llms.txt version 2 explained: does your business need one?
A short file for language models, revised in August 2026. What it is, what it is not, and how to decide.
Am I Ready for Agents? editors · 23 September 2026 · Reviewed September 2026
llms.txt is a proposed markdown file at the root of a website that tells language models what the site is and which pages matter. Version 2, published in August 2026, lets any page point to its markdown version and to the llms.txt file that covers it. It is optional, it does not control crawling, and deciding not yet is a valid answer.
What is llms.txt?
A plain text file, written in markdown and placed at the root of a website, for example at /llms.txt. It gives a short description of the site and a list of the pages that matter most, with a line on each. It is written for language models and the agents that use them, which often work better from a short, clean summary than from a full site of HTML pages.
It is a proposal published at llmstxt.org, not a standard issued by a standards body such as the IETF or the W3C.
What changed in version 2?
Version 2 was published in August 2026. The main change for a business reader is that any page can now point to its own markdown version and to the llms.txt file that covers it, so an agent that arrives on a deep page, rather than the home page, can find the site's guide from there.
A version 1 file remains a valid starting point. If you already publish one, nothing breaks.
What does it not do?
- It does not control crawling. That is the job of robots.txt and your bot rules.
- It does not grant or refuse permission to use your content for AI training. robots.txt rules and content signals express that.
- It does not guarantee that any particular AI product reads it.
- It does not replace readable pages. Agents still need prices, policies and products as text on the pages themselves.
Who is it most useful for?
Sites where a short guide saves an agent real effort: publishers with large archives, businesses with many product categories or much documentation, and anyone whose most important pages are hard to reach from the home page. For a small site with a clear structure and a current sitemap, the gain is smaller.
If you are unsure, look at where visitors land today. If most arrive on deep pages rather than the home page, the page-level links added in version 2 are the part most likely to help.
How should you decide?
- Do we know which ten or twenty pages matter most to an agent acting for a customer?
- Who would write the descriptions, and who would keep them current when pages change?
- Do we want to offer markdown versions of key pages, which version 2 lets each page point to?
- Does publishing it match our position on AI use of our content?
If the answers are clear, publish a short file and review it whenever you review your sitemap. If they are not, not yet is a reasonable decision. The self-check counts a deliberate not yet the same as yes: question 4 asks only whether you have decided.
How does it relate to markdown for agents?
They work together. llms.txt points to pages; a markdown version gives an agent a clean copy of a page. Cloudflare's Markdown for Agents can produce markdown at the edge when a client asks for it, and when a site has not set its own policy it adds a content signal saying yes to AI training, search and AI input. Decide your content signals before switching it on. The lesson on making pages readable covers both.
Does this site have one?
Yes. Our own llms.txt lists the main pages of this site with a line on each. It is an example of the format, not a template to copy.