llms.txt
A community proposal (not a W3C or IETF standard, and Google has said Search ignores it) for a plain Markdown file at the site root that tells language models and agents what the site is and which pages matter most.
Analogy Mental model
If the robots file is the doorman's exclusion list and the sitemap is a warehouse manifest of barcodes, this is a one-page guide written for a guest, listing only what matters and what it is.
In practice What you actually do
Create it in half an hour: a heading with the project name, a one-line positioning statement, a short background paragraph, then grouped links to the twenty to fifty pages that matter, each with a sentence explaining what it contains. An optional companion file can hold the full documentation as plain text, though that filename is a community convention rather than part of the original proposal.
Where people get burned Pitfalls
The common mistake is treating it as a switch that earns citations. As of 2026 no major AI vendor has publicly committed to reading it in production, and Google's search team says Search does not use it. Anthropic, Cloudflare and others publish one for their own documentation, which is not the same as consuming yours. Its real value is that it costs almost nothing and forces you to state, in a single sentence, what the site is. What actually decides citation is whether a page can be sliced into self-contained answers.
This entry was written in English from the Chinese original, not machine-translated. Read the Chinese version →