llms.txt: the file, the format and who actually reads it

A convention for putting a markdown file named llms.txt on your site that summarizes it and links to the pages an AI agent would need. Jeremy Howard published the proposal on llmstxt.org and on the Answer.AI blog on September 3, 2024.

The reasoning: web pages are built for people, and an agent wading through menus, scripts and ads wastes its limited context. A small index in markdown, plus clean markdown copies of the pages it links to, gives the agent the facts without the page furniture.

A second version went up on August 10, 2026. By the author's account, thousands of sites now publish the file and documentation platforms generate it automatically. Version 2 lets a file sit at any path and cover the pages under it. It accepts markdown copies at page.md as well as page.html.md, and it recommends advertising both with standard link relations: rel="alternate" type="text/markdown" for a page's markdown copy and rel="describedby" for the llms.txt that covers it.

It also dropped a version 1 idea. The "Optional" section no longer tells tools what to leave out, though it remains a convention for links an agent can skip.

Only the H1 is required. In order, the file holds:

llms.txt is an index; llms-full.txt is the whole content in one file. Only the first is part of the proposal.

llms-full.txt is a convention from documentation platforms. Mintlify, for example, says its llms-full.txt combines an entire documentation site into a single file, with each page's title, source URL, description and full markdown content. The original 2024 proposal gave a similar example under other names, llms-ctx.txt and llms-ctx-full.txt, built by a command-line tool; version 2 dropped that tooling.

The full files can be large. When we fetched them on October 7, 2026, the llms-full.txt behind Anthropic's developer docs was about 9 MB and Perplexity's about 3.6 MB. That is fine for a tool that searches a file, and a lot to load into a model's context at once.

For a SaaS marketing site, an llms-full.txt makes sense when your facts fit in a few thousand words: what you do, plans and prices, limits, policies and answers to common questions. Ours is about 19 KB.

No. Google's guide to optimizing for generative AI features on Google Search, last updated July 10, 2026, says you don't need to create machine-readable files, AI text files, markup or Markdown to appear in Google Search, including its AI features, "as Google Search itself doesn't use them."

The same guide says it is completely fine to keep an llms.txt for other services that use such files, and that doing so will neither harm nor help your visibility or rankings in Google Search, which ignores them. Its summary lists llms.txt among the tactics you can ignore for Google Search.

Google's John Mueller said it more bluntly on Bluesky on June 17, 2025: "FWIW no AI system currently uses llms.txt."

Few fetchers, and no major AI assistant says it reads the file. As of October 7, 2026, the crawler documentation from OpenAI, Anthropic and Perplexity describes their bots and how they treat robots.txt, and none of the three says those bots fetch or use llms.txt.

All three publish llms.txt files for their own developer docs, and the docs pages of OpenAI and Perplexity open with a note pointing agents to that index. That shows they expect coding agents to use such files on documentation. It is not a promise that their assistants read yours.

Server logs point the same way. Ahrefs looked at the 137,210 domains in its Web Analytics product that had traffic in May 2026 and published the results on June 15, 2026. By its count, 28% of them publish an llms.txt (a figure Ahrefs calls an upper bound, since its customers skew technical), and 97% of those files received no requests at all that month. SEO audit tools sent more of the requests that did arrive than any other kind of bot. Named AI tools sent 19.5%, led by OpenAI's GPTBot crawler and Anthropic's Claude-Code agent, while retrieval bots such as OAI-SearchBot and PerplexityBot barely registered. No AI bot asked for llms.txt on sites that did not have one.

Chrome's Lighthouse does check for the file, in an experimental Agentic Browsing category that needs Chrome 150 or later. Its llms.txt audit flags a server error; a missing file is marked not applicable, "as providing the file is optional at the moment."

The proposal's author says coding agents now use llms.txt files reliably, which fits a coding agent being among the top fetchers in Ahrefs' logs. We found no published evidence that the file gets a product cited in ChatGPT, Claude or Perplexity answers.

We serve an llms.txt, an llms-full.txt, a markdown copy of every page and an MCP server, because they are cheap to keep current and they put our prices and rules in plain text. We do not expect them to move our rankings, and we cannot tell you that any AI assistant reads them. As of October 2026:

Write it by hand from your pricing page and docs. A SaaS marketing site rarely needs more than 40 lines.