Knowledge base

A practical llms.txt guide

What to put in llms.txt, what to leave out, and how to keep bilingual versions consistent.

llms.txt is a small text file placed at the site root that tells language models which pages matter, what each page answers, and where the canonical version lives. It is not a ranking factor and it does not replace robots.txt. It is a curation file: it removes ambiguity when a model has to choose between ten similar pages.

Keep it short. A useful file lists the homepage, the main service or product pages, the pricing page, the policy pages, and a handful of answer pages. Each entry needs a plain-language title and a one-line description. If a page is not worth citing, leave it out; a list of two hundred URLs is treated as noise.

Do not put secrets in it. llms.txt is public. Internal documentation, unpublished drafts, customer data and staging URLs do not belong there, and listing them creates a leak that robots.txt cannot undo. Treat the file exactly like a public sitemap with commentary.

Bilingual sites need two consistent versions. The English file should describe English pages, the Chinese file should describe Chinese pages, and both should point to the same canonical structure. If the two files disagree about which page is canonical, models will pick one at random and may cite the thinner version.

Finally, treat llms.txt as documentation that must stay current. When a product page moves or a policy page is replaced, update the file in the same change. An outdated curation file is worse than none: it sends models to dead or redirected URLs and teaches them to distrust the rest of the site.