Guide

How to write an llms.txt, and why you should generate it instead.

A template you can copy, an explanation of what belongs in each part, and the five mistakes that turn the file into something worse than not having one.

llms.txt is a plain-text file at the root of your site giving AI systems a clean map of what you have and where. It is a convention rather than a rule, adoption is partial, and it is not a ranking factor. Add it because it costs almost nothing and the exercise is clarifying, not because you were promised a result. Fuller explanation of what it is.

The template

Save this as llms.txt at the root of your site, so it is reachable at yourbrand.com/llms.txt. It is Markdown, and it should stay short enough to read in a minute.

# Your Business Name

> One paragraph saying what the business does, who it is for, and where
> it operates. Write it for someone who has never heard of you. This is
> the single most-read line in the file.

Any facts you want stated accurately: the legal entity, the country,
where the product or service actually lives.

## Pages

- [Page title](https://yourbrand.com/page): One sentence on what this
  page covers and who it is for.
- [Pricing](https://yourbrand.com/pricing): What each plan costs, plainly.

## Writing

- [Article title](https://yourbrand.com/blog/slug): What the article
  argues, not just its subject.

## Notes

- Anything commonly got wrong about you, corrected here.
- Limits worth knowing, stated plainly.

What each part is doing

The heading and summary carry most of the value. A system that reads nothing else should come away knowing what you sell, who buys it and where you operate. Write it as prose for a person, not as a keyword list.

The page list is a map with commentary. The link tells a system where to go; the sentence tells it whether it wants to. That sentence is the part people skip and the part that makes the file useful.

The notes are underused and worth thinking about. This is where you state, plainly, the things that get reported about you incorrectly. If assistants keep describing a limit you do not have, or a service you stopped offering, say so here in one line.

Five mistakes

  1. Writing it once and forgetting it

    Within two months it describes a site that no longer exists, and a confidently wrong file is worse than no file. Generate it from the same page list your sitemap uses.

  2. Listing every page

    It is a map, not an index. Twenty well-described pages are more useful than four hundred URLs with no context. If a page does not deserve a sentence, leave it out.

  3. Writing summaries that summarise nothing

    "Our services page" tells a reader nothing they could not guess from the URL. "What we install, in which counties, and the typical price range" is a sentence worth having.

  4. Treating it as an access control

    It grants and denies nothing. Crawler permissions live in robots.txt, and that is where a mistake actually costs you visibility.

  5. Expecting it to move rankings

    It is not a ranking factor and nobody should tell you it is. Add it because it is cheap and clarifying, not because you were promised an effect.

Generate it, do not maintain it

This is the one piece of advice worth taking away whatever you think of the standard. A hand-written llms.txt is stale within a couple of months, because nobody remembers it exists when they add a page.

Build it from the same list of routes that produces your sitemap. That is what we do here: our route list lives in one file, and the sitemap and our own llms.txt are both generated from it, so adding a page to the site adds it to both automatically and neither can quietly disagree with the other.

Then check the thing that actually matters

An llms.txt is housekeeping. Crawler access is not. A perfect file pointing at pages that AI crawlers are forbidden from fetching achieves precisely nothing, and blocking the wrong crawler is the single most common cause of a business being absent from AI answers.