A useful llms.txt file does not try to explain everything about your company. It gives an AI system a fast orientation layer: what the site is, which pages matter most, and what should count as canonical.

Start with a plain-English summary

The opening line should make the purpose of the site obvious. If a human can read the summary and still not know what the business does, the file is not doing its job.

Point to the pages you would cite in a sales call

Good files usually link to the same pages a careful human would reach for first: product, pricing, docs, security, support, and contact. These pages should be stronger than any old launch post or secondary article that mentions the company.

  • Use product pages for what the product does.
  • Use pricing pages for current packaging and plans.
  • Use docs and support pages for operational details.
  • Use trust or security pages for compliance and risk claims.

Keep the scope small enough to be trusted

The fastest way to weaken the file is to overload it. A short list of high-signal URLs is more useful than a giant catalog of everything on the site.

The surrounding site still matters

The file can only summarize what already exists. If the pricing page is vague, or the docs are hard to find, or the product naming changes from page to page, the root file cannot fix that. It can only route the model into the cleanest version of what you publish.

What to do next

Build your draft in the builder, publish it with the platform guide, then compare the result against examples in the directory.