llms.txt: what it does and who reads it

What the file contains

Jeremy Howard proposed the format on September 3, 2024. Neither the W3C nor the IETF, the bodies that publish web standards, backs it: it is a convention that each site adopts or skips.

The file is written in Markdown, plain text that reads the same for a person and for a program. Only the title is required. The rest follows in this order:

• a title, the name of the site or project;
• a short summary, set as a quote;
• if needed, a few paragraphs of detail;
• headed sections that list links, each followed by a brief description;
• by convention, an "Optional" section that an agent can skip when it runs short of space.

The proposal is aimed first at agents that read a site at the moment they answer, especially coding assistants looking up documentation. It says nothing about ranking in a search engine.

A real file: the one on this site

This site publishes its own llms.txt. It is one summary paragraph and a list of links. The summary says who sells (a French company, with its SIREN registration number), what the audit covers, how pages are examined, what is delivered and when payment is collected. The links go to the offer, the methodology, pricing, the FAQ and the sample reports, each with a label that says what the page holds. The file quotes no price: it points to the pricing page, which is the reference.

The site exists in French and English. The format only allows one file at the root, so ours opens with the English version, then adds a "Français" section with the addresses of the other language.

Our first file was written by hand. It soon referred to "the method note" while the site said "the methodology note". Nobody reads the file on screen, so nobody noticed. It is now generated from the same text as the pages, and changes with them.

Who reads it, who ignores it

Google Search ignores it. Its guide to optimizing for generative AI features says you don't need machine-readable files or AI text files to appear in Google Search, including AI Overviews and AI Mode, "as Google Search itself doesn't use them". Creating one "will neither harm nor help" your visibility or rankings. Google added this note on June 15, 2026.

The same guide says it is completely fine to keep the file for other services that use it. It forbids nothing: it says Search does not take it into account.

Chrome, on the other hand, checks it. The audit tool built into the browser has an experimental category for AI agents, and its documentation calls llms.txt an "emerging convention". A missing file is marked not applicable there, since providing one is optional. Only a server error gets flagged.

For other AI assistants, we found no official documentation that says whether they read the file to build their answers, or how.

What matters more for being read by AI

Google's guide also explains that its generative AI features draw on the Search index. A page only shows up there if it is indexed and eligible to appear with a snippet. What helps therefore comes from the pages themselves, not from a separate file:

• content that crawlers can reach, and that stays readable when JavaScript does not run;
• clear rules for crawlers in your robots.txt file, including those run by AI companies;
• pages organized in sections, with headings that say what each section holds;
• content that brings a point of view or first-hand experience, not a summary of what already exists;
• pages that say who wrote them and who runs the site, what Google calls E-E-A-T.

Google adds that structured data is not required for its AI features, and that no special markup is needed. It remains useful for rich results.

What the audit checks

The audit looks for llms.txt at the root of the site, along with its variants llms-full.txt and ai.txt. A successful response is not enough. Many sites return their home page for any address, and an llms.txt that answers with a page of the site does not exist. The audit therefore compares the response with those of addresses that cannot exist, and checks its type and size. Only a real text file, distinct from the fallback page, counts as published.

The audit also records what your robots.txt tells AI companies' crawlers: allowed, blocked, or not mentioned. That is an editorial choice, and the report describes it without scoring it.

A missing file is not a fault: no rule in our scoring penalizes it. When present, it counts among the signals of readability by AI agents, with little weight. What carries this pillar is the pages: what an automated system can read there, extract and attribute to you.

What this check doesn't tell you

By Quentin Mathis, Z29K · updated September 28, 2026