Call · 30 min
MarketingMattia Esposito10 November 202613 min read

llms.txt. What it really is, who reads it, and an example file annotated line by line.

llms.txt is sold as the way into ChatGPT, and Google says its Search ignores it. Here is the format explained properly, an annotated example and a checklist, in Word and PDF.

In brief

llms.txt is a Markdown text file, at the root of a site, that introduces the site to a language model: a heading with the name, a summary and links to the pages that matter. The proposal was published on llmstxt.org in 2024 and updated to version 2 in August 2026.

Google Search doesn’t use it. In its May 2026 guide Google says creating one neither helps nor hurts. OpenAI, Anthropic and Perplexity publish one for their documentation, but don’t state that they read it to choose sources for answers.

It makes sense if you have documentation, technical products or agents reading your site. For everyone else it takes half an hour and replaces nothing.

The example file and the checklist can be downloaded below, in Word and PDF, with no email required.

Written by: Mattia Esposito, AI and marketing consultant, founder of Itria AI. From Puglia, based in Turin since 2023, he works remotely in Italy and abroad. LinkedIn profile.

The example llms.txt file, to download

A complete llms.txt file for a made-up company, Conserve Esempio, with a note on every line and a ten-point checklist to publish and check it. In Word you change names and addresses, the PDF is printed and ticked off. The example follows version 2 of the proposal on llmstxt.org.

The example llms.txt file annotated line by line, with the checklist to publish and check it
FileContentsLink
Wordto adapt, in Italian

The example file, what each line does, the ten-point checklist.

llms-txt-modello.docx

PDFto print, in Italian

The same template, with the checklist to tick off with a pen.

llms-txt-modello.pdf

Example llms.txt, line by line

# Conserve Esempio
The H1 heading with the name. For the proposal it is the only required element, and it must come first.

> Conserve Esempio has made tomato passata and sauces in the province of Taranto since 1987 and sells to shops, restaurants and importers in Italy and abroad.
The summary, in a blockquote. Who you are, what you do, for whom and where, in one or two sentences.

Minimum export order: one pallet. Price list and spec sheets on request by email.
The context: free text, no headings. Put here what is needed to understand the rest.

## Products
An H2 section: it opens a list of links.

- [Catalogue](https://www.tuaazienda.it/catalogo): the twelve products, sizes and packs
One entry: the link in brackets is required, the note after the colon isn’t.

- [Spec sheets](https://www.tuaazienda.it/schede-tecniche): ingredients, allergens, shelf life
Better a page with the data in plain text than a PDF to download.

## Company and contacts
Name, address, VAT number and contacts written exactly as on the site.

- [Contacts](https://www.tuaazienda.it/contatti): email, phone, company name, VAT number
The page an agent opens when someone asks how to reach you.

## Optional
By convention, secondary material an agent can skip when it needs to keep things short.

- [Recipes](https://www.tuaazienda.it/ricette): serving ideas for consumers
Useful, not essential: that’s why it sits here.

On 7 September I opened the websites of 142 Italian companies that also sell abroad and counted those with the file. This is how I put it in that day’s post: «Twenty-eight out of a hundred and forty-three have the file that explains the company to an assistant, and in twenty-four a plugin wrote it.» 19.6% had it, and in 86% of cases nobody had written it by hand.

I had already written mine, along with the structured data, and it wasn’t enough. On 7 August ChatGPT, asked who had founded itria.io, said it couldn’t find me. In my post of 21 September I put it like this: «I had written the llms.txt, the schema, all the nice things. As far as AI was concerned, I didn’t exist.»

The file is still there and I keep it up to date, because it costs little. In the following weeks I wrote 78 new pages, many with a file to download, and the site’s impressions went from 24 to 1,222 in 28 days. This page separates what llms.txt does from what it is credited with.

What llms.txt is

llms.txt is a Markdown file published at a site’s /llms.txt address, designed so a language model can quickly understand what the site is about and where the useful pages are. The proposal came out on llmstxt.org in September 2024, signed by Jeremy Howard of Answer.AI. Version 2, from August 2026, is the current one.

The problem behind it is practical. A web page carries menus, scripts, ads and footers, and a model answering a question about a site has little room to read. A short file written for it gives the name, a summary and an index. The proposal frames it for on-demand use, while an agent is helping someone, more than for training.

It shares only one idea with robots.txt and sitemap.xml: a fixed name at the root. robots.txt tells crawlers where they may go; the sitemap lists every page for search engines; llms.txt picks the pages that matter and explains them in words. None of the three replaces the others, and llms.txt neither grants nor removes permissions.

The llms.txt format, section by section

Version 2 of the proposal sets five parts in a fixed order, and only one is required: the H1 heading with the name. The specification page says so in so many words: «This is the only required section». Everything else is optional, but the rest is where the file becomes useful.

PartContentsRequired
BOM

The encoding marker at the start of the file, allowed but not required.

No

# Name

An H1 with the name of the project or site.

Yes, the only one

> Summary

A short blockquote with the key information needed to understand the rest.

No

Details

Paragraphs or lists of any kind, as long as they have no headings.

No

## Sections

H2 headings, each with a list of links in the form [name](url): note.

No

## Optional

By convention, secondary links an agent can skip.

No

Version 2 adds three things. The file also works in a subpath, such as /docs/llms.txt, and covers the pages below it. It proposes clean Markdown versions of pages at the same address with .md on the end. It suggests rel="alternate" and rel="describedby" links, also as an HTTP header, so agents can find those versions on their own.

Is llms.txt a standard?

No: it is a proposal published on a website, with no body approving it, unlike the IETF’s RFCs. Chrome for Developers, in the Lighthouse documentation, calls it «an emerging convention». Plugin vendors often call it a «standard», and the difference matters: nobody is obliged to read it.

A convention, though, spreads through adoption. The v2 page describes thousands of sites publishing the file, documentation platforms generating it automatically and AI labs using it for their own developer docs. A rule many follow without anyone having made it binding.

Google and llms.txt: Search ignores it

Google Search doesn’t use llms.txt, not even for AI Overviews and AI Mode. In its 15 May 2026 guide to generative AI features, updated on 10 July, Google says creating one is fine for other services but neither helps nor harms visibility, because «Google Search ignores them».

Among the tactics to drop, the same guide lists creating unnecessary AI text files, and mentions llms.txt in brackets. The page on AI features says the same in other words: you don’t need new machine-readable files to appear. According to the trade press, Google’s John Mueller and Gary Illyes have spoken along the same lines.

Another part of Google does look at it, though. Lighthouse, Chrome’s tool for measuring pages, has an experimental category for in-browser agents with an llms.txt audit. It flags an error if the server can’t deliver the file; if the file is missing, the audit is marked not applicable, because for now it is optional. It concerns agents, not rankings.

Two paths: an agent that has to answer opens llms.txt and from there the pages listed; Google Search goes through the crawler and the index and ignores the file A questionto the agent or IDE /llms.txtname, summary, links The pagesalso as .md The answerwith the right context A searchon Google Googlebotrobots.txt, sitemap The indexresults and AI Overviews llms.txtignored by Search use stated by the proposal (llmstxt.org, v2)use stated by Google Search (guide of 15 May 2026)
Top: the path llms.txt was proposed for, an agent that has to answer about a site. Bottom: Google Search, which according to its documentation doesn’t use it.

Do ChatGPT, Claude and Perplexity read llms.txt?

None of the three says so in the public documents read for this page, and none denies it. What can be checked is something else: OpenAI, Anthropic and Perplexity publish an llms.txt for their developer documentation. On 10 November 2026 every address in the table returned a 200 code.

WhoWhat it states or publishesWhere
Google Search

Ignores the file; creating one neither helps nor hurts.

Guide of 15 May 2026

Chrome Lighthouse

Checks it for in-browser agents; optional.

Agentic browsing audit

OpenAI

Publishes one for its documentation, with an llms-full.txt.

developers.openai.com

Anthropic

Publishes one: about 99 KB, plus an llms-full.txt of about 688 KB.

platform.claude.com

Perplexity

Publishes one: about 43 KB, plus an llms-full.txt of about 3.9 MB.

docs.perplexity.ai

Google Gemini API

Publishes one for the API documentation, about 68 KB.

aistudio.google.com

Stripe

Publishes one with instructions for agents at the top; no llms-full.txt.

docs.stripe.com

Publishing one for your own docs helps developers working with a coding assistant: the agent opens the file and finds the right API page. It isn’t proof that the same providers’ search crawlers read it when choosing sources. For that, the rules gathered on the page about how to show up in ChatGPT apply.

llms.txt example: three real files to read

The quickest way to understand the format is to open three published files: the proposal’s, Anthropic’s and Stripe’s. The example on the llmstxt.org page, for the FastHTML library, has the full structure: heading, summary, important notes, a Docs section and an Optional section, with links to the .md versions of the pages.

Anthropic’s file is a long index of the documentation, with a note for every page. Stripe’s file opens with instructions for agents, such as checking the latest package version. OpenAI’s file is short, about 6 KB, and points to one file per product.

itria.io publishes one too, at itria.io/llms.txt. It is 22,876 bytes, with 16 H2 sections and 40 links; it opens with the name and a summary in a blockquote, carries the date of the last update and points to the extended version, llms-full.txt. It is written for a services company, the least documented case.

What I learned from my llms.txt

The file didn’t get me found. In August llms.txt and structured data were already there, and ChatGPT couldn’t find the founder of itria.io. What changed things were the pages, starting with the mistakes told on the page about the email address AI assistants can’t see.

I keep it for agents. A summary written by me, with VAT number, address and cases, saves an agent from rebuilding the company from scattered pages. It carries a date, and I change it when the site changes.

On other sites it is almost always automatic. Of 28 Italian company sites that had the file, 24 got it from a plugin.

llms.txt generator: WordPress, Shopify and the rest

On many platforms the file generates itself, and llmstxt.org lists Mintlify, GitBook, Yoast SEO, AIOSEO and Wix. A generator saves time on the list of pages; the summary, the part that says who you are, has to be written or corrected by hand. The file itself costs nothing: it’s text.

PlatformHow you get the fileWhat to check
WordPress

A plugin such as Yoast SEO or AIOSEO generates it; or you upload it by hand to the root folder.

The generated summary

Shopify

On two Shopify stores checked on 10 November 2026, /llms.txt already returned a file of instructions for shopping agents.

What your own /llms.txt returns

Wix

Generates it for every site, according to the llmstxt.org list.

The text and links

Mintlify, GitBook

Serve it automatically for documentation sites.

Excluded pages

Static site

You write the file and upload it to the root, like robots.txt.

200 code and encoding

Online generators

They read the sitemap and suggest a draft to copy.

Everything, line by line

On WordPress a plugin is the quickest route, and the 7 September count says it is also the most common. The risk is a file listing a hundred pages without saying what the company does. On Shopify, open your own /llms.txt address first: there may already be a file.

llms.txt checker: how to check it

An llms.txt works if the address returns a 200 code, the server sends it as UTF-8 text and the first line is an H1 heading. The check takes a minute: type yoursite.com/llms.txt into the browser, then, from a terminal, curl -I followed by the address, and read the code and the content-type line.

A 404 means the file isn’t there, and for Lighthouse that’s fine: the audit is marked not applicable. A server error, a 500 or a 403 from the firewall, Lighthouse flags as an error: the file exists but doesn’t arrive. Garbled accented letters mean the wrong encoding, fixed by serving the file as UTF-8.

Then the content. Every link must open a real page, the summary must say the same things as the site and the company details must match the footer. The checklist in the template lines up ten checks, from the file name to the last line, with where to look for each.

llms-full.txt and .md pages

llms-full.txt is a second file with the entire documentation in one text, and the v2 proposal page doesn’t mention it. The convention was spread by platforms such as Mintlify and GitBook and by the providers themselves: Anthropic, OpenAI and Perplexity publish one, Stripe doesn’t. Perplexity’s file is over 3.9 MB.

For a services company an llms-full.txt only makes sense if the site has many pages of data an agent might need to read all at once, such as product sheets or terms of sale. More useful, in v2, is the .md version of the important pages: the same content without menus and scripts, at the page address with .md on the end.

When to write llms.txt and when not to

It is worth writing when someone will read it: technical documentation, APIs, products with long spec sheets, sites that agents and coding assistants open on request. For a ten-page brochure site it does no harm and brings no visits: half an hour’s work, provided everything else is already in place.

First come the things Google and the others document: search crawlers let in, the site in Google’s and Bing’s indexes, text in the HTML, facts with their source. You’ll find them on the pages about how to get a website indexed on Google, how to appear in AI Overviews and generative engine optimization.

The rules don’t change between Italy, the European Union, the United Kingdom and the United States: llms.txt is a voluntary file, with no legal obligations, and can be written in any language. The technical terms, from language model to crawler, are explained in the language model glossary. How to build a real company’s presence on search engines and assistants is on the page about presence on search engines and AI assistants.

Questions and answers

What is the llms.txt file?

A Markdown text file, placed at the root of a site, that introduces the site to a language model or an AI agent: the name, a summary and links to the pages that matter. The proposal was published on llmstxt.org in September 2024 and updated to version 2 in August 2026, not a standard from an official body.

Does Google use llms.txt?

Google Search doesn’t. In its May 2026 guide to generative AI features Google says Search ignores these files: creating them neither helps nor hurts rankings, not even in AI Overviews and AI Mode.

Lighthouse, Chrome’s tool, does check it among its audits for in-browser agents, where the file is optional.

Does it help you show up in ChatGPT?

No official document read says so. OpenAI, Anthropic and Perplexity publish llms.txt for their developer documentation, but don’t state that their assistants use it to choose sources. To show up, what counts is letting search crawlers in, the index, and pages with checkable facts.

How do you create llms.txt on WordPress?

With a plugin that generates it, such as Yoast SEO or AIOSEO, listed among the integrations on llmstxt.org, or by writing it by hand and uploading it to the site’s root folder. Open and correct the generated text: it is often just a list of pages, with no summary written for customers.

How do you open and check an llms.txt file?

Open it in the browser by typing the site address followed by /llms.txt, or in any text editor. To check it: it must return a 200 code, be served as UTF-8 text, start with an H1 heading and have links that open real pages.

Notes on sources

  1. llmstxt.org, The /llms.txt file, version 2 of August 2026, and the changes page: the format, the quotation, the FastHTML example, the integrations. Read on 10 November 2026.
  2. Google Search Central, Optimizing your website for generative AI features on Google Search, 15 May 2026, updated 10 July 2026, and AI features and your website, updated 10 December 2025, content licensed under Creative Commons Attribution 4.0.
  3. Chrome for Developers, Lighthouse llms.txt audit, agentic browsing category. Read on 10 November 2026.
  4. The llms.txt files of Anthropic, OpenAI, Perplexity, Gemini API and Stripe, opened on 10 November 2026: response codes and sizes are as of that day.
  5. Shopify’s behaviour comes from two stores opened on 10 November 2026; it isn’t an official platform announcement. The positions of John Mueller and Gary Illyes are as reported by the trade press, not from a Google document.
  6. The count of 28 sites out of 143 comes from the post of 7 September 2026; the figures for itria.io’s file from the published file, read on 10 November 2026.
·Who does it for you

A text file describes the site. Getting the company chosen over its competitors takes something else. Positioning is built starting from the market.

Mattia Esposito, who wrote this guide, also works with Itria on positioning: in-depth market analysis to find where competitors are exposed and take that ground. He did it on his own site first. In August, asked who had founded itria.io, ChatGPT couldn’t find him. From 11 October to 7 November 2026 itria.io pages appeared 13,803 times in Google and Bing results and 3,818 times in AI answers: Google’s AI Overviews and AI Mode, and Microsoft Copilot.