Jeremy Howard, the Answer.AI developer behind llms.txt, shipped the first update to the spec since its 2024 launch. Llms.txt v2, released August 10, doesn’t change what the file is for. It fixes a specific gap: nothing in the original version told agents where a page’s Markdown copy actually lived.
What Llms.txt V2 Actually Changes
The original proposal covered two things: the llms.txt file itself, and Markdown versions of individual pages published at the same URL with .md appended. Llms.txt v2 keeps both but adds a second URL pattern for those Markdown pages. Sites can still append .md to the full filename, so /docs/tutorial.html becomes /docs/tutorial.html.md, or they can now replace the extension outright and serve it at /docs/tutorial.md. Anyone already running llms.txt only needs a minor edit to move to v2, not a rebuild.
Two Link Relations Tell Agents Where To Look
The bigger addition is a pair of formal link relations that let agents discover the Markdown version of a page without guessing at the URL pattern. A rel="alternate" type="text/markdown" attribute links a page directly to its Markdown counterpart. A rel="describedby" attribute points to the llms.txt file that covers it, since one llms.txt file can describe every page under a given path.
Sites can deliver both relations as standard HTML link elements in the page head, or as an HTTP Link response header. The header option is the more flexible of the two: it works on files that aren’t HTML at all, including the Markdown pages themselves, and it can be set at the server or CDN level without touching page templates or code.
The Gap Howard Was Trying To Close
Howard said the update reflects two years of real-world use. Thousands of sites now publish an llms.txt file, and hosting platforms like Mintlify generate one automatically for every site they run. Google’s Chrome team has gone as far as adding a check for the file to Lighthouse’s Agentic Browsing category, a feature we covered when it landed in May. Anthropic, OpenAI, and Google’s own Gemini team all publish llms.txt files for their developer documentation.
As adoption spread, the gap became obvious: the file told agents which pages existed, but nothing told them where the Markdown versions of those pages actually sat. Llms.txt v2 is the fix for that missing pointer, not a redesign of the format’s purpose.
Google Search Still Isn’t Using It
None of this changes where Google stands. Google’s Search team has said for well over a year that supporting llms.txt isn’t on its roadmap, and the company’s own AI optimization guidance states plainly that Search doesn’t read these files at all — maintaining one, the guide notes, “will neither harm nor help your site’s visibility or rankings.” That position has held even as Google staff weigh in on adjacent Markdown questions, including John Mueller’s recent caution about relying on Markdown for SEO purposes.
Chrome’s Lighthouse tool is a separate story. Its Agentic Browsing check tries to fetch a site’s llms.txt file and flags server errors; a missing file that 404s is scored N/A rather than as a failure, since the file is optional. Lighthouse treats this as a way to help browser agents map a site’s layout faster — but it doesn’t yet test for the new link relations or for Markdown-page discovery specifically, so v2’s additions sit ahead of the tooling that would actually verify them.
Who This Update Is Really For
The skepticism around llms.txt as a ranking lever isn’t new — it’s the same argument laid out in the case that a satirical cats.txt file exposed how thin the evidence for llms.txt’s SEO value really is. Llms.txt v2 doesn’t add anything to that side of the ledger. Sites maintaining the file specifically to influence AI Overviews or AI Mode citations still have no data, and no statement from Google, suggesting it works.
What v2 does serve is the audience that was already using llms.txt for something else: coding agents, IDE tools, and documentation platforms that read the file to speed up API lookups and pull in reference content. That group now has a clearer spec and a defined way to route from a page to its Markdown copy and back to the llms.txt file that describes it. It’s a similar dynamic to other structured-trust efforts aimed at agents rather than rankings, like Google’s own Open Knowledge Format and its follow-up trust-layer revision — genuinely useful machine-readable plumbing that isn’t a search visibility play.
Howard’s changes page documents every addition to the spec along with the reasoning behind it, and v2’s link relations are still open for feedback on GitHub. Sites adopting the syntax now should expect it may still shift before the format reaches a stable release.
Frequently asked questions
Do I need to migrate to llms.txt v2 right away?
No. Sites already running the original spec can adopt v2 with a minor update — adding the new link relations — rather than rebuilding their llms.txt setup from scratch.
Does llms.txt v2 change anything for Google rankings?
No. Google’s Search team has said support for llms.txt isn’t planned, and Google’s AI optimization guidance confirms Search doesn’t use the file for ranking, positive or negative.
What are the two ways to link a page to its Markdown version under v2?
A page can use a rel="alternate" type="text/markdown" HTML link element, or the equivalent can be delivered as an HTTP Link response header — the latter also works on non-HTML files and can be configured server-side without changing templates.
Who actually benefits from llms.txt v2 today?
Coding agents, IDE tools, and documentation platforms that already read llms.txt files to retrieve reference content and speed up lookups — not sites hoping for AI Overviews or AI Mode citation gains.