What is llms.txt, and should your site have one?

6 min read · Updated

In short

What the llms.txt proposal says, how to write a valid file, and a realistic view of its status as an optional convenience.

What llms.txt is, and its status

llms.txt is a proposal by Jeremy Howard, published on 3 September 2024. The idea is a Markdown file at /llms.txt (or in a subpath) that gives a language model a short, readable guide to a site: what it is and where the useful pages are.

More

The reasoning is practical. Web pages carry navigation, scripts and layout that get in the way when a model has limited context. A small curated file with links to clean pages is easier to use at the moment someone asks a question. The proposal describes it as used on demand at inference time.

The proposal page presents llms.txt as open for community input. It does not state that search engines have adopted it. This guide does not claim that a file improves rankings or leads to citations, and no major search engine has been shown to rely on it.

The file format

The format is plain Markdown with a fixed order of parts. Only the first part is required.

More
  1. An H1 with the name of the project or site. This is the only required part.
  2. An optional blockquote with a short summary.
  3. Optional paragraphs or lists with more detail, but no headings.
  4. Optional H2 sections, each holding a list of links written as [name](url): notes.
  5. An H2 named 'Optional' for secondary links that can be skipped when context is short.
Example llms.txt
# Example Co

> Example Co makes scheduling software for small clinics.

Plain-language notes about the product go here.

## Docs

- [Getting started](https://example.com/docs/start.md): Set up in five steps
- [API reference](https://example.com/docs/api.md): Endpoints and errors

## Optional

- [Changelog](https://example.com/changelog.md): Release notes

Writing one that is worth having

  • List only pages you would want quoted: documentation, pricing facts, key guides.
  • Write each note as one factual line. Leave out marketing claims.
  • Link to pages that are readable without scripts. Markdown versions are suggested by the proposal where you can offer them.
  • Keep it short. A file that lists every URL repeats the sitemap and adds nothing.
  • Update it when pages are added, moved or retired. A stale file points readers at dead links.
More

Check the result by opening the URL in a browser and reading it as a stranger would. If the first screen does not tell you what the site is for, rewrite the summary. Also confirm that the file is served as plain text or Markdown, is not behind a login, and is not blocked by a redirect loop, since a file nobody can fetch helps nobody.

How it differs from sitemaps and robots.txt

FileAudiencePurpose
robots.txtCrawlersStates which paths may be fetched
sitemap.xmlSearch enginesLists URLs for discovery and freshness
llms.txtLanguage models and the people who build tools with themA short, readable guide to the most useful pages
More

The three do different jobs. llms.txt does not grant or block access, and it does not replace a sitemap. If you want to control which automated agents may fetch your pages, that is a robots.txt matter.

A short maintenance routine, and whether to add one

  1. When you publish or retire a key page, update the list.
  2. Once a quarter, open each link and confirm that it still loads and says what the note claims.
  3. Re-read the summary against your homepage. If your positioning changed, change the summary.
  4. Remove pages that are out of date instead of keeping them in the Optional section.
More

Because there is no evidence of a measurable benefit, keep the effort proportionate. Fifteen minutes a quarter is plenty for a small site.

The decision is low stakes. If you maintain documentation or reference pages that you would like a model-based tool to find easily, a short file is a reasonable convenience. If your site has little structured content, the file will have little to point at.

  • Add one if you have a clear set of documentation or guides and someone to keep the file current.
  • Skip it if you cannot maintain it. A stale file is worse than none.
  • Do not expect to see an effect, and do not report one without evidence. If you want to test, sample answers before and after, as described in the guide on measuring AI search visibility, and treat the result as anecdotal.

A related point about FAQ markup

People often group llms.txt with FAQPage structured data. Google's changelog says the FAQ rich result stopped appearing in Google Search from 7 May 2026, and its documentation page was removed on 15 June 2026. FAQPage markup should therefore not be added as a rich-result tactic. It remains valid schema.org vocabulary that software can read, so adding it is harmless where real questions and answers already exist on the page.

Where BlogFactory fits

BlogFactory generates an llms.txt from the site's indexed pages and published posts, and the panel has a Copy button. It also produces BlogPosting and FAQPage JSON-LD for a post. Ghost drafts receive the FAQPage JSON-LD automatically when the article has at least two real question-and-answer pairs. Other CMS targets do not get this yet.

More
  • Needs: indexed pages or published posts to build the list from.
  • Cannot do: place the file on your domain. You copy it and publish it at /llms.txt yourself.
  • Cannot do: show that any AI product reads the file or that it changes citations or rankings.

Try it in BlogFactory

Questions

Does llms.txt improve my Google rankings?

There is no evidence that it does, and no major search engine has been shown to rely on it. Treat it as an optional convenience.

What is the minimum valid llms.txt?

A single H1 with the site or project name. Everything else in the proposed format is optional.

Where does the file go?

At /llms.txt on your domain, or in a subpath, as the proposal describes.

Is it the same as robots.txt?

No. robots.txt tells crawlers what they may fetch. llms.txt is a readable guide to useful pages and does not grant or block access.

Should I still use FAQPage markup?

Not for rich results, which Google stopped showing from 7 May 2026. It remains valid vocabulary for machines, so use it only on pages with genuine questions and answers.

Sources

Plan it, draft it, approve it.

BlogFactory does the research and drafting steps in this guide. You approve every draft, and nothing publishes itself.