llms.txt explained: what it is, who reads it, how to write one

The idea

HTML pages are written for browsers. A language model that reads one has to wade through navigation, scripts, cookie banners and footers to find the content. llms.txt offers a shortcut: a plain Markdown file that says what the site is and links the pages that matter, ideally to clean Markdown versions of them.

The format

The file lives at /llms.txt and follows a fixed order:

# Example

> Example is a hosted service that checks websites for SEO problems.

Example does not modify the sites it checks and never needs a login.

## Docs

- [Getting started](https://example.com/docs/start.md): install and run a first check
- [API reference](https://example.com/docs/api.md): every endpoint with examples

## Policies

- [Pricing](https://example.com/pricing.md)
- [Privacy](https://example.com/privacy.md)

## Optional

- [Changelog](https://example.com/changelog.md)

1. An H1 with the name of the site or project. The only required part.
2. A blockquote with a short summary.
3. Optional paragraphs with anything a model should know.
4. H2 sections of Markdown links, each optionally followed by a colon and a note.
5. A section named Optional for links a tool can skip when it is short on room.

llms-full.txt

llms-full.txt is the same idea with the content inlined: every linked page’s text in one file, so a tool can read the whole site in one request. It suits documentation sites. Keep it to a size a model can use; a file of many megabytes defeats the purpose.

Who reads it

Honestly: some developer tools, IDE assistants and agent frameworks read llms.txt when you point them at a site, and documentation platforms generate it. The major consumer assistants have not publicly committed to reading it, and search engines do not use it for ranking.

That makes it a small, cheap improvement with a real audience among developers and agents, not a lever for AI visibility. If your site has documentation or an API, publish one. If it is a local business site, your effort is better spent on the basics of AI search readiness.

Write one

1. Use the llms.txt generator to write the file from a few fields.
2. Link the pages that answer the questions people ask about you: what it is, how it works, how much it costs, the policies.
3. Where you can, link Markdown versions of the pages.
4. Publish at /llms.txt, served as text.

Validate it

The llms.txt validator checks that the file exists, is served as text or Markdown, starts with a title and a summary, and that a sample of its links answer. Run it after every docs release, because urls that move break the file quietly.