llms.txt was first proposed back in September 2024. At that point it was a proposal, a theory. Since then, AI agents have become a reality, and fast.
Then, on 10 August 2026, the creator of llms.txt, Jeremy Howard, published a version 2 of the file at llmstxt.org. It is the first revision since the format launched.
What version 2 adds
Version 1 described one file at the root of a site. Version 2 keeps that file and adds the plumbing around it, so an agent can get from any page to a clean Markdown copy, and from any page to the llms.txt that covers it.
Markdown copies of pages
Pages agents might need should offer a clean Markdown version. Two address patterns are allowed: append .md to the existing filename, or replace the extension entirely.
rel="alternate" type="text/markdown"
A link relation from an HTML page to its Markdown copy, so an agent does not have to guess the address.
rel="describedby"
A link relation from a page to the llms.txt file that describes it. Large sites can point different sections at different files.
HTTP Link headers
Both relations can be sent as an HTTP Link response header instead of HTML, which works for non-HTML resources and can be added at the CDN without touching page code.
None of this changes what an llms.txt file looks like. A version 1 file is still valid. What changes is how discoverable it is, and how cheaply an agent can get a readable copy of the page it is standing on.
Why now
This sparked a new journey for llms.txt. The time has come for the file to be taken seriously. AI agents represent a systemic shift away from the novelty of a new file, and are the signal that it is here to stay.
The spec's own page now says the AI labs publish llms.txt files for their developer documentation: OpenAI, Anthropic and Google's Gemini team. Thousands of sites publish one, and documentation platforms such as Mintlify create one for every site they host.
Google's Lighthouse adding an llms.txt check in May 2026 was another great signal. Lighthouse 13.3 shipped an Agentic Browsing category that looks for the file at the domain root, with the explanation that without it "agents may spend more time crawling the site to understand its high-level structure and primary content." Google Search still says the file is not needed for its own AI features. Chrome's tooling checks for it anyway.
Sources
The /llms.txt file, v2, Jeremy Howard, published 3 September 2024, revised 10 August 2026. llms.txt v2 adds formal Markdown linking for AI agents, Search Engine Journal. Lighthouse llms.txt audit, Chrome for Developers.
What to do with yours
If you already publish an llms.txt, it still works. The upgrade is to give your key pages a Markdown copy and point at it with the new link relation, and to point each page back at the file that describes it. If you have not published one yet, the case for doing so just got stronger. Run the AI Readiness Audit to see whether AI systems can find and read yours today, and look up any domain in the registry to see what it publishes.
Questions
Is a version 1 llms.txt still valid?
Yes. Version 2 changes nothing about the file's format. It adds link relations and Markdown page copies around it, so an existing file keeps working and can be upgraded gradually.
What does rel="describedby" do?
It is a link relation from a page to the llms.txt that covers it, sent as an HTML link element or an HTTP Link header. It lets large sites point different sections at different files.
Does Google use llms.txt?
Google Search says the file is not needed for its AI features. Chrome's Lighthouse has checked for it since May 2026 in its Agentic Browsing category. Two products, two answers.