llms.txt

A proposed file that gives AI models a short, Markdown map of your site. The short answer, as of October 2026: Google says Search ignores it, no AI company we could find says its crawler reads it, and it can still help an agent that is already on your site. It costs little. Here is what it is, who reads it, and when we would write one.

What the file is

llms.txt was proposed by Jeremy Howard of Answer.AI in September 2024 (llmstxt.org). It is a Markdown file at the root of a site, /llms.txt, meant as a curated overview a language model can read in one go instead of crawling the whole site. The proposal asks for:

an H1 with the name of the site or project, the only part that is required;
a short summary in a blockquote;
sections under H2 headings, each a list of links, [name](url) with an optional note;
a section called Optional for links an agent can skip when it needs less context.

It also suggests serving a clean Markdown copy of each page at the same address with .md added. A small llms.txt looks like this:

# Example Shop

> Hand-made boots, shipped across Europe. Prices in euros.

## Help
- [Sizes](https://example.com/sizes.md): how our sizes compare
- [Returns](https://example.com/returns.md): 30 days, free

## Optional
- [Our story](https://example.com/about.md)

Who reads it

Google Search does not. Google’s guide to generative AI features in Search lists llms.txt under what you don’t need to do: Search “doesn’t use” such files, and creating one “will neither harm nor help your site’s visibility or rankings in Google Search, as Google Search ignores them” (Google Search Central, updated July 2026). Its page on AI features says the same without naming the file: no new “machine readable files, AI text files, or markup” are needed (AI features and your website).

AI assistants, as far as anyone has said. In June 2025 Google’s John Mueller wrote “FWIW no AI system currently uses llms.txt” (Bluesky). We read the crawler pages of OpenAI, Anthropic, Perplexity and Meta in October 2026: each explains its bots and robots.txt, and none says its crawler reads llms.txt. That is not proof that nobody does; it is the absence of anyone saying so.

Documentation sites publish it. OpenAI’s developer docs point to their own llms.txt as the index of the documentation. That is the use the proposal had in mind: a developer, or a coding assistant they are using, is sent to the docs and needs the map.

Chrome’s Lighthouse looks for it. Its experimental Agentic Browsing category checks for “a machine-readable summary at the domain root” (Chrome for Developers). Google calls the category experimental and based on proposed standards. So one Google product checks for a file another says it ignores; they answer different questions, search ranking and agents on a page.

llms.txt is not robots.txt

	robots.txt	llms.txt
What it is	Rules: which crawler may fetch which path	A map: which pages matter, in Markdown
Status	A standard, RFC 9309 (2022)	A proposal (2024)
Who reads it	Search and AI crawlers, by their operators’ own pages (a fetcher acting for a person may not follow it)	No crawler confirmed; some agents and tools
Can it keep a bot out?	Yes, as a request the bot is asked to honour	No: it grants and refuses nothing

The practical consequence: whether ChatGPT, Claude or Perplexity can find and cite your pages is decided by robots.txt and by your firewall, not by llms.txt. A robots.txt that keeps OAI-SearchBot out is not undone by a perfect llms.txt. That is what the Trakis audit checks as AI-005, and what the robots.txt validator shows, bot by bot; the list of AI crawlers has each one’s own documentation.

Should you write one?

For Google Search: no. Google says it ignores the file; it neither helps nor hurts.
For documentation, an API or a developer tool: it is worth an hour. People point coding assistants at docs, and a map with Markdown copies of the pages is what those tools can use.
For a shop, a blog or a local business: don’t expect anything from it. Spend the hour on robots.txt, on pages that load without JavaScript, and on what your pages actually say.
If you write one: list the pages that answer real questions, keep it short, and keep it current. A map to pages that moved is worse than no map.

The Trakis audit does not check for llms.txt and does not score it. Until a search engine or an AI assistant says it reads the file, a missing one is not a problem we would report.

Ours

We publish one at trakis.io/llms.txt: the checks, the tools, the guides, written by hand and tested so that every link in it works. It costs us nothing, Lighthouse looks for it, and an agent sent to Trakis finds the map. We expect no traffic from it.

Sources

Read on 14 October 2026.

Last changed 14 October 2026.