llms.txt

The proposal was first written in 2024. A file at /llms.txt opens with one H1 and a blockquote summary, followed by H2 sections listing links with short notes. It is ordinary Markdown so that a model can read it without parsing a page's layout.

It answers a different question from robots.txt and from a sitemap. Those deal with access and with listing every page. This file is an editor's shortlist of what is worth reading.

Why it matters for a marketing site

An agent working through a site has limited context, and pages are built for people. A short index can point it at the right material. The value depends on something choosing to read the file, though, so treat it as cheap and possibly useful rather than as a ranking move.

How I handle it

This site publishes one, and it is generated rather than hand-kept. An earlier hand-written version quoted a stale hourly rate and described a process the site no longer followed. Now the page list comes from the content collections and the rates from one facts file. The site check reports llms.txt as a note, not a score, because Google says its AI features do not need it.

Example

Say a migration studio's file opens with one line on what it does, then lists its guides, its integrations and its rates as links with short notes. An agent asked what a move costs can go straight to the rates page. A person who opens the file reads a tidy list.

Common mistakes

• Expecting it to raise rankings. Google says you do not need AI text files to appear in its AI features.
• Writing it by hand and leaving it. A file nobody reads gets out of date first. Generate it from the same data as the pages.
• Using plain URLs. The format is Markdown. Bare URLs read fine to a person and are not links to a parser.