How to Write an llms.txt File (With a Template for Founders)

To write an llms.txt file, publish a markdown page at /llms.txt with an H1 naming your site, a blockquote summary and H2 lists of the pages an agent should read first. Keep it short, describe each link in a line and choose pages that answer real buyer questions. It takes an afternoon, guarantees nothing and gives an agent a cleaner first read of your product.

TL;DR

• llms.txt is a proposed markdown file at /llms.txt that curates the pages an AI agent should read.
• Version 2 requires an H1, adds a blockquote summary and organises links into H2 file lists.
• It is not robots.txt and not a sitemap: an agent uses it on demand when it needs information.
• Test it by asking an agent questions with only your llms.txt as the starting point.
• It is a proposal, so it cannot guarantee crawling, citation, inclusion or ranking.

Disclosure: ZeroRaised publishes this guide and is a launch site for bootstrapped products. Facts come from the linked sources as checked on 28 Oct 2026; ZeroRaised has no commercial relationship with the companies named.

llms.txt is a proposed markdown file, usually at /llms.txt, that gives AI agents a short curated map of a site: an H1 with the site name, a blockquote summary and lists of useful pages. A file list is an H2 section of links written as [name](url): notes, where the note tells the agent why the page is worth opening.

What is llms.txt and who proposed it?#

llms.txt was proposed by Jeremy Howard and first published on September 3, 2024. It is a markdown file at /llms.txt, or at any subpath, intended to help agents use a website (llmstxt.org). Version 2 was updated on August 10, 2026 (llmstxt.org).

What does the version 2 format require?#

The specification is small. An H1 with the site name is the only required section. Everything else is optional but useful (llmstxt.org):

• a blockquote summary directly under the H1;
• optional paragraphs or lists, without headings, for context;
• H2 sections that contain file lists of [name](url): notes;
• an "Optional" section for links an agent can skip.

The notes matter: they help an agent choose between your quick start, your API reference and a blog post that mentions the same words.

What does a good llms.txt look like for a small SaaS?#

Here is a sample for a fictional support inbox. Copy the block, replace the name and links, and keep the notes to one line. The block is indented by one space; remove it when you copy the sample.

 # Acme Desk

 > Acme Desk is a shared support inbox for small SaaS teams, with email, chat and a help centre in one queue.

 Acme Desk helps a two to ten person support team answer customer questions faster without adding another layer of process.

 ## Docs

 - [Quick start](https://acme.example/docs/quick-start): connect an inbox and send a first reply.
 - [API reference](https://acme.example/docs/api): endpoints, authentication and rate limits.
 - [Pricing](https://acme.example/pricing): plans, seat limits and overage rules.

 ## Company

 - [About](https://acme.example/about): who builds Acme Desk and where it runs.
 - [Changelog](https://acme.example/changelog): shipped changes, newest first.

 ## Optional

 - [Status](https://acme.example/status): uptime and incident history.
 - [Careers](https://acme.example/careers): open roles.

Notice what is missing: no history, no list of every post, no marketing language. The file answers three questions: what is this, where are the docs, and what does it cost.

How is llms.txt different from robots.txt and sitemap.xml?#

robots.txt says what access is acceptable, and llms.txt is used on demand when an agent needs information (llmstxt.org). One is a permission signal, the other is a content guide.

A sitemap.xml lists everything and is not a substitute for llms.txt (llmstxt.org). A sitemap helps a crawler discover pages. An llms.txt helps an agent choose among them and understand why each one exists. Publish both, and keep the pages themselves crawlable.

What else did version 2 add?#

Version 2 also proposes clean markdown versions of pages at the same URL with .md, plus the rel="alternate" type="text/markdown" and rel="describedby" link relations, which can be sent as an HTTP Link header (llmstxt.org). A markdown copy of your pricing page gives an agent clean text instead of a rendered layout to parse.

How do you test an llms.txt file?#

The specification says to use concise language, brief and informative link descriptions, no jargon, and to test by asking an agent questions with only your llms.txt as a starting point (llmstxt.org).

1. Open a fresh agent session with no memory of your product.
2. Give it only your llms.txt URL, or paste the file.
3. Ask what a buyer would ask: what does it do, what does it cost, how do I connect it.
4. Note every question it cannot answer, fix the file or the pages, then ask again.

If the agent has to guess, the file is incomplete. A dead link is worse, because it wastes the one chance you had to be useful.

Does llms.txt guarantee AI citations?#

No. It is a proposal, not a standard. Thousands of sites publish one, Chrome's Lighthouse audits sites for one as part of its agentic browsing checks, and OpenAI, Anthropic and Gemini publish llms.txt files for their developer docs. Mintlify, GitBook, Yoast SEO, AIOSEO and Wix can generate one (llmstxt.org).

The honest framing is ZeroRaised's own line: "This file is a concise discovery aid. It does not guarantee crawling, citation, inclusion, or ranking by any system." (ZeroRaised llms.txt). That is not modesty, just an accurate description of a convention systems may read, ignore or never request.

Treat it as one more way to make your site legible, not as a ranking tactic. The guide to optimizing for AI search answers covers the parts that carry more weight.

A practical checklist for shipping llms.txt#

1. Write the H1 and the summary. Name the site, then say in one sentence what it does and who it is for.
2. Pick five to ten pages. Docs, pricing, about, contact and the page that explains your main use case.
3. Add a note to every link. Say what the agent will find, not how excited you are about it.
4. Use Optional for skippable links. Status, careers, legal pages and old announcements.
5. Test with real questions and fix the gaps.
6. Revisit when facts change. A stale price is worse than no file at all.

Where ZeroRaised fits#

ZeroRaised is a launch and discovery site for products built without institutional venture capital or accelerator equity. If you have a product people can try, you can submit it here. A launch page gives agents and buyers a stable set of facts: what the product does, who made it, how to try it and what it costs.

Keep your llms.txt, launch page and site agreeing on the same claims. The guide to marketing an API covers how to present technical products without overselling, and our directory submission guide covers where else to list the facts. For more analysis, join the ZeroRaised newsletter. We promise no traffic or results, from this file or a launch.

Sources#

• llmstxt.org: the llms.txt specification, version 2
• ZeroRaised llms.txt