MMOOZZGG.com

LLM

What is llms.txt, and does your website need it?

7 min read3 sources

llms.txt is a short site map in Markdown for language models and AI agents: who you are and which pages to read first. Jeremy Howard proposed the format on 3 September 2024 at llmstxt.org. Search engines have made no official statement of support, but the file takes an hour to make and breaks nothing on your site.

On this page · 6 sections

What is llms.txt

llms.txt is a text file at an address like example.com/llms.txt. Inside is Markdown, a simple text markup where a hash sign marks a heading and a dash marks a list item: the site name, a short description and lists of links to the main pages.

The file’s job is to help an AI agent, a program built on a language model, understand a site fast. The authors of the standard explain the problem this way: an HTML page is wrapped in menus, ads and scripts, it is hard to turn into clean text, and a model’s context window is still too small for a whole site.

The file is meant mainly for the moment of an answer, not for model training. Example: you ask an assistant to study a library’s documentation, and the agent first reads llms.txt and then opens only the pages it needs.

File format: what is required and what is optional

Only a level-one heading (H1, a Markdown line with one hash sign) with the name of the project or site is required. The rest is optional, but in a recommended order.

1. A quote, a line starting with “>”: who you are and what you do, in one or two sentences.

2. Paragraphs and lists without headings: details that help to understand the rest of the file, such as how the site sections are organized.

3. Level-two sections (H2, a line with two hash signs) with lists of links. Each list line is a link in the form [name](address), then a colon and a short note about the page.

An example of a list line: “- [AI solution map](https://mmoozzgg.com/ai-roadmap): which company tasks to give AI and what effect it brings”. The note is short, but it tells the agent whether to open the page.

4. A section titled Optional. It holds secondary links, and an agent skips them when it needs a shorter context.

The file goes in the site root or in a subfolder, where it describes only the pages of that subfolder. Separately, the standard advises serving a clean Markdown copy of each page at the same address with .md added, and index.html.md for addresses with no file name.

Example: the file of our Russian-language site

The llms.txt file of our Russian-language site is built from the list of the site’s published pages, so a new page gets into it by itself. It starts with an H1 line with the brand name and a one-sentence quote about what the company does.

Then come five sections: “## Services”, “## Prices”, “## Sections”, “## AI glossary”, “## Contacts”. Each holds a list of links with a short note, as the format requires.

“Services” holds the service pages, “Prices” the pages where prices are published, and “Sections” the main pages of the site’s sections, including this section on language models. The prices themselves are not copied into the file: when a plan changes, the file will not contradict the site.

“AI glossary” links to the glossary’s main page and its categories, not to every term. This keeps the file short: the categories are enough for an agent to find the right word.

Only published pages go into the file; drafts do not. This rule is worth copying: a link in llms.txt to a page that does not exist is worse than no link.

How llms.txt differs from robots.txt and sitemap.xml

robots.txt tells automated programs which access to the site is allowed. llms.txt neither forbids nor allows anything: it explains what is on pages that are already open.

sitemap.xml lists all pages for indexing, often thousands of addresses. llms.txt gives a chosen overview: the main pages with notes, so the list fits in a model’s context window.

The format differs too. Programs read robots.txt and sitemap.xml by strict rules, while llms.txt is written in Markdown, which a model reads as normal text. The authors of the standard call it an addition to the two older files, not a replacement.

Who reads it: what is known as of the check date

As of 16 October 2026 there are no official statements of support. We checked Google Search Central help and the webmaster help of Russia’s largest search engine: neither mentions llms.txt as a search signal.

In its section on AI search features, Google says it directly: to appear in them, you do not need to create new machine-readable files, AI text files or markup. The name llms.txt is not used in that text, but the file fits the description.

At the same time, Lighthouse, the page audit tool in the Chrome browser, has an llms.txt check in its group for AI agents. It flags a server error when the file is requested, and if there is no file it reports “not applicable”: the documentation says the file is optional for now.

Who actually reads the file shows only in your server logs: requests to /llms.txt from assistants and agents. Look at the logs a month after you publish it, before you spend time on Markdown copies of pages.

If you want AI answers in search to quote your company, the usual things work: clear text, facts with sources, open access for crawlers. This is the work of GEO, optimization for the answers of generative models.

How to make the file in an hour

Step one. List the main pages: services, prices, contacts, the key sections and the articles people come to you for. Do not take everything: the file should take a minute to read.

Step two. Write an H1 heading with the company name and a one-sentence quote: what you do and for whom.

Step three. Sort the links into H2 sections and add a note after a colon to each one: what a person will find on the page. Move secondary links to the Optional section.

Step four. Save the file in UTF-8 as llms.txt in the site root and check that it opens at example.com/llms.txt as plain text, with no redirects and no password.

Step five. If your site is built from code, generate the file from the same list of pages as sitemap.xml: then the links will not go stale. This is how our site’s file is made.

Step six. Check the file through the eyes of a model: give an assistant the link to it and ask what the company does and where to see prices. If the answer is inaccurate, fix the quote and the link notes rather than adding new pages.

Do not repeat prices and terms in the file itself; link to the page with prices. More on how models read and understand text is in the section on LLMs.

Questions and answers

Does llms.txt affect search rankings?

There is no confirmation. As of 16 October 2026, search engines have made no official statement of llms.txt support, and Google writes that its AI search features need no special files. Make the file for AI agents and assistants, not for rankings.

Where do you put the file?

In the site root: example.com/llms.txt. The standard also allows a file in a subfolder; then it describes only the pages of that subfolder. Serve it as plain text in UTF-8 encoding, with no redirects and no login.

Do you need .md copies of all pages?

The standard advises it but does not require it: a clean Markdown copy of the page at the same address with .md added. For a small site, llms.txt itself with links to the normal pages is enough. Copies make sense for documentation and help pages that agents read in full.

Sources and fact-checking

Checked by the editors on 16 October 2026.

  1. llmstxt.org, 3 September 2024 (llmstxt.org)
  2. Google Search Central, 10 December 2025 (developers.google.com)
  3. Chrome for Developers, Lighthouse, 5 May 2026 (developer.chrome.com)

Found a mistake? Tell us

Publications

Contact us