Guide · GEO
What is llms.txt, and how do you set it up?
llms.txt is a plain text file in Markdown, placed at the root of a website, that tells AI language models in a few lines what the site is about and which pages matter most. It is a proposed convention, not a standard: AI providers are not obliged to read it, and it does not control what their crawlers may access. It takes little effort to add and is worth doing as one part of wider work on AI visibility, not as a substitute for that work.
What is llms.txt for?
A website is built for people in a browser: menus, banners, scripts, cookie notices. A language model that reads a page has to find the content in between, and it can take in only a limited amount of text at a time. The llms.txt proposal offers a shortcut: one short file, at a fixed address, that says who you are and links to the pages worth reading, each with a line of explanation.
What goes in the file?
The file is written in Markdown and is published at /llms.txt, next to robots.txt. The proposal sets out a fixed order:
- A title: the name of the business or the site. This is the only required part.
- A short summary, written as a quotation block, that lets a reader place you: what you do, for whom and where.
- Optionally, a few plain paragraphs of further context.
- Sections with a heading each, containing a list of links. Every link has a name, the address and a short note on what the page covers.
- A section called Optional, for links that may be skipped when space is short.
Keep it short and factual. It is a map, not a brochure: a short list of well-described links is worth more than a copy of the sitemap.
How do you set it up?
- List the pages that best explain the business: the services, the main guides, the about and contact pages.
- Write one sentence for each that says what the page answers.
- Save the file as plain text and publish it at the root of the domain, so that it opens at your address followed by /llms.txt.
- The proposal does not say how to handle several languages. A workable approach is one file per language version of the site.
- Where possible, have the content system generate the file, so that it does not go stale when pages change.
- Open it in a browser and follow every link.
Some sites also publish a Markdown version of each page, or a second file, llms-full.txt, holding the full text. Both are extensions; start with the basic file.
What does llms.txt not do?
- It is not robots.txt. It grants and refuses nothing. Whether a crawler may fetch your pages is still decided in robots.txt.
- It is not a ranking signal. No search engine has said that it uses the file to rank pages, and Google has said that it does not.
- It is no guarantee. At the time of writing, none of the large AI providers has committed to reading it when answering. Some tools and agents do fetch it, mostly for technical documentation.
Is it worth doing?
Yes, with the right expectations. The effort is small, and writing the file forces a useful exercise: stating in a few lines what you do and which pages show it. But a file cannot repair pages that crawlers cannot read or that answer no question. We add it as one step in our GEO work, after access and content. What that wider work involves is set out in What is GEO?
Frequently asked questions
Is llms.txt an official standard?
No. It is a community proposal, not a standard adopted by a body such as the W3C or the IETF. Its format may still change, and AI providers are free to ignore it.
Does llms.txt stop AI crawlers from using my content?
No. The file only points to pages; it cannot refuse access. To refuse a crawler you name it in robots.txt, and even that works only for crawlers that choose to respect it.
Where exactly does the file go?
At the root of the domain, so that it opens at your address followed by /llms.txt. A tool that looks for the file looks there, and will not find it in a subfolder or under another name.
How often should llms.txt be updated?
Whenever a page it lists is added, removed or moved. A file with dead links or a description of last year's services does more harm than no file, which is why generating it from the content system is the safer route.
Let's talk about your project.
Tell us where you are and where you want to be. We'll come back within one working day with a first view. A person reads your message and answers it.