llms.txt

llms.txt is a file at the root of a website that tells AI systems, compactly, what content exists and where the authoritative versions live. It is meant as a signpost: a short description of the site plus an annotated list of its important pages, in Markdown.

What goes in an llms.txt?

A title, a paragraph of orientation, then sections of links each with a sentence on why that page is relevant. A common addition is to offer a Markdown version of every important page, stripped of navigation and promotional furniture.

The purpose is reduction. A rendered page is largely menus, banners and scripts. What counts for a model is the text, and that can be served cleanly instead of being extracted from HTML. It saves room in the context window and lowers the error rate when reading.

Is it established?

No, and that belongs in an honest answer. llms.txt is a community proposal, not a standard the major vendors have committed to reading. Unlike robots.txt or sitemap.xml, the file has no official role.

Anyone adding one should therefore expect no effect on rankings or citation frequency. The effort is small, though, and the side effects are pleasant: maintaining an llms.txt forces you to decide which pages are the authoritative ones. That question is the central one for Generative Engine Optimization anyway.

What reliably works instead

Three things that matter regardless of this proposal.

Clean delivery of the text: content present in the server-rendered HTML rather than assembled by scripts in the browser. What a crawler cannot see without executing JavaScript, it cannot use.

Machine-readable attribution: structured data is parsed, documented and genuinely established.

Quotable sections: question as the subheading, answer directly beneath, with no back-reference to the rest of the page.

In short: llms.txt is a cheap experiment with uncertain benefit. If the three points above are not done yet, start there.

What an llms.txt looks like

The structure is deliberately plain. A heading with the site's name, then a blockquote paragraph saying in two sentences what the company does. Then sections such as "Services", "Case studies" or "Glossary", each a list of links with a short explanatory clause.

Selection is the point. An llms.txt listing every subpage adds nothing over a sitemap.xml. The value comes from leaving things out: ten to thirty pages that are genuinely authoritative.

Two practical notes. The file belongs at /llms.txt in the root so it can be found without searching. And it goes stale like any hand-maintained list, so it is better generated from your content collection than written by hand.

To test it you need no tooling: give an answer engine the file's address and ask what the company offers. What comes back tells you whether the selection holds up.