llms.txt Explained: What It Is, What It Does, and How to Write One

Why it exists

Websites are built for humans with eyes and a mouse. Navigation, hero images, animations, and marketing copy all make sense visually and are noise to a model trying to answer "what does this company do and who is it for?" The llms.txt proposal offers a curated summary: the company in a paragraph, the key pages with one-line descriptions, and links to the documents that matter most, all in clean Markdown.

Adoption is uneven and no engine has published a guarantee that it reads the file. But the cost is an hour, and we have seen cases where the file is the single clearest statement of a company anywhere on the web, which is exactly the thing an answer engine needs when it is deciding whether to name you.

What belongs in it

A good llms.txt is short and factual. A typical structure:

# Company Name

> One paragraph: what you do, for whom, and the one fact that distinguishes you.

## Key facts
- Founded: 2019
- Headquarters: Austin, Texas
- Pricing: from $X per month; full details at /pricing
- Integrations: list the ones buyers ask about

## Main pages
- [Pricing](https://example.com/pricing): current plans and what each includes
- [Product overview](https://example.com/product): features, with limits stated
- [Comparisons](https://example.com/compare): how we differ from the main alternatives
- [Docs](https://docs.example.com): setup and API reference

## Optional
- [Changelog](https://example.com/changelog)
- [Security](https://example.com/security)

The rules that matter: state facts the way you want them quoted, keep it current, and link only pages that are themselves clear and accurate. An llms.txt that points to a vague homepage adds nothing.

What it will not do

It will not make an engine recommend you. It will not override wrong information on third-party sites. It will not substitute for Organization and Product schema, which engines do reliably parse. And it will not help if the pages it links are blocked to AI crawlers or render only with JavaScript. Treat it as one layer of a consistency strategy: your site, your schema, your llms.txt, your Wikidata entry, and your review-site profiles all saying the same true things.

Common mistakes

• Writing marketing copy instead of facts. "The leading platform for modern teams" gives a model nothing. "Payroll and benefits for US companies with 10 to 200 employees, from $40 per month" gives it everything.
• Letting it drift. Pricing changes, the file does not, and now you have published a contradiction.
• Linking everything. Twenty links dilute the signal. Six good ones concentrate it.
• Forgetting the companion file. The proposal also describes /llms-full.txt, a longer version with full page contents, useful for documentation-heavy sites.

How we use it

At AEO GEO Labs an llms.txt is part of the standard "make it impossible to mix you up" pass, alongside Organization and Product schema and a sweep of every third-party profile to align the facts. We write the file from the same source of truth as the schema, so a model reading either gets the same answer, and we revisit it whenever pricing or positioning changes.

Write yours today

Draft the paragraph first: what you do, for whom, and the one distinguishing fact. If you cannot write that sentence cleanly, that is a positioning problem the file will only expose. Once you have it, the rest takes an hour. AEO GEO Labs offers a free llms.txt generator among its tools, and a free report that shows how engines currently describe you, which is a useful "before" picture.