Guides › What Is llms.txt? A Straight Answer

What Is llms.txt?

By Michael Emery · reviewed against the measurement record · 30 July 2026

llms.txt is a plain Markdown file, published at the root of a domain (yourdomain.com/llms.txt), meant to give an AI system a short, structured summary of a site: what it is, what it offers, and links to the pages that matter most. The idea is a proposed convention, not a web standard ratified by anyone. There is no governing body and no browser requirement behind it, only a format that a growing number of sites have chosen to adopt.

The spec, in full

The format is deliberately small:

• A single H1 with the site or company name.
• An optional blockquote with a one-line summary.
• Free-text sections in Markdown.
• Optional H2 sections containing Markdown link lists, pointing to further pages (often a fuller llms-full.txt for long-context readers).

That is the entire requirement. Nothing enforces it: a file can be malformed, outdated, or missing the H1 entirely and still sit at /llms.txt without error. The one place that requirement now gets checked mechanically is Google's own audit tooling, covered below.

How many sites actually publish one

Adoption is real but still a minority behavior: roughly one in ten sites now publish an llms.txt, per SE Ranking's survey of 300,000 domains. In our own nine-site census, seven of nine already had one published, but not one of the seven had placed a pointer to it where our testing shows it actually gets found (Note 09). Writing the file turns out to be the easy half.

The two-sided Google reality

Both of the following are true at the same time, and neither cancels the other out.

Google Search does not use it for rankings or AI features. Google's AI-optimization guidance, updated 15 June 2026, states plainly: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them." Google has also named llms.txt in a June 2026 Search Central mythbusting update, and Google's own Gary Illyes has said there is no support and no plan to add it; John Mueller has compared it to the keywords meta tag, a file the site operator controls and search engines learned decades ago to ignore.

Google's Lighthouse tooling now grades it anyway. PageSpeed Insights and Lighthouse added an "Agentic Browsing" audit category (verified July 2026) that checks, among other things, whether llms.txt exists and follows the expected Markdown-with-H1 format. That is a page-quality grade, not a ranking signal. See our explainer on the Agentic Browsing audit for the full breakdown of what it checks.

So: not a ranking mechanism, and simultaneously something Google's own tools now score you on. Both statements are accurate; they are just answers to two different questions.

What we measured when real agents try to use it

The Google question is about crawlers and rankings. A separate, and for most businesses more urgent, question is what happens when an AI agent with a browsing tool (the kind fielding a buyer's question about your business) actually tries to use your llms.txt. We ran controlled tests to find out, and the file performed worse than most advice assumes.

Unpointed, it is rarely found. Across corrected test runs on a real client site with no visible signpost to its llms.txt, agents reached the file 0 times in 18 visits (Note 15). Earlier results suggesting agents "guess" the conventional path on real sites turned out to be an artifact of a fetcher bug reading garbled binary noise instead of the actual page; corrected, the guess-rate is zero (Note 15). Head-tag pointers (<link rel="alternate">, <meta name="llms-txt">) scored even worse than nothing, because browsing tools receive extracted visible text, and head markup does not survive extraction (Note 06).

Even when it is fetched, agents often don't use what's inside. On a real site whose llms.txt contains an actual price, six of ten model runs fetched the file, and zero of those ten reported the price in their final answer (Note 19). Finding the file and using its contents turned out to be two different events. A file reached late in a long crawl rarely makes it into the answer the agent actually gives.

The fix that works

Putting the same fact directly in the first few hundred words of the page, not in a linked file, took recovery from 30% to 78% in our strongest test, with agents needing an average of 1.1 page fetches instead of hunting for a file (Note 41). That is the mechanism that moves the number. llms.txt is not it.

So should you still publish one?

Yes, for reasons that have nothing to do with driving traffic. It is inexpensive to generate, it is the artifact Google's audit tooling now checks for, and it is a real resource for the agents that do go looking for it once they've exhausted your visible pages. Treat it as a compliance file that costs little and helps you pass a grading check, not as the thing that gets your business found or correctly described. That work happens on the page itself. If you need the file itself, our free llms.txt generator and validator build and check one in the correct format.

Get your numbers

Run a free audit and see whether an AI agent can actually find and use the facts on your own site: no assumptions, a measured recovery score.

Request your free AI visibility audit · or see what the full agent-readiness package includes.