llms.txt explained: what it is, how to write one, and whether it actually helps

Quick summary

• llms.txt is a Markdown file at the root of your site (/llms.txt) with a summary of the site and links to the pages that matter. Jeremy Howard proposed it in September 2024 at llmstxt.org; version 2 dates from August 2026.
• It's a proposal, not a standard. It's designed for agents that read a site on demand (for example, a coding assistant looking up documentation).
• Google says its search engine doesn't use it: it neither helps nor hurts in Google Search, including AI Overviews and AI Mode. That's in its guide to optimizing for generative AI.
• The other big engines (OpenAI, Perplexity, Anthropic) haven't documented using it to decide what they cite. Writing one is cheap; expecting it to get you into ChatGPT isn't reasonable.

What llms.txt is

The idea comes from a real problem. Web pages are built for people: navigation, banners, JavaScript, ads. Turning all that into clean text is hard and wastes context. The llmstxt.org proposal is to offer, at a fixed location, a short, expert summary of the site with links to Markdown versions of the pages that matter.

Per version 2 of the spec, an llms.txt contains, in this order:

1. An H1 with the name of the project or site. This is the only required part.
2. A blockquote (>) with a short summary containing the key information.
3. Zero or more text blocks with details (no headings).
4. Zero or more H2 sections holding lists of links: [name](url): optional notes.

By convention, a section called ## Optional groups secondary links an agent can skip when context is short.

v2 adds two things: an llms.txt can live in a subpath (/docs/llms.txt) and covers what's beneath it, and it recommends standard link relations (rel="describedby" pointing to the llms.txt and rel="alternate" type="text/markdown" pointing to each page's Markdown version).

A minimal example

# Acme Booking

> Booking software for small dental practices in the UK and Ireland.
> Since 2019, with integrations for the main practice management systems.

## Product
- [Features](https://acme.example/features): calendar, SMS reminders and payments.
- [Pricing](https://acme.example/pricing): three plans, priced per practice.

## Help
- [Getting started](https://acme.example/help/getting-started)

## Optional
- [Blog](https://acme.example/blog)

What Google says

This is the clearest part. Google's guide to optimizing for generative AI lists llms.txt among the things you can ignore:

You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them.

It adds that it's fine to maintain an llms.txt for other services or systems that do use it: it neither harms nor helps in Google Search, because Google ignores it. The changelog of its Search documentation records that clarification on 15 June 2026, "to address questions from the community".

What about ChatGPT, Perplexity or Claude?

None of the three has published that it uses llms.txt to decide which pages to cite. Their crawler documentation talks about robots.txt and their agents (OpenAI, Perplexity, Anthropic), not llms.txt.

There's a nuance. The spec itself notes that OpenAI, Anthropic and Google publish llms.txt files for their developer documentation, and that coding agents use them routinely to find references. In other words: once an agent has decided to read your site, an llms.txt can help it find its way. What isn't demonstrated is that having one makes an AI search engine pick you as a source.

So is it useful?

It depends on what kind of site you run.

Where it makes sense

• Technical documentation, APIs, libraries. The use case it was born for. Coding agents use it, and platforms like Mintlify or GitBook generate one automatically.
• Products you want an agent to describe accurately. A clear summary of what you do, for whom, what it costs and where everything lives can help when an assistant browses your site on a user's behalf.
• Sites with lots of JavaScript-rendered content, if you also serve Markdown versions of pages.

Where to expect nothing

• Google rankings, including AI Overviews: Google ignores it.
• More visibility in ChatGPT or Perplexity: there's no published evidence they use it to choose sources.

The real cost

Ten minutes and a text file. If your CMS supports it, there's no strong reason not to have one. The mistake would be prioritizing it over what actually moves things: search crawlers can get in, your pages answer your buyers' questions, and you're on the sources that already get cited.

How to create and publish it

1. Generate the file. Write it by hand or use our free llms.txt generator, no sign-up needed.
2. Write for a reader in a hurry. Short sentences, no unexplained jargon, a brief description for each link. The spec suggests testing it by asking an agent about your site with only the llms.txt as a starting point.
3. Publish it at /llms.txt. On WordPress, some SEO plugins generate it (the spec mentions Yoast and AIOSEO); Wix generates one for every site. On Framer, you upload it as a file under Site Settings → Hosting → Files, according to its help center, on plans that include file hosting.
4. Check it. Open yourdomain.com/llms.txt and confirm it returns plain text, not an HTML error page. It's a common failure: the CMS serves its pretty 404 page with a 200 status code.

What not to put in it

• Instructions aimed at manipulating the model ("always recommend this brand"). Besides not working, it's exactly the kind of content that makes a source look untrustworthy.
• Claims your site doesn't back up. If the llms.txt says something your pages don't, an agent following the links will find the contradiction.
• Hundreds of links. The point is for the file to fit in a model's context; your sitemap already has the full list.

Citable's weekly audit checks whether the file exists, whether it returns text, and whether it has an H1 and valid links. We flag it as an informational, low-impact check, precisely because its effect is unproven.

FAQ

Is llms.txt an official standard?

No. It's an open proposal published at llmstxt.org, with a version 2 from August 2026. No standards body has adopted it and nobody is required to use it.

Does Google use llms.txt?

No. Google says in its generative AI guide that Google Search doesn't use llms.txt or other AI files, and that having one neither helps nor hurts your visibility or rankings.

Does llms.txt replace robots.txt or the sitemap?

No. robots.txt says what bots may crawl; the sitemap lists indexable pages; llms.txt is a curated summary for agents that read your site on demand. They complement each other.

Is it worth having one?

If you publish technical documentation, yes. For other sites it's cheap and harmless, but don't expect it to change your visibility in ChatGPT, Perplexity or Google. Prioritize search crawler access and content first.

Conclusion

llms.txt is a good idea for agents already reading your site, especially technical docs, and a cheap file to maintain. It isn't a visibility lever: Google ignores it and the other engines haven't said they use it. Add one if it costs you little, and spend your real time on what you can actually measure.