llms.txt: What It Is, How to Write One, and Whether It Actually Gets You Cited

Updated October 18, 2026: We corrected what the evidence shows about who uses llms.txt: Google Search ignores it, AI search crawlers request it rarely, and its use by coding agents rests mainly on the proposal author's account. We also added vendor documentation, sourced adoption and citation studies, a valid example file, the August 2026 v2 changes, corrected hosting steps and a robots.txt step that separates AI search crawlers from training crawlers. Why the GEO evidence changed.

llms.txt is a plain Markdown file you publish on your site to point AI tools to the pages worth reading and explain why. It is an optional, proposed convention, not an official web standard, and it is no shortcut to getting cited by ChatGPT or Google — but it takes under an hour to create and costs nothing to maintain. Just know the limits before you spend that hour: Google says Google Search ignores it, no major AI company documents reading it for its search or answer product, and log studies show that few requests reach these files. Its author says it is used most for software documentation that coding agents follow.

Here is what you need to know to create a valid one.

What llms.txt is — and what it is not

Jeremy Howard, co-founder of Answer.AI, published the original llms.txt proposal on September 3, 2024 (Answer.AI, September 3, 2024); he revised it as v2 in August 2026 (llmstxt.org changes). The current text lives at llmstxt.org. The proposal's premise: language models and AI agents working on a user's request can't take in a whole website. They have to parse HTML, get past navigation, ads and scripts, and guess which of your two hundred pages actually answers the question being asked. llms.txt offers them a curated map.

Think of it as a curated index — a document you write that says: here is what my site does, here are the pages that matter, here is a one-line explanation of what each page answers.

Three files, three completely different jobs:

robots.txt tells crawlers where they are allowed to go. sitemap.xml lists the destinations you want found. llms.txt is meant to tell AI tools which stops are actually worth making. robots.txt and sitemap.xml are the two that search engines rely on; llms.txt is optional, and it does not replace either.

llms.txt doesn't grant or block anything: it sets no permission to read, train on or cite your content, and it doesn't replace robots.txt. Access and training preferences go in robots.txt, using each AI company's crawler names; our robots.txt guide for AI crawlers lists them.

What the format looks like

The llmstxt.org spec (v2, August 2026) calls for plain Markdown in a fixed order: an optional byte-order mark, an H1 with the site or project name (the only required section), a blockquote summary, optional details without headings, and H2 sections listing links as [name](url): notes. Here is a valid example for a small business (example.com is a placeholder; use your own domain):

# Example Bakery

> Family-owned bakery with a bilingual (English/Spanish) website. This file lists the pages that answer common questions about our menu, hours and ordering policies.

Spanish-language versions of these pages are under /es/.

## Main pages

- [Menu](https://www.example.com/menu/): Full menu, updated monthly
- [Hours and location](https://www.example.com/visit/): Opening hours, address and parking
- [Ordering policy](https://www.example.com/orders/): Custom cakes, lead times and cancellations

## Resources

- [Cake care guide](https://www.example.com/blog/cake-care/): How to store and transport a custom cake

## Optional

- [Blog](https://www.example.com/blog/): Recipes and seasonal news

Only the H1 is required; everything else is optional, but whatever you include must follow this order. The H1 is your site name. The blockquote is your site summary — write it as if answering "what does this website do?" in one or two sentences. The plain paragraph after it can add context, but no headings. H2 sections group related pages. Each bullet is a Markdown link, optionally followed by a colon and a short, honest annotation. In v2, an H2 named "Optional" is a convention for secondary links an agent can skip when it needs a shorter context (llmstxt.org changes).

What makes a good annotation: state exactly what the page answers, not what it sells. "Step-by-step checklist for setting up Google Business Profile for a service-area business" is useful to an AI tool deciding whether to read your page. "Our comprehensive guide to dominating local search" is not.

About llms-full.txt: some documentation platforms also generate an llms-full.txt with the full text of your pages in one file (for example Mintlify); that file is a platform convention, not part of the llmstxt.org spec, which instead recommends offering a clean Markdown version of each page at the same URL plus .md. Both take more effort and mainly pay off on documentation-heavy sites — for most small businesses, a clean llms.txt is enough.

Where to host it

The usual place is https://www.example.com/llms.txt (with your domain), at the root of your domain, but the llmstxt.org spec (v2, August 2026) also allows a file at any subpath, such as /docs/llms.txt, which covers the pages under that path; a subdomain such as docs.example.com is a separate site and can have its own file at its root. For most small-business sites, the root is the right choice. The file must be public, with no login.

How to put it there on common platforms:

• WordPress: upload the file through FTP or your host's file manager to the folder where WordPress is installed (the one that holds WordPress's index.php), so it loads at /llms.txt on your domain; don't use wp-config.php as the landmark, because WordPress lets you move it one folder up. If WordPress sits in a subfolder but your site address is the domain root, put the file in the web root instead.
• Next.js: put the file at public/llms.txt in your project root; Next.js serves files in public from the base URL, so it will load at /llms.txt (Next.js docs).
• Webflow, Squarespace and Wix now have built-in llms.txt settings (as of September 2026): Webflow lets you upload the file in the site's SEO settings and serves it at your domain's root (Webflow, July 23, 2025); Squarespace 7.1 has an LLMS.txt tab under SEO/AI Visibility > SEO Settings (Squarespace Help, updated July 22, 2026); and Wix generates the file automatically on eligible sites, editable under SEO & GEO in the dashboard (Wix Help Center). Custom-code fields that inject HTML into pages don't create a file.
• Static site generators: put it in the root of the build output folder (same level as index.html).

Verify by loading the URL in a browser. You should see plain text — not HTML, not a file download prompt. If the browser downloads the file instead of showing it, check the server headers: serve it as text/plain (or text/markdown) with charset=utf-8 and without Content-Disposition: attachment, which tells browsers to download (MDN); the llmstxt.org spec itself does not set a MIME type, so this is a practical recommendation.

Does it actually work today?

Honest answer: as of September 2026, none of the major AI search or answer engines documents using it, the studies that measured citations found no clear link, and the use case its author points to, coding agents, rests mainly on his claim and limited log data.

Google Search ignores it. Google's AI optimization guide, last updated July 10, 2026, says creating llms.txt files "will neither harm nor help" your visibility in Google Search, "as Google Search ignores them"; Google added that note on June 15, 2026 (Search Central updates). That covers Google Search, including AI Overviews and AI Mode, so don't treat the file as a lever for either.

ChatGPT, Perplexity and Claude don't document reading it. As of October 2026, OpenAI, Anthropic and Perplexity document only robots.txt controls for their crawlers and say nothing about reading llms.txt (OpenAI; Anthropic; Perplexity). Their bots do fetch these files now and then, but rarely: in Ahrefs' May 2026 logs, GPTBot (a training crawler) and Claude-Code (a coding agent) were among the top AI fetchers, while OAI-SearchBot, PerplexityBot and Claude's search crawler combined made only a couple of hundred fetches across thousands of sites, and 97% of files got no requests at all (Ahrefs, June 15, 2026). Ahrefs, which sells SEO and analytics tools, studied the 137,210 domains in its own Web Analytics base, which skews technical, and notes that a fetch doesn't prove the file was read.

No clear link to citations. The published studies that measured citations found no clear link between having llms.txt and being cited by AI. Across about 300,000 domains, SE Ranking found "no correlation between AI citations and LLMs.txt", with citations measured as how often each domain appeared as a source across a sample of AI answers (SE Ranking, November 2025; the page now shows October 15, 2026). Trakkr (March 14, 2026) compared 37,894 domains that appeared at least twice in its AI-citation snapshots: sites with and without llms.txt averaged almost the same number of citations (6.8 vs 6.7; median 3 for both), while Trakkr's full-sample test found a statistically significant difference with a small effect size, and among the 5,000 most-cited domains the difference was not statistically significant (Trakkr, March 14, 2026). Neither study names the AI platforms it sampled, and Trakkr warns against reading the snapshot as proof of either a benefit or no effect. Both are observational vendor studies, so neither shows what adding the file would cause.

Adoption depends on who counts. In a November 2025 analysis of about 300,000 domains, SE Ranking found llms.txt on 10.13% of them (SE Ranking); a May 2026 Ahrefs count found it on 28% of the 137,210 domains in its Web Analytics base, a figure Ahrefs calls an upper bound (Ahrefs, June 15, 2026), so adoption depends heavily on the sample. Don't read either number as an opening: with none of the major AI engines documenting use of the file, having one gives you no proven edge.

Coding agents: the use its author points to. The llmstxt.org author says the files are "used most heavily for software documentation, where coding agents follow them," but that is his claim, not measured data, and Ahrefs' May 2026 logs found that AI bots never requested llms.txt files that didn't exist (Ahrefs, June 15, 2026), so don't assume an agent will find the file on its own. When an agent does read it, that is a different kind of visibility — not a consumer search result, but a developer tool reading your content. To help agents find it, Ahrefs advises linking to the file from your site or documentation, and v2 of the proposal recommends a rel="describedby" link pointing to it (llmstxt.org); both are advice, not evidence that agents will read the file.

llms.txt is an optional, proposed convention, not an official standard. Robots.txt followed a similar path: Martijn Koster proposed it in 1994, it worked as an informal convention for decades, and the IETF published it as a standard only in September 2022 (RFC 9309); as of September 2026, llms.txt is still a proposal "open for community input" (llmstxt.org) and no standards body such as the IETF or W3C has adopted it. It will not make up for weak content or an untrustworthy domain. If developers or AI coding agents use your site, it earns its hour of setup time; otherwise, spend that hour on your content.

The bar to publish llms.txt is low, but so is the payoff in search. The real question is not "should I?" — it is "who will read it?" If AI coding agents or developers use your site, write it for them and publish it. Either way, don't count on it for Google or AI citations, and revisit quarterly in case a major AI engine documents using it.

If you want to know where your business actually stands in AI answers, request a visibility assessment. Before any work begins, we reply with a fit recommendation, the scope and the fee.