llms.txt: a step-by-step method and prompt

llms.txt is a markdown file at the root of your site that gives AI assistants a short, curated guide to your most useful pages. It starts with your name and a one-line summary, then lists key links with a sentence each. It is cheap to make and helps models find clean, authoritative pages instead of guessing.

What it is

The format is described in the llms.txt proposal. The file has an H1 with the site or product name, a blockquote summary, optional notes, then H2 sections containing markdown link lists. Each link can carry a short description after a colon. An optional section titled "Optional" marks links a model can skip if context is tight.

A minimal file for a hypothetical product looks like this:

# Ledgerly

> Invoicing software for EU freelancers that applies VAT rules automatically.

## Docs
- [Getting started](https://ledgerly.example/docs/start): Create an account and send a first invoice.
- [VAT rules](https://ledgerly.example/docs/vat): Which VAT rate applies to which client.

## Optional
- [Changelog](https://ledgerly.example/changelog): Release notes by month.

Some sites also publish llms-full.txt, which contains the full text of key pages in one file. Start with the short version.

For background on why this matters for AI search, read llms.txt explained.

When to use it

• You want AI assistants to describe your product accurately.
• You have docs, pricing or comparison pages that models should read first.
• Your site is heavy on JavaScript or navigation, which makes clean extraction harder.
• You are running a GEO audit and want quick groundwork.

Step-by-step method

1. Choose 10 to 30 pages. Pick pages that answer real questions about you: home, product overview, pricing, top docs, key use cases, comparisons, policies.
2. Write the summary. One blockquote sentence that says what you do and for whom, with no slogans. Example: "Ledgerly is invoicing software for freelancers in the EU that handles VAT rules automatically." (Ledgerly is hypothetical.)
3. Group links into sections. Typical H2s: Product, Docs, Pricing, Guides, Company. Keep each section short.
4. Describe every link. After each link, write one factual sentence about what the page contains. Models use these to decide what to fetch.
5. Prefer clean URLs. Link canonical pages that render content in HTML. If you offer markdown versions of docs, link those.
6. Publish at the root. Serve it at /llms.txt with a text content type. Make sure robots.txt does not block it.
7. Maintain it. Update when you launch pages, change pricing or retire products. Add it to your release checklist.

Ready-to-use prompt

Write an llms.txt file for [COMPANY] ([DOMAIN]) following the llmstxt.org format.

What we do, in plain words: [ONE OR TWO SENTENCES]
Audience: [WHO]
Pages to include (url and what the page contains): [PASTE LIST]

Rules:
- Start with "# [COMPANY]" then a one-sentence blockquote summary with no marketing adjectives.
- Group links under H2 sections: Product, Docs, Pricing, Guides, Company (drop empty ones).
- Format each line as "- [Page title](url): one factual sentence about the page."
- Put nice-to-have pages under "## Optional".
- Use only the URLs I gave you. Output the file contents only.

Checklist

• 10 to 30 high-value pages chosen
• H1 with the product or company name
• Plain, factual one-sentence summary
• Links grouped under clear H2 sections
• One factual description per link
• Only canonical, indexable URLs
• File served at /llms.txt and not blocked by robots.txt
• Update added to the release checklist

How Thig automates it

The GEO agent drafts an llms.txt from your crawled pages and strategy documents, and the Coding agent can ship it to your repo as a pull request. To make one by hand in a minute, use the free llms.txt generator.