What is llms.txt? Definition, format and examples

Generate or validate an llms.txt

The format, required versus conventional

The spec, published at llmstxt.org by Jeremy Howard of Answer.AI, is deliberately small. In its own words, the H1 is the only required section. Everything else is convention or optional:

• An H1 with the site or project name. Required.
• A blockquote summary right after the H1. A strong convention, expected but not strictly mandated.
• Free-form detail paragraphs, with no headings, so they stay distinct from the sections below.
• H2 sections, each a markdown list of [name](url): optional description links.
• A reserved ## Optional section, a naming convention for links that can be skipped when a shorter context is needed.

The canonical example

This is the textbook structure, taken verbatim from llmstxt.org:

Does anyone actually read it? The honest answer

This is where most guides oversell. Publishing an llms.txt on your own site is common and easy; AI systems consuming and prioritising other sites' llms.txt files is a different claim, and the evidence says it is mostly not happening as of mid-2026.

• Google's John Mueller, quoted by two independent outlets, said no AI system currently uses llms.txt, comparing it to the long-ignored keywords meta tag.
• A 137,000-domain server-log analysis found 97% of llms.txt files received zero requests; among the requests that did occur, the top category was SEO audit tools, not AI systems.
• A 300,000-domain study found no measurable relationship between having an llms.txt and being cited by AI models.
• The one place it plausibly is used is developer tooling, where coding assistants fetch a library's docs. That is why Anthropic, Stripe, Vercel and Cloudflare publish theirs on developer docs, not marketing pages.

The fair summary: llms.txt is real, growing, and cheap to publish, and it may help coding agents read your docs. But as of mid-2026, no major consumer AI system is confirmed to fetch or prioritise third-party llms.txt files during general search. Publish it if you like; do not expect it to move your AI visibility.

How widely is it adopted?

Adoption figures vary by roughly two times depending on the dataset, so the honest presentation is a range, not a single number: about 5.6% to 10% of sampled domains carry one as of mid-2026, growing five to nine times year over year off a small base.

llms-full.txt is a separate convention

You will also see llms-full.txt, which inlines a site's full documentation rather than linking to it. It is not part of the original spec; it was popularised separately, notably by Mintlify. Adoption is inconsistent: Anthropic and Cloudflare ship one, while Stripe publishes llms.txt but not llms-full.txt. Treat it as a vendor convention layered on top of the proposal.

Real-world examples

Real files deviate from the strict spec in small ways, which is useful texture. llmstxt.org and Mintlify are textbook-compliant. Anthropic and Stripe skip the blockquote and open with a plain paragraph. Vercel places the blockquote before the H1, a real ordering deviation. Seeing these makes clear the format is used more loosely in practice than it reads on paper.