What is an llms.txt file?
An llms.txt file lives at the root of a website, the same address pattern as robots.txt, and it is written in plain Markdown. It gives an AI agent or a coding assistant a short, structured summary of what a site is and where to find the parts that matter: documentation, key pages, whatever the site owner decides is worth pointing at.
The format comes from a proposal, not a standard. Jeremy Howard published it on 3 September 2024, and it was updated to version 2 on 10 August 2026, three days before we wrote this (llmstxt.org). Nobody voted on it. No browser requires it. It is one developer's idea for a convention, and enough of the industry has picked it up that ignoring it completely would be careless.
Does Google use llms.txt?
Google's own search documentation answers this directly, if you read the actual sentence rather than a summary of it. It says: "You don't need to create new machine readable files, AI text files, or markup" to appear in AI Overviews or AI Mode (developers.google.com, last updated 10 December 2025). That sentence does not name llms.txt. It is a general statement about machine readable files, and an llms.txt file is one example of a machine readable file, so treat it as covering the category rather than as a specific verdict on this one format.
Google's browser team takes a different angle. Chrome's Lighthouse tool audits for an llms.txt file under its agentic browsing checks, and calls it "an emerging convention" (developer.chrome.com, page last updated 5 May 2026). A missing file does not fail the audit, which matters. Lighthouse marks it Not Applicable instead, "as providing the file is optional at the moment."
These two facts sit in different parts of Google, and they do not contradict each other. Google Search says it does not need files of that kind to decide what goes into its own AI features. Google's browser tooling checks for it anyway, because agents that are not Google Search, the coding assistants and browser-based agents that read a page directly, might use it. Both things are true at once.
What actually reads it, then?
Adoption is real, even without Google Search needing the file. Thousands of sites now publish an llms.txt file, several documentation platforms generate one automatically, and coding agents use them reliably when a project has one (llmstxt.org/changes.html). OpenAI, Anthropic and Gemini all publish one for their own developer docs.
"Coding agents use them reliably" describes developer tools reading developer documentation. It is not a statement about chat assistants ranking an ordinary business higher because that business has one. We have not seen evidence for that second claim, from Google or from anyone else.
Worth separating, too: a coding agent reading a project's llms.txt while it writes code against that project's API is a different act from a chat assistant deciding which businesses to mention in an answer about, say, plumbers in Leeds. The file format was built for the first case. Everyone hoping it also solves the second case is hoping, not reporting a result. If you want the wider picture of how AI systems decide what to cite at all, read our piece on how AI Overviews choose which brands to cite, or the broader explainer on what answer engine optimisation actually is.
What changed in version 2, published this month
Version 1 of llms.txt was a summary file and not much else. Version 2, published 10 August 2026, adds something more useful: discoverability (llmstxt.org/changes.html).
Two new link relations do the work. rel="alternate" type="text/markdown" points from a normal page to a Markdown version of that same page. rel="describedby" points from a page to the llms.txt file that covers it. Either can be set as an HTML <link> element in the page head, or as an HTTP Link: header on the response. An agent landing on any page can now follow a signpost to the right file, instead of guessing one exists at the root and hoping it does.
Version 2 also allows two naming patterns for a page's Markdown version, page.html.md and page.md, and it defines subpath scope: an llms.txt file covers the pages under its own path, and where more than one file could apply, the most specific one wins. We checked the two most authoritative explainers ranking for this term on 13 August 2026, from Ahrefs and Semrush. Neither mentions version 2, the link relations, or anything else on this list. Both were written against version 1.
llms.txt and robots.txt are not the same job
robots.txt is a fence. It tells crawlers what they may and may not touch, and it has done that job for as long as there have been crawlers.
llms.txt is a signpost. It does not block or permit anything. It points an agent toward a plain summary of a site, if that agent chooses to look for one. Confusing the two leads people to think an llms.txt file changes what can crawl their site. It does not touch access at all.
So does your site need one?
Split it by what kind of site you run.
If you run documentation or a software product with an API, the case is clear. Your users already include AI agents and coding assistants, the adoption pattern above backs that up, and an llms.txt file gives them a shortcut instead of forcing a scrape of the whole site.
If you run an ordinary business marketing site, the honest answer is smaller. Building one is a short job and harms nothing. Google's documentation says you do not need a file like it to appear in Google's AI features, and no one has published evidence that it moves rankings, AI citations or traffic anywhere else either. Note what that does not say: Google has not said the file hurts you, and nothing here suggests it does.
Do it anyway, once the bigger jobs on your site are done. An hour is a small enough cost that the case against bothering is weak, even with no upside proven. Just do not let anyone tell you it is the lever that gets you cited by ChatGPT, because that is not what the evidence in this article says. If someone is selling you an "AI visibility" package built mainly around adding this one file, they are overselling what the file does. Building your own is a short, well-documented job. Our piece on making your website readable to AI walks through it alongside the rest of the groundwork, so we will not repeat the tutorial here.
What ours looks like, and what it is still missing
We publish one. It is live at thesocialtarget.com/llms.txt, 74 lines of plain text, formatted to the spec: a title, a short summary, and linked sections pointing to the pages that matter on our site. We are not going to tell you it got us cited anywhere, because we have not measured that, and nobody has published a dataset that would let anyone else make that claim either. What we can tell you is that the file exists, matches the format, and took about an hour to write.
Here is the part most pieces on this topic will not show you. We checked our own site on 13 August 2026 against the version 2 discoverability signals described above. We found none of them. No Link: header. No rel="describedby". No rel="alternate" type="text/markdown" among the 18 <link> tags on a live blog page we checked.
Three days after version 2 shipped, our own file is still a version 1 file wearing a version 2 date. We will add the discoverability signals. Until then, this is the honest state of it: a file that exists and matches the letter of the spec, sitting on a site that has not yet done the newer part.
Adding this file is a small job with an unproven payoff, and that is fine. The harder question is which jobs on your site deserve your next hour, and that one is worth getting right. Tell us about your business through the intake form and we will give you a straight answer on where your effort belongs.
↳ Frequently asked
01How do you create an llms.txt file?
It is a plain text file, not a code project. Create a file called llms.txt, add an H1 with your site name, a one-line summary in a blockquote, then H2 sections linking to the pages you want an agent to find, and upload it to your site's root so it loads at yourdomain.com/llms.txt. We cover the full walkthrough, including where people get the formatting wrong, in our companion piece on making your website readable to AI.
02Do AI assistants actually read what's in an llms.txt file?
Coding agents and assistants that work with developer documentation read them reliably, according to the file format's own maintainers. Whether a general chat assistant like ChatGPT or Gemini reads your llms.txt file before answering a question about your business is not documented anywhere, by Google, by OpenAI, or by anyone else who has tested it publicly. For the fuller picture of what does influence whether a chat assistant cites you, see [how to get cited by ChatGPT](/blog/how-to-get-cited-by-chatgpt).
03Will adding an llms.txt file affect my Google ranking?
There is no evidence that it will, in either direction. Google's search documentation says you do not need machine readable files, AI text files or markup to appear in its AI features. That is a statement about what you need, not a promise that the file does nothing, and it does not name llms.txt specifically. Chrome's Lighthouse tool audits for one, but a missing file does not fail that audit. It is simply marked not applicable.
04What is the difference between llms.txt and a sitemap?
A sitemap is a list of URLs built for search engines to index. An llms.txt file is a short, human-readable summary written for an AI agent to understand what a site is about and where the useful pages sit. They do different jobs.
05Do I need to update my llms.txt file often?
Update it when your site's structure changes meaningfully: new product lines or pages you have retired. No published schedule governs how often anything fetches it, so there is no crawl cadence to keep pace with the way there is with a sitemap.