The proposed llms.txt standard shipped its first revision on August 10, 2026 — nearly two years after launch. V2 keeps the curated root file but adds a second URL pattern for Markdown pages and two formal link relations (rel="alternate" and rel="describedby") so agents can reliably discover the clean-text version of a page.
2026-08-18 · by Moe Ameen
Jeremy Howard of Answer.AI — the developer who proposed llms.txt in September 2024 — published version 2 of the specification on August 10, 2026. It is the first revision since launch, and Howard framed it as an update "based on two years of adoption." The core idea is unchanged: a site publishes a curated Markdown file at `/llms.txt` that gives a language model or AI agent a clean, linkable map of its most important pages, plus optional Markdown versions of individual pages so the agent reads structured text instead of scraping cluttered HTML.
The gap V2 closes is discovery. The original spec pointed agents at pages and suggested publishing a Markdown copy at the same URL with `.md` appended (`/docs/page.html.md`), but nothing told an agent where that Markdown version actually lived or how to find it. V2 adds a second, cleaner URL pattern — replacing the extension outright (`/docs/page.md`), with `index.md` for directory paths — and, more importantly, two formal link relations so agents can locate the files without guessing. `rel="alternate" type="text/markdown"` points from an HTML page to its Markdown twin; `rel="describedby"` points to the llms.txt file that covers it. Both can be delivered as HTML `<link>` elements or as an HTTP `Link:` response header, so a site can adopt them without touching page markup.
Adoption context matters for reading this update. Howard says thousands of sites now publish an llms.txt file; documentation platform Mintlify generates one for every site it hosts, and Anthropic, OpenAI, and Google's Gemini team publish llms.txt files for their own developer docs. At the same time, the standard remains proposed, not official: Google Search has repeatedly said supporting llms.txt is "not on its agenda," even as Chrome's Lighthouse tool added an "Agentic Browsing" check that tests whether a site ships the file. Verify the current spec text and link-relation syntax on llmstxt.org, which is the primary source.
The honest read on llms.txt V2: it is plumbing. A better map from an AI agent to your clean Markdown is genuinely useful, but a map is worthless if it points at pages nobody would cite. The work that decides whether an agent quotes you is the same work it has always been — publishing original, well-structured, experience-rich pages, and enough of them, across the surfaces where agents and audiences actually look. The file is a five-minute add; the content library it describes is the year of work.
That library is what [Kompozy](/) produces. Point it at one source — a talk, a podcast, a long post — and it generates a full spread including [Blog Articles](/glossary/content-repurposing) and Email Newsletters written in one voice through your [Persona Brief](/glossary/persona-brief), with a banned-word filter that strips the generic AI-tell phrasing agents and readers both discount. Those blog pages are exactly the clean, structured, Markdown-friendly assets an llms.txt file is meant to point at, and Kompozy finishes them as a draft you add your own examples and point of view to — substance an agent has a reason to cite, not a doorway page. From the same run it also builds the social spread — captioned [Persona Shorts](/glossary/persona-shorts), brand-exact carousels, quote cards, text posts — and schedules the lot across the eight social platforms plus blog and email on [Autopilot](/glossary/autopilot). Ship the llms.txt file for the agents that read your site; use Kompozy to make sure there is something on the other end of every link worth reading. See [Generative Engine Optimization](/glossary/generative-engine-optimization) for the wider playbook this fits into.
V2, published August 10, 2026, keeps the curated /llms.txt root file but adds a second URL pattern for Markdown pages (replacing the extension, e.g. /page.md, or /index.md for directories) and two formal link relations so AI agents can discover those files: rel="alternate" type="text/markdown" points to a page's Markdown version, and rel="describedby" points to the llms.txt file. Both can be delivered as HTML <link> elements or an HTTP Link: response header.
Jeremy Howard of Answer.AI proposed llms.txt in September 2024. The V2 update, published August 10, 2026, is the first revision to the specification and is based, in Howard's words, on two years of adoption experience.
No. Google has repeatedly said supporting llms.txt is not on its agenda, so the file is not a ranking signal. It is an agent-readability convention that makes your content cleaner for AI agents and language models to read and cite. Chrome's Lighthouse tool does now include an Agentic Browsing check that tests whether a site ships the file, but that is a readability check, not a search-ranking one.
If you publish docs, a blog, or a knowledge base an AI agent might read, it is a low-effort add worth doing — thousands of sites, including Anthropic, OpenAI, and Google's Gemini docs, now publish one. Just remember it is a map, not the content: it only helps if the pages it points to are original and genuinely worth citing. The standard remains proposed, not official, so treat it as a readability improvement rather than a guaranteed traffic channel.