llms.txt Field Reference

The llms.txt specification requires only an H1 project name and a blockquote summary, plus optional H2 sections containing markdown link lists. Companion conventions include page-level .md mirrors and a single-file llms-full.txt for inlined documentation.

TL;DR: A valid llms.txt file is a markdown document at the site root. Only two elements are required: an H1 with the project or site name, and a single blockquote on the lines immediately following it. Everything else — H2 sections, link lists, and an explicit Optional section — is optional and used to direct LLMs at deeper content.

Where the Spec Comes From

llms.txt was proposed by Jeremy Howard and Answer.AI in September 2024 and is maintained at llmstxt.org with a reference implementation at github.com/AnswerDotAI/llms-txt. The format is intentionally minimal so it stays readable by humans and parseable by simple programs.

This page is a field reference. For a conceptual overview, see the llms.txt overview. For a working starter, see the llms.txt template. For the step-by-step rollout, see How to Create llms.txt.

Required Fields

A conforming llms.txt file must contain, in order:

Field	Format	Required	Notes
Project name	# Name (H1)	Yes	Single H1 at the top of the file. Identifies the site, product, or project.
Summary	> One or more lines. (blockquote)	Yes	Appears immediately after the H1. Should fit in one or two sentences.

If either of these is missing, downstream parsers may reject the file or fall back to heuristic extraction.

Optional Pre-Section Content

After the required H1 and blockquote, a file may contain free-form markdown (paragraphs, lists, blockquotes) before the first H2. Use this for short context that does not fit in the summary, such as scope, intended audience, or licensing notes. Keep it brief; LLMs treat the top of the file as highest-signal content.

Optional H2 Sections

The rest of the file is a series of optional H2 sections. Each H2 contains zero or more markdown link list items in this exact form:

Section Name

Title: optional notes about the page

Common section names in the wild:

Section	Typical content
## Docs	Primary documentation entry points (getting started, guides, API).
## Examples	Code samples, tutorials, integration walk-throughs.
## API Reference	Endpoint or SDK references.
## Policies	Terms, licensing, attribution notes.
## About	Two to three sentences of organizational context.
## Optional	Lower-priority links — see semantics below.

Section names are not enumerated by the spec; they are conventions. The spec only requires that each H2 section, if present, contains a markdown link list.

The Optional Section

A section literally titled ## Optional has special meaning: parsers and LLM agents may skip its contents when context is constrained. Use it for nice-to-have references such as changelogs, blog archives, or marketing pages. Treat anything outside ## Optional as in-scope by default.

Link List Item Format

Each link list item follows the pattern - Title: notes. Notes are optional but encouraged because they help LLMs decide which links to follow. Keep titles unique within a section. URLs should be absolute (https://example.com/docs/foo) so parsers do not need a base URL.

Companion Files

The llms.txt proposal includes two companion conventions:

File / Convention	Purpose	Source
*.md page mirrors	Append .md (or index.html.md for index URLs) to any page URL to serve a clean markdown version of that page for LLMs.	Answer.AI proposal
llms-full.txt	A single file that inlines the full markdown of the most important documentation, designed for one-shot context loading. Originated at Mintlify with Anthropic and was folded into the official proposal.	llmstxt.org, Mintlify

Use llms.txt as a navigation index, *.md mirrors so individual pages can be fetched cleanly, and llms-full.txt when context size and freshness allow inlining.

Conformance Checklist

Before publishing, verify:

File is served at /llms.txt over HTTPS with Content-Type: text/plain or text/markdown.
Exactly one H1 at the top.
A blockquote summary immediately after the H1.
All H2 sections, if present, contain only link list items in the canonical form.
All URLs are absolute and resolve.
File is small enough to be useful as a navigation index (commonly under 10 KB; larger files often belong in llms-full.txt).
For each linked URL, a .md mirror is available where practical.

Common Mistakes

Using multiple H1s. The spec defines exactly one.
Writing a long opening paragraph instead of a single blockquote summary.
Mixing prose into H2 sections. Keep them link lists; put prose before the first H2.
Linking to HTML-only pages without a .md mirror. LLMs benefit from the markdown variant.
Treating llms.txt as access control. It is a discovery aid, not a policy file; use robots.txt for crawl directives.

FAQ

Q: Which fields are strictly required in llms.txt?

Only two: a single H1 with the project or site name, and a blockquote summary on the lines immediately after it. Everything else is optional.

Q: Can I add custom H2 sections?

Yes. The spec does not enumerate section names. Use any H2 names that match how you organize content, but keep each section's body to a markdown link list.

Q: What does the ## Optional section mean?

It marks lower-priority links that LLMs may skip when context is constrained. Use it for nice-to-have references rather than core documentation.

Q: What is the difference between llms.txt and llms-full.txt?

llms.txt is a small navigation index that links out to detailed pages. llms-full.txt inlines the full markdown of the most important pages into a single file for one-shot context loading. Many sites publish both.

Q: Do I still need page-level .md mirrors?

The proposal recommends them. They let LLMs fetch a clean markdown version of any page by appending .md to its URL, which is more reliable than re-parsing HTML.