llms.txt: What It Is, How to Write One (With Example), Who Uses It

It was proposed by Jeremy Howard of Answer.AI in September 2024, and the spec lives at llmstxt.org¹.

Before you write one, here's the key point: as of November 2026, Google says its Search ignores llms.txt, and we haven't found official documentation from OpenAI, Perplexity, or Anthropic saying their search engines use it to decide what to cite. Where it is used is technical documentation read by coding agents. Having one doesn't hurt and costs little, but it isn't a GEO lever. Details below.

What llms.txt is for

Web pages are built for people: menus, ads, JavaScript, cookie banners. A model trying to read your site has to strip all that out, and its context window can't fit the whole site anyway. The llms.txt proposal tackles both problems:

• One small file that explains what the site is and where the important stuff lives.
• Links to clean Markdown versions of key pages.

The spec (version 2, updated in 2026) also proposes that each page offer a Markdown version at the same URL ending in .md (for example, /pricing.html.md or /pricing.md), linked with rel="alternate" type="text/markdown" and rel="describedby" in the HTML or in an HTTP Link header.

The format, step by step

Per the spec, the file contains these sections, in this order¹:

Each list item is a Markdown link, [name](url), optionally followed by a colon and a note. The Optional section has a specific meaning: an agent can skip it when it needs a shorter context.

The file can sit at the root or in a subfolder (/docs/llms.txt), in which case it covers only the pages under that path.

A valid llms.txt example

Here's an example for a fictional marketing agency. Feel free to copy and adapt it:

# Example Agency

> Digital marketing agency in Austin, Texas, specializing in SEO and GEO for dental and medical clinics. Serving clients across the US since 2015.

Prices are in US dollars. Service pages list scope, timelines, and case studies published with client permission.

## Services

- [GEO audit](https://exampleagency.com/services/geo-audit.md): what we review, turnaround, and price
- [SEO for clinics](https://exampleagency.com/services/clinic-seo.md): monthly retainer, what's included and what isn't

## Case studies and pricing

- [Case studies](https://exampleagency.com/case-studies.md): client results with verifiable data
- [Pricing](https://exampleagency.com/pricing.md): packages and terms

## Optional

- [Blog](https://exampleagency.com/blog.md): articles and guides
- [Team](https://exampleagency.com/team.md): bios and experience

Tips for writing a good one:

1. The summary does the heavy lifting. The opening blockquote should say who you are, what you do, for whom, and where, with specifics.
2. Few links, good links. It's not a sitemap. Link what an AI would need to answer questions about you.
3. Link to Markdown if you can. If you don't have .md versions, link to regular HTML pages; that's valid too.
4. Keep it consistent with the site. Nothing in llms.txt that isn't on your pages. Mismatched facts cost you credibility.
5. Serve it as plain text at https://yourdomain.com/llms.txt, with a 200 status code.

Some sites also publish llms-full.txt, with all their content in a single file. That isn't part of the llmstxt.org spec; it's a convention some documentation platforms use.

Who actually uses llms.txt (November 2026)

This is the part most guides gloss over. Here's what we could confirm from official sources:

Google's wording in its official guide: creating llms.txt or similar files "will neither harm nor help your site's visibility or rankings in Google Search, as Google Search ignores them." It adds that Google may crawl and index the file like any other, which doesn't mean it gets special treatment.

The fact that AI labs publish llms.txt for their own docs tells you they value it for agents and developers. It doesn't tell you their search products read it to decide whom to cite. Those are different things, and it's worth keeping them apart.

Common llms.txt mistakes

• Using it to block crawlers. llms.txt grants and blocks nothing. That's what robots.txt is for.
• Turning it into a sitemap. Hundreds of links with no notes help no one; you already have sitemap.xml for that.
• Writing marketing copy. The summary should inform, not sell. Lines like "we're industry leaders" give a model nothing to work with.
• Forgetting about it. If your prices or services change and the file doesn't, you're feeding stale facts to whoever reads it.
• Expecting results in Google. Google has already said it ignores the file.

Is it worth creating one?

Our take:

• Yes, if you have technical docs, an API, or a software product. Coding agents use it, and it can save you support headaches.
• Yes, if it takes you 30 minutes. It's cheap, it forces you to decide which pages matter, and it prepares your site for agents that may use it later.
• Not as a GEO strategy. Don't expect more mentions or citations from having one. Your time pays off more in data-rich content, your robots.txt, and your presence off-site.

How to check yours

1. Open https://yourdomain.com/llms.txt in a browser. It should load without errors.
2. Confirm it starts with a # Title and that the links work.
3. Run Lighthouse in Chrome and look at the agentic browsing section, if your version has it.
4. Or use our free GEO audit, which checks llms.txt, robots.txt, structured data, and more in one pass.

Keep going

llms.txt is a small piece of technical GEO. The bigger levers are in our other guides: what GEO is, AI crawlers and robots.txt, and how to measure AI visibility. In the GEO Plus Academy course we devote a whole module to the technical side, covering what works and what doesn't, with a hands-on task in every lesson.