llms.txt is a plain-text file at the root of a website, written in Markdown, that tells AI assistants what the site is and where its important pages are. It is a proposal, published at llmstxt.org, not a standard that every assistant has agreed to follow. It is cheap to produce and does no harm, it is read by some AI tools and ignored by others, and Google has said its search ignores it. Your site should have one if you want to make it as easy as possible for assistants to understand you, as long as you treat it as one part of the job and not the whole of it.
What the file looks like
The proposal at llmstxt.org sets a simple shape. The file begins with a single H1 heading, the name of the site or project, which is the only required part. A blockquote underneath gives a short summary. Then come sections under H2 headings, each a list of links, where every link can carry a short note after a colon. The proposal also suggests a companion file, llms-full.txt, that holds the full text of the pages rather than just the links, and the current version of the spec describes Markdown twins of pages at the same address with .md added.
A short example, in the shape the proposal describes:
- The H1: the name of the business.
- The blockquote: one paragraph saying what the business does, for whom and where.
- Sections: services, each with a link and a one-line note; the main pages; the articles.
That is all. It is deliberately something a person can write in an editor.
What it does
It gives an assistant a map. When a tool fetches a site to answer a question, a well-written llms.txt tells it in a few hundred words what the site is about and which pages matter, instead of leaving it to work that out from the navigation and the home page. For a site with many pages, the full-text version lets a tool read everything in one request.
What it does not do
It does not make you rank. Google's guidance for its AI features says explicitly that creating AI text files such as llms.txt neither helps nor harms a site in Google Search, because Google Search ignores them. It does not replace structured data, clear pages or crawler access. And it does not force any assistant to read it. The proposal's own page says that thousands of sites publish one, and the major AI companies publish their own, which tells you the idea is taken seriously, not that every answer engine consults it.
Who reads it
Developer tools and documentation assistants were the first users, which is why so many software companies publish one. Beyond that, treat it as useful for any agent that fetches your site directly, and as a low-cost signal of clarity. We publish one for orqen.si and for the sites we build.
A real example
getoperators, an executive search practice whose site Orqen built, publishes an llms.txt that opens with the practice's name, a one-paragraph description, and a note on the other names the business goes by. It then lists every public page by section and points to a full-text version. Operator profiles are not listed one by one; the file points to the bench as the index of them. That last choice matters: the file is a map, not a copy of the site.
How to write a good one
- Name and summary first. One H1, one paragraph that says what you do, for whom and where, in the same words you use on your home page.
- Sections that match how buyers think. Services, then the pages that answer questions, then the rest.
- A note on every link. A link with no note is a mystery. "Fractional CFO: day rates and how engagement works" is useful.
- Keep it current. Generate it from the site if you can, so new pages appear automatically. We build ours that way.
- Offer the full text. A llms-full.txt with the page text gives tools that want everything a single place to get it.
- Link it. Put a "For AI assistants" link in your footer and a link element in the page head, so it is discoverable rather than hidden.
Does your site need one?
If you want to be found and understood by AI assistants, yes, because it costs almost nothing and removes guesswork. Do it after the things that matter more: allowing the search crawlers, writing pages that say what you do, and adding structured data. Our guide to getting recommended by ChatGPT puts those in order, and schema for AI search covers the structured data. Orqen builds all of this into the websites we make and run.
Frequently asked questions
Will llms.txt improve my Google ranking?
No. Google states that its search ignores these files, and that they neither help nor harm. Its value is for assistants that fetch your site directly.
Can I just paste my whole site into it?
The proposal separates the index, llms.txt, from the full text, llms-full.txt. Keep the index short and put the full text in the second file.
Do I need to block anything in it?
No. It is not an access control file. Crawler permissions belong in robots.txt.
How do I check mine is right?
Open yoursite.com/llms.txt in a browser. It should start with your name as a heading, have a summary, and list real pages with notes. If it does not load, it is not there.
Talk to us about your site
If you would like an llms.txt written and generated from your site, as part of making it readable to AI assistants, book a call with us and we will take a look at what you have.
