AI search
llms.txt that a model can actually walk
· Configuration notes, not a newsroom.
How to write llms.txt as a URL map, not a slogan. What SiteRune scores, what we refuse to invent, and why a 200 OK file can still fail the chapter.
A file is not a map
Most llms.txt files we fetch are a paragraph of marketing and a link back to the homepage. That is a brochure. Answer engines need a list of URLs they are allowed to treat as source material: docs, pricing, about, policy, the pages that contain facts.
SiteRune fetches llms.txt on the origin, then walks those URLs before ordinary nav links — up to the plan cap (Observer 5, Operator 10, Studio 20). If the file 404s, we say so. If it exists but names the homepage twice, we do not pretend it is a knowledge graph.
What to put in it
A useful llms.txt is closer to a curated sitemap than a brand story. Minimum viable:
- H1 with the legal or trading name of the organization — the same string as your JSON-LD name.
- A short paragraph a model can quote without hallucinating staff or offices.
- Markdown links to 5–15 durable URLs: product, pricing, docs, security, about, contact.
- No tracking parameters. No login walls. No 'click here'.
What SiteRune will not do
If your origin already ships a stronger llms.txt than the template we generate, we do not replace it. We do not invent authors, editors, or a newsroom. We do not tell you to buy Reddit threads. The AEO chapter is a configuration review of what a polite crawler sees.
Run it on a live URL
The brief will name your CMS and attach the files. Three guest scans, no card.