llms.txt examples from real websites, with a template
·4 min read
The best llms.txt example to copy is a short one. On October 10, 2026 we fetched the files of nine developer sites with curl, and the most useful ones are short files that send an agent to the right Markdown in one hop. Svelte and Vercel do this well. Cloudflare splits a huge docs site into per-product files. Stripe and Anthropic ship 700 to 800 lines that read more like a sitemap.
This post is only the examples. For the format itself, read what llms.txt is, and for writing and hosting your own, the llms.txt creation guide.
llms.txt examples at a glance
Each file was fetched with curl -sL on the date above. Excerpts below are trimmed, with link notes shortened.
| File | Lines | Links | Worth copying |
|---|---|---|---|
| svelte.dev/llms.txt | 23 | 7 | Size variants of the full docs |
| vercel.com/llms.txt | 65 | 25 | Plain instructions for agents |
| developers.cloudflare.com/llms.txt | 144 | 113 | One child llms.txt per product |
| developers.openai.com/llms.txt | 53 | 40 | A "Common tasks" section |
| supabase.com/llms.txt | 55 | 41 | llms-full.txt linked on line 3 |
| docs.stripe.com/llms.txt | 710 | 455 | Version rules up top, little else |
| docs.anthropic.com/llms.txt | 830 | 760 | Not much |
| docs.perplexity.ai/llms.txt | 181 | 171 | A real Optional section |
platform.openai.com/llms.txt returned a 404 with an HTML page. OpenAI's file lives on developers.openai.com. The Anthropic URL redirects to platform.claude.com/llms.txt, so an agent has to follow a redirect to read it.
Small llms.txt examples that route agents well
The small files share one habit. They tell an agent which bigger file to open next, and stop.
Svelte
# Svelte Documentation for LLMs
## Documentation Sets
- [Abridged documentation](https://svelte.dev/llms-medium.txt): ...
- [Compressed documentation](https://svelte.dev/llms-small.txt): ...
- [Complete documentation](https://svelte.dev/llms-full.txt): ...Seven links in 23 lines. The size variants matter because the complete file is about 1.26 MB, too much for many context windows. An agent with little room takes the small one. The weak spot is the blockquote, which calls the framework "breathtakingly concise". That is a line no model needs.
Vercel
## How agents should use Vercel
- For research, fetch the Markdown documentation indexes below ...
- For REST API calls, read the OpenAPI description ... before choosing an operation.
Ask for approval before changing account resources.Vercel spends a section on behavior, and it is specific. Read the OpenAPI spec before calling the API. Ask before changing anything. It also ends with ## Optional, which the spec reserves for links an agent may skip.
Cloudflare
## Developer platform
- [Workers](https://developers.cloudflare.com/workers/llms.txt): Build, deploy, and scale serverless applications ...
- [Workflows](https://developers.cloudflare.com/workflows/llms.txt): Build durable, multi-step applications ...Every link points to another llms.txt, one per product. That is the right shape for a docs site with thousands of pages. The root stays at 144 lines and each product keeps its own index current. Its notes are the product taglines, though, which say little about what each index covers.
Large llms.txt examples that read like sitemaps
Big companies tend to dump their whole docs tree into the file. It works as an index, but an agent has to read tens of kilobytes to pick one page.
Stripe
# Stripe Documentation
When installing Stripe packages, always check the npm registry for the latest version
rather than relying on memorized version numbers. ...The opening paragraphs are the best part of any file here. They fix the exact mistake a model makes, installing an old SDK version from training data. Later, a section titled "Instructions for Large Language Model Agents" tells agents to prefer Checkout Sessions and never recommend the Charges API. Then come 455 links in 92 KB. Copy the top 10 lines and skip the rest.
Anthropic
### Messages
- [Overview](https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview.md) - Agent Skills
- [Features overview](https://platform.claude.com/docs/en/build-with-claude/overview.md)
- [Quickstart](https://platform.claude.com/docs/en/agents-and-tools/agent-skills/quickstart.md) - Get started ...
- [Quickstart](https://platform.claude.com/docs/en/get-started.md) - Get started with ClaudeThis file has no blockquote summary. Its links sit under H3s inside one ## English section, and 625 of the 760 have no note at all. Sixteen are titled "Overview" and five "Quickstart", so the title alone cannot tell an agent which one it wants. The notes that exist follow a spaced hyphen instead of the colon the spec uses.
Perplexity
- [Quickstart](https://docs.perplexity.ai/docs/getting-started/quickstart.md): Generate an API key ...
## Optional
- [Community](https://community.perplexity.ai)Most of Perplexity's 171 links carry a clear note and a .md URL, and community links sit under Optional where they belong. The OpenAPI Specs section breaks the pattern. Its six links are relative paths like /openapi.json, which point nowhere once an agent reads the file away from the domain. The spec examples use full URLs.
A minimal llms.txt template
Take Svelte's size, Stripe's opening rules and Vercel's agent section. For a small product, this is enough.
# Product name
> One sentence on what it is and who it is for.
Facts an agent needs first, such as the current API version, currency or a
deprecated method it must not suggest.
## Docs
- [Quickstart](https://example.com/docs/quickstart.md): Create a key and make the first call
- [API reference](https://example.com/docs/api.md): Endpoints, auth and rate limits
## How agents should use this site
- Read the API reference before writing requests.
- Ask the user before changing account data.
## Optional
- [Changelog](https://example.com/changelog.md): Releases by dateKeep every URL absolute, give every link a note that names the page's facts, and move bulk content into llms-full.txt or per-section files the way Cloudflare does.
Do AI search engines read llms.txt?
None of the major AI search engines has said it uses llms.txt to choose sources. Google's AI optimization guide says Search ignores the file. The companies above publish one for their developer docs, which suggests the real readers are coding agents working with an API, not crawlers deciding whom to cite. That is also why the best examples here talk to agents in the imperative.
If your site is a store or a local business, a short file is fine and the long ones above are the wrong model. Crawler access in robots.txt still matters more, and our AI crawler list has the user agents.
Check your llms.txt against these examples
The AI Agent Readiness Checker fetches your /llms.txt, or a draft you paste, and parses it. It flags a missing H1 or blockquote summary, a heading in the preamble, empty sections, items that are not Markdown links, relative, non-public or duplicate URLs, links without notes, an HTML response and an unexpected content type. On the files above it would flag Anthropic's links without notes and Perplexity's relative URLs.
The same run applies Lighthouse's Agentic Browsing checks, tests whether your pages answer Accept: text/markdown, and looks for WebMCP form coverage and MCP or A2A agent cards. It reads up to 128 KiB of the file and does not open the linked pages, so test those with curl. A run costs 10 credits.