What llms.txt is, with an example

A web page is built for a browser: menus, scripts, styling and the content somewhere among them. An AI system reading the page has to dig the content out. An llms.txt file gives it a shortcut: a single plain-text file that says what the site is and links to the pages that matter, preferably in a clean text form.

The proposal sets a simple order, written in markdown.

1. A top-level heading with the name of the site or project. This is the only required part.
2. A short summary, written as a quotation block.
3. Any further paragraphs of explanation.
4. Sections under second-level headings, each a list of links with a note beside each link.

A small file looks like this.

# Example Plumbing

> Plumbing and heating repairs in Leeds, with fixed call-out times.

## Services

- [Boiler repair](https://example.com/boiler-repair.md): what is covered and what it costs
- [Emergency call-outs](https://example.com/emergency.md): hours and areas

## About

- [Contact](https://example.com/contact.md): phone, address and opening hours

The proposal recommends that the links in the file lead to markdown versions of pages, published at the page’s address with .md added. A markdown copy carries the page’s words and headings with none of the furniture around them.

This site publishes one. Open hadalmedusa.com/llms.txt to see it: the name, a one-paragraph summary, then every page and guide with its markdown copy beside it. It is generated from the same data as the pages, so it never lists a page that does not exist.

Three limits are worth knowing before adding one.

• The file controls nothing. Access is still set in robots.txt.
• Google says its AI features require no such file.
• Reading it is each tool’s choice. It helps the agents that look for it and changes nothing for the ones that do not.

An llms.txt file costs little and does no harm, and it suits a site whose pages are already clear and well organised. It is not a substitute for that work. A site that is slow, thin or closed to crawlers gains nothing from a map, so the order is: make the site readable first, as how ChatGPT search chooses websites sets out, then add the file.