Give AI agents a map of your site
llms.txt is a real, spec-defined markdown file that tells an AI agent what your site is and which pages matter, build a spec-compliant one below.
- llms.txt is a real markdown file at
/llms.txtthat gives an AI agent a structured summary of your site, proposed by Jeremy Howard in September 2024, now on version two. - It's not a ranking signal and not a permissions file like robots.txt. It's a map: what your site is, and which pages actually matter.
- Fill in your details below, copy the generated file to
/llms.txton your own domain.
What llms.txt actually is, and isn't
The spec is deliberately small. It requires exactly one thing: an H1 with your site or project's name. Everything else is optional but expected, a one-line blockquote summary underneath the H1, then any number of plain paragraphs, then any number of H2-delimited sections listing links in the form [name](url): note. That's the whole format.
It's easy to lump this in with robots.txt since both are plain-text files at a predictable path, but they do opposite jobs. robots.txt says what a crawler is allowed to access, pure permissions, no description of what any of it means. llms.txt assumes an agent is already reading your site and hands it a summary and a short list of the pages that actually matter, the same way a table of contents helps a person, not a bouncer at the door.
Worth being honest about where adoption actually stands: there's no confirmed evidence that a specific major AI assistant fetches and prioritizes llms.txt today. What's real is that it's a voluntary convention with genuine, growing adoption, thousands of sites publish one, documentation platforms like Mintlify generate it automatically, and Chrome's own Lighthouse now audits for it as part of its agentic-browsing checks. It costs little to have and does nothing to hurt you, which is a reasonable bar even before broader model adoption is confirmed.
Build your llms.txt
Your llms.txt
Doing it right
- Keep the summary literal. One sentence that says what the site actually is, this is for a machine deciding whether to keep reading, not a marketing tagline.
- List only the pages that matter. A handful of genuinely important links beats a dump of your entire sitemap, that's what your real sitemap.xml is already for.
- Use the "Optional" convention for secondary links. By convention, a section named "Optional" signals links an agent can skip when it needs a shorter context, useful for anything real but non-essential.
- Keep it current. A stale map pointing at retired pages is worse than no map at all.
Where it goes
Save the output as plain text at yourdomain.com/llms.txt, no server configuration beyond hosting a static file, the same mechanism as robots.txt. A file scoped to one section of a large site can also live at a subpath, like /docs/llms.txt, covering everything beneath it.
FAQ
Who created llms.txt, and is it an official standard?
Jeremy Howard (of Answer.ai and fast.ai) proposed it in September 2024; the spec is now on its second version. It isn't an official standard ratified by a body like W3C, but it has real adoption, thousands of sites publish one, documentation tools like Mintlify generate it automatically, and Chrome's own Lighthouse audits for it as part of its agentic-browsing checks.
Does having an llms.txt file guarantee ChatGPT or Perplexity will read it?
No. There's no confirmed evidence that any specific major AI assistant currently fetches and prioritizes llms.txt when answering questions. It's a low-cost, forward-looking convention rather than a guaranteed input to any one model today.
How is llms.txt different from robots.txt?
robots.txt is permissions. It tells crawlers what they may access. llms.txt is a map. It describes what your site is and points an agent that's already reading it toward the pages that matter most. They do different jobs and having one says nothing about the other.
Where does the file need to live?
At your domain's root as /llms.txt (or a subpath like /docs/llms.txt, covering everything beneath it), served as plain text, the same mechanism as robots.txt. No server configuration beyond hosting a static file.