llms.txt Examples: Real Files From Real Sites (and What to Copy)

What does a real llms.txt file look like?

The fastest way to understand llms.txt is to read files that companies actually publish. Several developer-focused companies do, and their files are public: you can open any of them in a browser and read the raw Markdown. That is the best way to calibrate what yours should look like.

The short version: the format holds up everywhere, but the files differ widely in size and intent. A short link index and a sprawling documentation dump both count as valid. What matters is that each one answers the same question for an AI system: what is this site, and where is the good stuff?

How real llms.txt files are structured

Here is how a handful of published files are organized, from the leanest to the largest. The point is the shape of each file, not a leaderboard of sizes.

• Supabase: the minimalist. One heading, a pointer to its llms-full.txt companion, then a flat list of documentation links. No prose at all, because the docs pages speak for themselves.
• Zapier: the hub. A short agent-oriented summary at the top, then links out to per-section indexes (and an expanded llms-full.txt), so an AI can fetch only the surface it needs. It explicitly labels itself agent-readable.
• Cloudflare: the sectioned guide. Developer docs grouped by product area, each group and link annotated so an agent can jump to the right product.
• Stripe: the agent brief. The file opens with direct instructions to AI coding assistants about how to integrate Stripe correctly, before any documentation links appear.
• Anthropic: the full index. Its developer documentation indexed in one large file, at a scale that shades into llms-full.txt territory.

The same format stretches from a bare list of links to an index of an entire documentation set, without changing shape. Size follows the site, not the spec.

The four patterns worth copying

Read enough of these files and four distinct patterns emerge. Pick the one that matches what your site is, not what the biggest company does.

• The minimalist index (Supabase): just links to clean content. Best for documentation and content sites where the pages speak for themselves.
• The curated business summary: name, one-paragraph pitch, key pages with notes. Best for small businesses, agencies, and local services. This is the original intent of the spec.
• The hub of indexes (Zapier): a root file pointing to per-section files. Best for companies with several products or subdomains.
• The agent brief (Stripe): instructions for AI systems that act, not just read. Best when getting the details wrong costs your users money or broken code.

Stripe's pattern is the one to watch. It treats the file less like a sitemap and more like an onboarding note for AI agents. As assistants do more real work, that framing will spread.

A small-business llms.txt example

Big-company files are instructive, but most sites reading this need the curated summary pattern. Here is a trimmed version of the file this site publishes, which you can adapt directly:

# JNAbear

> JNAbear is an SEO content agency run by Joseph Nicholas Abear in Portland, Oregon. It builds entity-based SEO content strategies designed to rank in Google and earn citations in AI search.

Contact: info@jnabear.com.

## Services

- [Local SEO Clusters](https://jnabear.com/local-seo-clusters): Interlinked content built to win the map pack. Projects from $1,800.
- [Content Audits](https://jnabear.com/content-audits): Page-by-page review with a ranked fix list. From $750.

## Guides

- [What Is llms.txt?](https://jnabear.com/what-is-llms-txt): The AI crawler file explained, with an honest read on whether it works.

Notice the prices. That is deliberate. When someone asks an AI assistant what a service costs, the file is your chance to answer in your own words. Leave prices out and the assistant guesses, or quotes a competitor who did include them.

What all the strong files get right

Across files from very different companies, three traits repeat. First, a clear H1 with the real name of the site or product. Second, a summary near the top that a stranger could repeat accurately. Third, a short note on every link explaining why it is worth fetching. That last one is the difference between a map and a pile of URLs.

The files also stay plain. No HTML, no styling, no tricks. Markdown headings and lists are the whole toolkit, which is exactly why every language model can read them.

How big should your llms.txt file be?

As small as it can be while still answering the questions people ask AI about you. A serious product can get by with a short index of a few dozen links. A local business or agency usually needs even less: a summary and its handful of key pages. You only need documentation-scale files when you have documentation-scale content.

If you want to skip the formatting work, my free llms.txt generator builds a valid file from a short form, and the full what-is-llms.txt guide covers where to upload it and how to keep it honest. Start small, keep it current, and let the file grow only when your site does.