What Is llms.txt, and Do AI Engines Use It?

llms.txt is a proposed Markdown file at the root of a website that gives language models a short summary of the site and a curated list of links to its most useful pages. It's a reading guide, not a permission system. It's cheap to publish, but no major answer engine has documented using it to pick sources.

Where did llms.txt come from?

Jeremy Howard of Answer.AI proposed llms.txt in 2024. The problem it targets is real. Websites are built for humans, full of navigation, scripts and layout, and a language model reading a site at question time has limited room in its context window. A short, clean Markdown file that says “here's what this site is and where the good stuff lives” is easier for a model to use than a crawl of the whole site.

The proposal also suggests offering clean Markdown versions of important pages at the same URL with .md appended, so a model can read the content without the HTML wrapped around it.

Note what it's designed for: helping a model use your content when it's already reading your site, typically at the moment it answers. It isn't a ranking signal, a sitemap replacement or a way to control crawling.

What does an llms.txt file look like?

It's plain Markdown with a light structure, in this order:

1. An H1 with the site or project name. This is the only required element.
2. A blockquote with a one- or two-sentence summary of what the site or product is.
3. Optional free text with key details, in paragraphs or lists, but no headings.
4. H2 sections containing link lists. Each item is a Markdown link, optionally followed by a colon and a short note.
5. An optional section titled “Optional” for secondary links a model can skip when it's short on space.

Here's an example for a fictional analytics company:

# Acme Analytics

> Acme Analytics is product analytics software for B2B SaaS teams. It tracks feature adoption, retention and account health without engineering work.

Acme is priced per tracked account, and every plan includes unlimited seats. It is not designed for consumer mobile apps.

## Product

- [Features](https://www.example.com/features.md): What Acme tracks and how it works
- [Pricing](https://www.example.com/pricing.md): Plans, limits and what each includes
- [Integrations](https://www.example.com/integrations.md): Supported data sources and setup

## Docs

- [Quickstart](https://www.example.com/docs/quickstart.md): Install the snippet and send a first event
- [API reference](https://www.example.com/docs/api.md): Endpoints, authentication and rate limits

## Optional

- [Changelog](https://www.example.com/changelog.md): Release notes by month
- [Customer stories](https://www.example.com/customers.md): How teams use Acme

How is llms.txt different from robots.txt and sitemaps?

They sound related and do completely different jobs:

The practical upshot: llms.txt can't open doors that robots.txt or your firewall closed, and it can't stop anyone training on your content. To control AI crawling, start with the AI crawlers and robots.txt guide.

Do ChatGPT, Google, Claude or Perplexity use llms.txt?

Here's the honest picture as of late 2026:

• No major answer engine has documented using llms.txt to decide what to retrieve, rank or cite. Their crawler documentation is about robots.txt.
• Google has said publicly that it doesn't use llms.txt. Its guidance for AI Overviews and AI Mode says there are no additional requirements or special optimizations beyond normal Search eligibility.
• Some site owners do see requests for /llms.txt in their logs. A fetch proves something read the file, not that it changed any answer.
• Where it clearly helps is on-demand reading. Coding assistants, agents and tools that accept a URL can use llms.txt and Markdown pages to find the right docs fast, which is why many developer platforms publish one.

The case for publishing one anyway is decent. It's cheap. Writing it forces you to state, in a few sentences, the facts you most want models to get right, which is a useful exercise on its own. And if an engine does start reading it, you're ready. The case against overinvesting is just as plain: it needs upkeep, and every hour spent polishing it is an hour not spent on crawler access, content or coverage, which demonstrably matter.

Should you publish an llms.txt file?

Probably, if it's cheap for you. The cost is an hour or two plus some upkeep, and the downside is close to zero. But be clear about the order of operations:

1. Make sure answer-engine crawlers can reach your site.
2. Make sure your key pages answer buyer questions directly.
3. Earn coverage in the sources engines already cite.
4. Then publish llms.txt as a clean summary of all of the above.

How much it's worth depends on what kind of site you run:

How do you write a good llms.txt?

• Lead with the facts you most need models to get right. Category, audience, pricing model, regions served, what you don't do.
• Curate, don't dump. Link the canonical pages that answer real questions, not every URL in your sitemap.
• Describe each link in plain words. The note after the colon should say what the page answers.
• Offer Markdown versions if you can. Some docs platforms and site generators can produce .md copies automatically.
• Keep it in sync. Generate it from your CMS, or review it whenever pricing, products or positioning change.
• Leave out anything private. The file is public. Don't list gated, staging or internal URLs.

Weak summary vs strong summary

The blockquote does the most work, so spend your time there. A weak one reads like a slogan: “Acme is the leading platform helping modern teams do more with data.” A model learns nothing from that. A strong one reads like a fact sheet: “Acme Analytics is product analytics software for B2B SaaS teams. It tracks feature adoption, retention and account health, and is priced per tracked account.” Category, audience, what it does, how it's priced. If a model quoted only that sentence, would a buyer understand what you sell? That's the test.

How do you generate and serve llms.txt?

There are three realistic routes, depending on the size of the site:

• Write it by hand. For a marketing site with a few dozen important pages, a hand-written file is the clearest option. Put it in version control so changes get reviewed like any other copy.
• Generate it from your CMS or docs platform. Some SEO plugins and documentation platforms can build the file for you. Check the output: an auto-generated list of every URL defeats the point of curation.
• Build it at deploy time. If your site is code, generate the file from the same source as your sitemap and product data, so pricing and page titles can't drift out of sync.

However you build it, serve it as a static file with a plain-text or Markdown content type, allow caching, and keep the URL stable. Treat changes to it the way you'd treat changes to your pricing page, because to a model reading it, that's roughly what it is.

How do you check that your llms.txt works?

• It loads at https://yourdomain.com/llms.txt with a 200 status and a text content type.
• Your CDN or firewall doesn't block or challenge it for AI user agents.
• Every link resolves, and Markdown versions return Markdown, not an HTML shell.
• Paste the URL into an assistant that can browse and ask it to summarize your product. If the summary is off, the file is unclear.

Elite Answers' AEO site audit includes an llms.txt check alongside per-bot robots.txt rules and firewall probing, so you see the whole access picture in one place. For the one-line definition, see the llms.txt glossary entry.