LLMS.txt Validator

Structural rules to check

Exactly one H1 carrying the site name. Section headings at H2. Link lists in standard markdown with a description following a colon. Any blockquote summary immediately after the H1. Multiple H1s or headings jumping straight to H3 break the expected shape, and a parser reading the file positionally will misattribute entries.

Broken links are the common fault

Entries pointing at 404s, at redirects, or at URLs that changed during a site migration. A file that recommends missing pages is worse than none, because it spends a model's limited budget on failed fetches. Every URL should return 200 directly rather than through a redirect chain.

Check against robots.txt and canonicals

A page listed here but disallowed in robots.txt will not be fetched. A page whose canonical points elsewhere should not be listed — the canonical target should be. These contradictions are invisible when the file is written by hand from a list of favourites.

Descriptions do the selection work

Entries with no description force a model to fetch a page to discover whether it is relevant. A specific one-line description is what allows the file to serve its purpose without any fetching, and it is the difference between a useful file and a decorated sitemap.

Absolute URLs, served as plain text

Relative paths resolve inconsistently for a fetcher that did not arrive through the site. The file belongs at the domain root, served as text/markdown or text/plain — serving it as HTML with a template wrapper defeats parsing entirely.