AI-Readable Files

Your domain llms.txt file

llms.txt is a single markdown file at your site root that tells AI systems what your site is about and which pages matter most: a title, a one-line summary, and short sections of described links. SuperSchema does not generate it for you today. This article covers the format, the rules that make one worth reading, and how to write your own.

View as plain text Updated 2026-08-14

Where alternate files give AI the clean content of a single page, llms.txt describes your whole site. It is one markdown file at your site root that gives AI systems a curated map: what your site is, and which pages are worth reading. Think of it as a short guided tour written for a machine, not a list of every URL you have.

Note SuperSchema does not generate llms.txt for you today. This article explains how to write your own. Generation is on the roadmap, and this page will be updated when it ships.

Who reads it, honestly #

llms.txt is a convention, not a standard that every AI company has agreed to. Anthropic has confirmed support and Perplexity has said it retrieves llms.txt to help decide which pages to look at. Coding agents such as Cursor and the Claude desktop app read it. Independent server-log studies suggest most published llms.txt files are rarely requested, so it is worth being clear-eyed: this is a low-cost file that helps in specific situations, not a switch that changes how often AI mentions you.

It is most useful when an AI system arrives at your domain without prior context and has to work out what you do and where to look. It costs a few minutes to write and nothing to host, which is why it is worth doing even while adoption is still settling.

How it differs from a sitemap #

A sitemap.xml is a complete list of URLs built for crawlers to discover pages. It says where your pages are, not what they contain, and it aims for coverage. llms.txt is the opposite: it is short, curated, and written to be read. It leaves most of your site out on purpose.

Sitemap.xmlllms.txt
AudienceSearch-engine crawlersAI systems and agents
FormatXML list of URLsMarkdown, readable by a person
GoalComplete coverageCurated selection
Typical sizeEvery page you have5 to 20 pages
ContainsURLs and last-modified datesSite summary plus described links

The format #

The file is plain markdown and the structure is fixed. A single H1 with your site name, a blockquote summary, then H2 sections containing lists of links. Each link is a markdown link, optionally followed by a colon and a description.

markdown
# Acme Robotics

> Industrial pick-and-place arms for warehouse automation.
> Last reviewed 2026-08-14.

## Products

- [AR-7 Arm](https://acmerobotics.com/products/ar-7): 7-axis arm, 12kg payload, 1.4m reach. Spec sheet and CAD downloads.
- [Gripper Kits](https://acmerobotics.com/products/grippers): Vacuum and two-finger end effectors in four sizes.

## Documentation

- [Install Guide](https://acmerobotics.com/docs/install): Mounting, power and network setup for AR-series arms.
- [ROS 2 Driver](https://acmerobotics.com/docs/ros2): Package install, topics and services.

## Optional

- [Company History](https://acmerobotics.com/about/history)
  • The H1 is the only required element. Use exactly one, and put it first.
  • The blockquote directly under the title is your summary. This is the line an AI is most likely to reuse when describing your company, so make it accurate rather than promotional.
  • Every list item must be a real markdown link. A bare URL on its own line is the single most common mistake and it makes the file unreadable to a parser.
  • A section named exactly "Optional" is reserved by the convention. It means these links can be skipped when a shorter answer is needed.

What makes one worth reading #

The descriptions are the part that does the work. An AI deciding which of your pages to read decides on the description, not the URL. "Product page" tells it nothing. "7-axis arm, 12kg payload, spec sheet and CAD downloads" tells it exactly when this page is the right answer.

  • Include 5 to 20 links. Past about 50 the file stops being a curated map and each entry gets diluted.
  • Keep the whole file under about 10kB so it stays cheap to read.
  • Write specifically. If a description could be pasted onto any other company's site, it is not earning its place.
  • Skip marketing language. Taglines and metaphors rely on context an AI does not have.
  • Date the file. A line such as "Last reviewed 2026-08-14" tells a reader whether to trust it.
  • Group links under headings that mean something: Products, Documentation, Pricing, Support.

Common mistakes #

  • Bare URLs instead of markdown links. The file will not parse as llms.txt.
  • More than one H1, or an H1 that is not the first line.
  • Dumping your whole sitemap into it. Volume is the opposite of what this file is for.
  • Links that have moved or 404. Stale links make a reader distrust everything else in the file.
  • Outdated product names or prices. An llms.txt with wrong information is worse than none, because the wrong information travels.
Warning Never list staging or preview URLs, and never list a URL with a token, key or signature in its query string. Your llms.txt is public and anyone can read it. If you find one, remove the link and rotate the credential.

Where to put it #

The file belongs at the root of your domain, served as plain text or markdown at https://yourdomain.com/llms.txt. On most platforms that means dropping it in the folder that serves static files: public/ for Next.js, Nuxt and Vite, static/ for Hugo and SvelteKit, the site root for WordPress, and the Files area for Webflow or Squarespace. Confirm it worked by visiting the URL and checking you get the raw text back rather than a styled page or a 404.

Tip Pair llms.txt with per-page alternate files. llms.txt gives AI the map of your site, and each alternate file gives AI the clean content of one page without the navigation, banners and scripts around it.

Still have questions?

Can't find what you're looking for? Our support team is here to help.

Contact Support