Ship an llms.txt that AI engines will trust.

What the validator checks

We request https://yourdomain.com/llms.txt, follow redirects, and confirm the response is served as readable text with a 200 status. A file that returns HTML, a 404 page, or a login wall is treated as missing, because that is exactly how an AI crawler would see it.

The parser then walks the markdown structure: a single H1 carrying your site or product name, an optional blockquote summary, and H2 sections grouping related links. Each link must be an absolute HTTPS URL with descriptive anchor text.

• HTTP status, content type, encoding and redirect chain
• Heading hierarchy and section grouping
• Absolute HTTPS links with human-readable descriptions
• Reachability of up to twelve sampled linked pages
• Duplicate, empty or orphaned sections

How to read your score

The compliance score is the share of structural checks that passed, weighted so that fatal problems — a missing file, unreachable links, or a broken heading tree — cost more than cosmetic ones. A score above 90 means an engine can parse your file without guessing.

Failing rows are more useful than the number itself. Each check lists exactly what we saw, so you can reproduce the issue with a single request rather than trusting a summary.

Common mistakes we see

The most frequent failure is a file that exists in a staging branch but was never deployed, so the production domain still serves a 404. The second is relative links, which resolve correctly in your editor but not for a crawler reading the raw file.

Bloat is the third. An llms.txt with several hundred links behaves like a sitemap and dilutes the signal; a curated list of the twenty to fifty pages that actually answer questions performs better in our own testing.

• Deploy to the production root, not a subdirectory
• Use absolute https:// URLs everywhere
• Describe each link in plain language
• Keep it curated, not exhaustive
• Re-validate after every site restructure

Where it fits in AI search

An llms.txt file does not guarantee citations. It removes ambiguity: when a model has already decided to reference your domain, a clean file increases the chance it reaches the page you intended and quotes accurate, current wording.

Pair it with server-rendered content, an accessible robots.txt and a valid sitemap. Those three signals together determine whether an answer engine can crawl, parse and trust your site.