llms.txt Examples
llms.txt examples from three real sites.
Real llms.txt files from Anthropic, Cloudflare and Vercel, what each does well, and a template you can adapt to your own site.
Short answer
What does a good llms.txt file look like?
Among real llms.txt examples, Cloudflare's file shows one clear pattern. A plain sentence sits under the H1, then a blockquote gives navigation tips. Each link follows [name](url): description and points to a product's own llms.txt. The spec allows per-path files, and agents use the most specific one.
In brief
Five things the examples show.
Only the H1 is required
The proposal says the only required element is an H1 with the name of the project or site.
Structure beats length
All three files group links under clear headings, each link with a short note.
Large sites split the file
Cloudflare's root file indexes per-product llms.txt files instead of listing every page.
Agent guidance is allowed
Vercel adds plain-language sections that tell agents how to use its platform.
Google doesn't use it
Google says Google Search itself doesn't use llms.txt files, so don't expect a ranking effect there.
The format
What the llms.txt format requires.
An llms.txt file is a Markdown index of links to your pages. AI referrals already matter: Mintlify reports Vercel attributes roughly 10% of new signups to ChatGPT referrals. Track those referrals to measure results.
Only the H1 is required. Everything else is optional, which is why real files differ so much.
| Part | Required? | What it does |
|---|---|---|
| H1 title | Yes | Names the project or site |
| Blockquote | Optional | Short summary of key facts |
| Free text | Optional | Details, with no headings |
| H2 file lists | Optional | Links as [name](url): notes |
| Optional section | Optional | Secondary links an agent can skip |
Example 1
Anthropic: a full documentation index.
Anthropic's file opens with an H1 and a one-line overview. Links point straight to Markdown (.md) versions of pages.
It is very large: it lists every English documentation page and API reference endpoint.
Language counts up front
English has 762 pages included; 11 other languages point readers to the website.
H3 sections by topic
Messages, Managed Agents, Admin, Best practices, Models & pricing, SDKs and API reference.
Direct .md links
Each link goes to a Markdown page, with a short title and often a description.
Slim index, full export
Mintlify notes Anthropic also publishes a larger llms-full.txt with the full documentation.
Example 2
Cloudflare: an index of indexes.
Cloudflare's root file doesn't list pages. It links to a separate llms.txt for each product and calls that the recommended way to explore one.
That keeps the root file short while each product file stays complete.
| Cloudflare's choice | Why it helps | |
|---|---|---|
| Per-product files | Links go to each product's own llms.txt | Root stays short; detail lives lower |
| H2 categories | Application performance, Application security, Developer platform | Readers scan by need |
| Link pattern | [Workers](url): one-line description | Each entry says what it covers |
| Docs for agents | A dedicated entry on agent access | Explains how agents read the docs |
Want to see if AI can read your site?
Our technical AEO work covers crawler access, llms.txt, schema and rendering, so the pages you want cited can be read.
Example 3
Vercel: guidance for agents.
Vercel's file starts with an H1 and a blockquote summary. It then adds free-form sections such as “When to use Vercel” and “How agents should use Vercel.”
It also links a full-content file of all docs and an OpenAPI description.
Rules for agents
It tells agents to read the OpenAPI and authentication docs first and ask approval before changing account resources.
Machine-readable extras
A complete docs file and REST API reference sit in one linked file.
Setup instructions
It documents an OAuth-authenticated MCP server and a plugin install command.
An Optional section
An experimental agent resource catalog sits under Optional, so it can be skipped.
Template
A simple llms.txt template.
This llms.txt template follows the proposal's structure. Replace each placeholder with your own pages.
# Your Company
Start with an H1 naming your company or product. This is the only required line.> One-line summary
Add a blockquote stating what you do, in plain words, with the key facts.Add short context
Write a few plain lines under no heading: who you serve and where you work.## Services, with links
Group your links under H2 headings, as Cloudflare does with its product categories. List each page as a link with a short description after it.## Optional
Put secondary pages here, so an agent can skip them when it needs less.
Writing yours
How to write your own.
The proposal asks for concise, clear language, brief link descriptions and no jargon. Link to clean Markdown versions of pages where you have them.
Curate, don't dump
The proposal calls llms.txt a curated overview. A sitemap lists everything; this file lists what matters.
Describe every link
Add a short note so an agent knows what a page covers before opening it.
Match your site's logic
Mintlify shows Stripe grouping by product, Cloudflare by vertical and Supabase by language.
Test it with an agent
Ask an agent questions using only your llms.txt as a starting point.
Not sure where to start?
The audit shows the buyer questions where you are missing and what to fix first.
Expectations
What llms.txt will and won't do.
LLMReach can help, and the first step is a free AI audit. The audit checks your visibility, mentions and citations for your buyers' questions. It shows which gaps to fix first.
The proposal's authors expected llms.txt to help mainly with inference, when an agent answers on demand, rather than training.
| Where it may help | Where it doesn't | |
|---|---|---|
| Agents and tools | Coding agents follow it to API references and tutorials | Google Search says it does not use it |
| Adoption | OpenAI, Anthropic and Gemini publish their own | Google says its AI features rest on core Search systems |
| Your priority | A clear map for tools that choose to read it | Crawlable, useful, indexed pages still come first |
FAQ
llms.txt examples: common questions.
What does llms.txt look like in practice?
A sample llms.txt is a real Markdown file a site publishes. Anthropic and Cloudflare both open theirs with an H1 title and a plain-text sentence. The spec suggests a blockquote summary there instead. Both then list links.
What should an llms.txt template include?
An llms.txt template needs an H1 with your name, since that is the only required element. Add a short blockquote summary, optional context, H2 sections of links written as [name](url): notes, and an Optional section for secondary pages.
Does Anthropic publish an llms.txt file?
Anthropic publishes one for its developer documentation. It lists English pages as Markdown links under topic headings. The llmstxt.org proposal notes that OpenAI, Anthropic and Gemini all publish llms.txt files for their developer docs.
How is Cloudflare's llms.txt different?
Cloudflare's root file links to a separate llms.txt for each product instead of listing pages. Each product file holds that product's full index, which Cloudflare calls the recommended way to explore it. The root stays short as a result.
Does Google use llms.txt?
Google Search does not use llms.txt. Its AI optimization guide says you don't need llms.txt files or other AI text files to appear in Google Search, including its generative AI features, and that you can ignore creating them.
Is llms.txt a replacement for a sitemap?
llms.txt does not replace a sitemap. The proposal says sitemaps list all pages for search engines, while llms.txt offers a curated overview for language models. It can also complement robots.txt, which tells tools what access is acceptable.
Where do I put an llms.txt file?
An llms.txt file goes at your site root, at /llms.txt. The proposal also allows it at any subpath, where it covers the URLs under that path. Where several files apply, agents should use the most specific one.
Free AI Audit
See how AI engines read and cite your site.
You leave knowing where the gap is, what is causing it, and which changes would matter first, whether we work together or not.
What to expect
Audit
Before the call, we run your buyers' questions through ChatGPT, Claude, Perplexity and Gemini.
Map
On the call, we show you where competitors are cited and you are not, and what's causing it.
Decide
An honest read on whether the Citation Stack fits. If it doesn't, you'll hear it on the call.
Don't see a time that works? Email Karim directly: contact@llmreach.ai