders and reads your real HTML. A separate Markdown file restating that content is redundant, and it is unverifiable, Google cannot confirm the summary actually matches the live page. That is the same loophole that retired the old keywords meta tag two decades ago. A signal a site owner can write to say anything, with nothing checking it against reality, is a signal a search engine learns to ignore.

So for the things most people mean when they say "AI search," appearing in AI Overviews, AI Mode, and Google's ranked results, llms.txt does nothing. It is not a ranking factor and it is not a citation factor.

What Chrome's Lighthouse says: it checks for it

Here is where the screenshots come from. Chrome's Lighthouse, the auditing tool built into Chrome DevTools, added a new audit category called Agentic Browsing. One of its audits checks whether your site serves an llms.txt file. That is a different Chrome team from Search, documenting a different thing.

It is worth reading what that audit actually does, because the detail kills the panic:

• It is optional. In Google's own words, providing the file is "optional at the moment." If your server returns a 404, the audit is marked Not Applicable, not failed.
• The category is not even scored. Google states that "unlike other Lighthouse categories, the Agentic Browsing category does not have a weighted average score." There is no 0 to 100 number to chase here.
• It is explicitly built for agents, not search. The docs describe llms.txt as "a machine-readable summary of a website's content, specifically designed for LLMs and AI agents," noting that "without this file, agents may spend more time crawling the site."

The other audits sitting alongside it in that category make the intent obvious. They check for registered WebMCP tools, accessibility for agents, and layout stability, all of which are about an AI agent navigating and operating your page, not a search engine ranking it. The Agentic Browsing category, in Google's framing, "evaluates how well your site is constructed for machine interaction."

Search engines read the web. AI agents operate it. llms.txt is being built for the second job, and sold as if it helps with the first.

Does Google use llms.txt on its own sites?

Here is the objection that has driven most of the noise: if llms.txt does nothing, why does Google use it on its own sites? Start with the premise, because half of it is wrong. Google's documentation site does not publish an llms.txt file at all. Request developers.google.com/llms.txt and you get a 404. What Google's developer docs do publish is Markdown versions of individual pages, the separate thing described above.

So the real question is narrower: why does Google serve Markdown copies of its documentation? In May 2026, Google's John Mueller answered that directly in a public thread, and it is worth reading closely.

His short version: "it's not done for search. There's more to websites than just SEO." The longer version is the distinction this whole article is built on. Mueller separates "discovery", being found by a search engine, from "functionality", helping a visitor or an agent do the task once they have arrived. Two different jobs. Markdown and llms.txt, where they help at all, help the second one.

So why does Google's developer site serve Markdown? Because AI coding tools are now everywhere, and they parse reference documentation constantly. A Markdown copy of a docs page is cheaper for those tools to ingest. Mueller is blunt that this is a convenience, not a necessity: "OF COURSE they can read HTML just fine," he wrote, calling the Markdown copies "more of a temporary crutch, perhaps to save some tokens."

And he was just as blunt that it does not generalize. "For non-developer sites, I don't think this makes much sense," he wrote, "even with more agentic traffic in the future." His example: "Making a markdown version of a shoe's specs is not going to get you more sales." On llms.txt specifically he went further, saying he has yet to see any AI service, agentic or not, actually use llms.txt files regularly on the sites he has access to, despite the file being proposed back in 2024.

His closing line is the one to keep: "Prioritize needs before dreams." Preparing for a future where agents are everywhere is the dream. Your site has real SEO and conversion work that needs doing now. That comes first.

Why this is not a contradiction

Mueller's discovery-versus-functionality split is the whole resolution. Apply it to the two surfaces everyone has been holding up against each other.

A search engine reads the web to answer a question. An AI agent operates the web to complete a task: it fills the form, clicks the button, books the appointment. Those are different jobs, and they need different tooling. llms.txt is being built for the agent job. Google Search guidance is about the search job. Both statements are true at the same time.

The reason this lands as a "Google contradiction" is that Google never put the two pieces of guidance in the same place. Search documentation lives on one developer property; the Lighthouse documentation lives on another. Nobody wrote the sentence "note: the other team means something different." So the internet found the gap and filled it with noise.

What you should actually do

Three things, in priority order.

1. Do not add llms.txt expecting AI citations

It will not move your position in AI Overviews, and there is no confirmed evidence it changes whether ChatGPT, Claude, or Perplexity cite you. When we measured AI citations across four engines and compared the tools that track them, the brands winning were winning on content depth and entity clarity, not on a text file. Anyone selling llms.txt as an AEO tactic is selling something the data does not support.

2. Adding one is cheap, but it is not a priority

A clean llms.txt costs an afternoon, or about ten minutes if you generate it with Claude Code, and carries no real downside, so if you want to be early on agentic browsing, it is a defensible low-cost bet. But be honest about where it sits. Google's own John Mueller would tell you not to bother for most sites: prioritize needs before dreams. Both things are true at once. It is a near-zero-cost optional experiment, and it is nowhere near the top of your list. File it under "future-proofing," not "ranking," and never let it feel like the AEO box is now ticked. For a worked example of what going further on agentic browsing actually looks like, see how I made propsaasgrowth.com agent-native in an afternoon against Cloudflare's agent-readiness audit.

3. Spend the real effort on what actually moves citations

The work that demonstrably lifts AI citation rate is structural, and it is the same work that has always earned organic visibility: direct-answer openings, FAQ blocks, named-entity binding, and off-site mentions in sources AI engines trust. That is covered in depth in our guide to AEO versus SEO for B2B SaaS, and engine-by-engine in our playbook on how to rank in ChatGPT. None of it is a file you can upload in an afternoon, which is exactly why a file you can upload in an afternoon keeps getting oversold.

For the broader technical foundation, see our Technical SEO service: CWV, schema, internal linking and AI crawler access.

Frequently asked questions