# Writing an llms.txt

A plain markdown file that tells AI models what's on your site, with a real one to copy.

Guide by Jon C. Phillips · November 16, 2026 · https://joncphillips.com/llms-txt-what-it-is-an-example-and-a-template/

This site has one, alongside markdown versions of every page. It's one of the cheapest ways to make a site easier for AI tools to read, and you can write one in 10 minutes.

## What llms.txt Is

llms.txt is a proposal from Jeremy Howard of Answer.AI, published in September 2024 at [llmstxt.org](https://llmstxt.org). The idea is that websites are built for people, with navigation and layout everywhere, and a language model trying to understand a site wastes most of its limited context window on all that.

So you give it a shortcut: a single markdown file at `/llms.txt` that says what the site is and links to the pages that matter, with a line about each one.

It's closer to a table of contents than to robots.txt. robots.txt tells crawlers what they may fetch, and llms.txt tells a model what's worth reading.

## The Format

The spec is short. In order, an llms.txt has:

1. An H1 with the name of the site or project. This is the only part that's required.
2. A blockquote with a one-line summary.
3. Optionally, a few paragraphs or lists with anything else a model should know, without headings.
4. Sections under H2 headings, each holding a list of links in the form `- [Name](url): notes`.

There's one special section. Anything under `## Optional` is marked as skippable, so a tool short on context can leave it out.

## A Real Example

Here's a trimmed version of the one on this site, which you can see in full at [joncphillips.com/llms.txt](https://joncphillips.com/llms.txt):

```markdown
# Jon C. Phillips

> Serial builder of side projects and open source software.

This is the personal website of Jon C. Phillips, a product engineer from
Montreal. The site publishes articles, a weekly newsletter, a free ebook,
and a portfolio of current and previous projects.

How to read it: every page is also available as markdown. Append .md to
its URL, or request the page with an `Accept: text/markdown` header.

## Main pages

- [Articles](https://joncphillips.com/articles/): Every article, newest first.
- [Tools](https://joncphillips.com/tools/): Free tools for people building on the web.
- [Work](https://joncphillips.com/work/): Current and previous projects.

## Articles

- [Two Days](https://joncphillips.com/two-days/): November 16, 2026. I built Runlight in a little over a month.
- [Software is Not Dead](https://joncphillips.com/software-is-not-dead/): November 8, 2026. Pieter Levels thinks we've entered the era where everyone builds their own software.

## Optional

- [Articles RSS feed](https://joncphillips.com/articles.rss): The 10 newest articles.
- [Sitemap](https://joncphillips.com/sitemap.xml): Every page on the site.
```

A few things about that file are deliberate. The paragraph about markdown versions tells an agent how to get clean text for any page, which saves it from parsing HTML. Every link has a short note after it, because a bare list of titles doesn't tell a model which page answers which question. The feed and sitemap go under Optional, since a model rarely needs them.

Mine is generated on every request, so a new article shows up in it the moment it goes live. If your site has a build step or a CMS, generating it is better than writing it by hand, because a stale llms.txt is worse than none.

## A Template You Can Copy

```markdown
# Your Site or Product Name

> One sentence on what it is and who it's for.

A short paragraph with the context a model needs: what you do, and
what a reader would get wrong without being told.

## Docs

- [Getting started](https://example.com/docs/start): Install and first steps.
- [API reference](https://example.com/docs/api): Every endpoint, with examples.

## Product

- [Pricing](https://example.com/pricing): Plans and what's free.
- [Changelog](https://example.com/changelog): What changed and when.

## Optional

- [Blog](https://example.com/blog): Announcements and longer posts.
```

Swap the sections for whatever your site actually has. A docs site might have Guides and Reference, and a personal site might have Writing and Projects. Keep each note to a line.

## llms-full.txt

You'll also see `llms-full.txt` on a lot of documentation sites. It wasn't in the original proposal, and it caught on anyway. It holds the full text of every page in one file, so a tool can load an entire set of docs in one fetch.

It makes sense for documentation, where someone might paste the whole thing into an assistant's context. For a blog or a marketing site it's usually overkill, and a big file of every post isn't something most tools need.

## Markdown Versions of Your Pages

The spec also suggests serving a markdown version of each page at the same URL with `.md` added (for an address that ends in a slash, it suggests `index.html.md` or `index.md`), and that's the part I'd prioritize if you only do one thing. An llms.txt points a model at pages, and those pages are still HTML unless you give it something better.

On this site every page has a markdown twin, like `/about.md` for `/about/`, and a request with `Accept: text/markdown` gets markdown from the normal URL too. I wrote about why in [An Audience of Agents](https://joncphillips.com/an-audience-of-agents/), after Cloudflare said agents now visit more pages than people do.

## Does Anyone Actually Use It?

**No major AI company has said its crawlers read llms.txt.** People at Google have been openly skeptical of it, and on this site the AI crawlers spend nearly all their visits on ordinary pages.

Where it does get used is by tools working on someone's behalf. Coding agents and assistants reading documentation are the main case, and several of them look for llms.txt or markdown versions when they're pointed at a site. That's why it's most common on developer docs today.

So I think of it as a small bet. It doesn't take long and can't hurt your search rankings. If the agents that read the web lean on it more over time, you're already set up.

## Check Yours

Once yours is live, run it through the [llms.txt Checker](https://joncphillips.com/tools/llms-txt-checker/). It checks the file against the llmstxt.org format and follows every link to make sure none are broken. If you don't have one yet, it can draft a starting file from your sitemap.

For the bigger picture, the [AI Agent Readiness Check](https://joncphillips.com/tools/agent-readiness-check/) looks at llms.txt alongside your robots.txt rules for AI crawlers and whether your pages come in markdown. And if you're deciding which of those crawlers to let in, my guide to [robots.txt for AI crawlers](https://joncphillips.com/robots-txt-for-ai-crawlers-gptbot-claudebot-and-the-rest/) covers each one.

Topics: [AI](https://joncphillips.com/topics/ai/), [Distribution](https://joncphillips.com/topics/distribution/)
