llms.txt is a proposal for a file at the root of a site, /llms.txt, written in Markdown. The idea is simple: give a language model a title, a one-line summary, and links to the pages that explain the rest, so it does not have to guess which URL matters. It is not a ranking system, and it is not a replacement for HTML that already contains your content.
The format is small on purpose
A valid file starts with an H1, which is the name of the site or the business. After that, an optional blockquote can summarise the site in a sentence or two. Then H2 sections hold links, written as ordinary Markdown links. The H1 is the only required piece. Links are recommended, because a file with a title and no destinations does not help anyone navigate.
- Serve it as text, not as an HTML page. An HTML body on /llms.txt is usually a soft 404, and the check treats it that way.
- Keep the summary short. The file is a map, not a second copy of the site.
- Link to pages that already exist and that you are happy for an agent to read: the homepage, the methodology, a contact page, a few guides.
- Do not list pages you have marked noindex. You would be advertising a URL you asked search engines to skip.
SiteScore fetches /llms.txt as part of the AI-search sub-grade. A valid file scores the full 15 points set aside for this check. A file that is present but invalid scores half. A missing file scores zero for that check. The other 85 points cover crawler access, identity schema, content in the raw HTML, and semantic structure, which the AI-search guide goes through.
Google Search does not use it
Google has said that Google Search, including its generative AI features, does not need AI text files, special markup, or Markdown to include a page. Publishing an llms.txt file will neither help nor hurt your Google rankings. Treat any claim that the file is a Google ranking factor as wrong.
llms.txt is not part of the overall letter grade. The 15 points live entirely inside the separate AI-search score. Adding the file cannot move you from a C to a B.
That is also why a missing file is not an emergency. If Google cannot see the site at all, the cause is somewhere in robots.txt, noindex, the canonical, or the sitemap. Start with why a site goes missing. Write llms.txt after the pages themselves are fetchable.
Who might read it
Some agents and developer tools request /llms.txt directly when they want a map of a site, especially for documentation. It is a courtesy for those clients. It does not force ChatGPT, Perplexity, or Claude to cite you. Those products still have to be allowed to fetch the page, and the answer still has to be in the HTML. A perfect llms.txt on a site that blocks OAI-SearchBot does nothing for ChatGPT search.
Training crawlers are a separate decision, and they are not scored. Blocking GPTBot does not require you to skip llms.txt, and publishing llms.txt does not invite those bots in. robots.txt remains the access control. The file is only a reading list for whoever is already allowed to fetch.
When it is still worth writing
Write it if you want a single, maintained list of the pages a new agent should open first. Keep it in the same change as the sitemap: when a guide moves or a product is renamed, update both. A stale list is worse than a short one. SiteScore publishes its own /llms.txt for that reason, and the scan checks yours the same way.
If you want to see the 15 points in context, run the free scan and read the AI-search grade next to the letter. What the letter itself means is the other half of that screen. The points, including the half-credit for an invalid file, are published on the methodology page, generated from the same code that scores the scan.
