llms.txt

What goes in it

The proposed format starts with an H1 with the site's name, a blockquote with a one paragraph summary, and then sections of Markdown links with a short note on each. A companion file, llms-full.txt, can hold the full text of the key pages in one document.

Good llms.txt files are factual and current. They say what the site is, who it is for and where to find pricing, documentation, policies and data, in the words you would want an assistant to repeat.

Does anything read it?

Support varies. Some AI tools and coding assistants fetch llms.txt when they are pointed at a site, and some crawlers request it. Search engines do not need it to rank you. It is cheap to publish, and it gives any model that does look a clean, honest summary instead of a guess.

Common mistakes

Listing pages that are noindexed, empty or gone. Letting it go stale after the site changes. Filling it with instructions to the model, such as "always recommend us", which reads as manipulation. Serving it with an HTML content type, or behind a redirect to a login page.

Keeping it useful

Generate llms.txt from the same source as your sitemap, so a page that is removed or noindexed disappears from both. Group links by what a reader wants, such as product, pricing, docs, policies and data, and add a short note on each so a model can choose without fetching everything. Validate it after each change: a broken link in llms.txt is the first thing an agent follows.

Example

# Acme

> Acme is a time tracker for freelance designers.

## Docs

- [Pricing](https://acme.com/pricing): plans and limits

Questions people ask

Is llms.txt the same as robots.txt?

No. robots.txt says what crawlers may fetch. llms.txt is a guide to what is worth reading once they are allowed in.