A small Markdown file at the root of your site promises to help AI models understand your business. Here is what it actually does today, what it does not, and why we publish one anyway.
llms.txt in two sentences
llms.txt is a plain Markdown file at the root of a website that gives AI models a short, curated map of who you are and which pages matter. It is cheap to publish and harmless to have, but on the evidence available in 2026 it does not improve your Google rankings or measurably increase how often AI assistants cite you, so treat it as housekeeping rather than a growth lever.
That is less exciting than the way the file is often sold. It is also the honest starting point for deciding whether it deserves an hour of your team's time, and what that hour should produce.
Where the file came from
The format was proposed in September 2024 by Jeremy Howard, co-founder of Answer.AI. His argument was narrow and practical: a language model cannot hold a whole website in its context window, and converting cluttered HTML into clean text is imprecise, so a site owner can help by publishing a short summary and a list of the pages worth reading.
The specification is deliberately thin. A single H1 heading with the name of the site is the only required element. Below it sits an optional one-paragraph summary in a blockquote, optional free-form notes, and sections of links, each with a one-line description. A section headed Optional marks material a model can skip when it is short on space. A separate convention, llms-full.txt, bundles the full text of the site into one file; it grew up around documentation platforms and is not part of the original proposal. Neither file has any standards-body recognition. It is a community convention, not a protocol.
What the evidence says so far
Google has been unusually direct. In July 2025 Gary Illyes of Google Search said Google does not support llms.txt and is not planning to, and John Mueller compared the idea to the keywords meta tag, the self-declared signal search engines stopped trusting precisely because site owners wrote it about themselves. That logic is hard to argue with: a file in which every business describes itself as the best answer cannot help a system choose between businesses.
The citation data points the same way. According to an SE Ranking analysis of roughly 300,000 domains, reported by Search Engine Journal in November 2025, about one in ten sites had published the file, and there was no measurable link between having it and how often a domain was cited by AI assistants; removing the llms.txt variable from their prediction model actually made it slightly more accurate. None of the major AI providers has publicly committed to reading the file as a ranking or retrieval signal.
Where the file does have readers is narrower. Developer platforms such as Stripe, Cloudflare and Anthropic publish one because coding assistants and documentation tools use it to orient themselves once they are already working inside a product's docs. That is a real audience. It is just not the audience most business owners are imagining.
Why we publish one anyway
Ummah Collective publishes an llms.txt at ummah-collective.com/llms.txt, and the reasoning is worth stating plainly because it is not the reasoning usually given. It costs almost nothing to maintain when it is generated alongside the sitemap. It forces a useful discipline: if you cannot describe your company, your services and your markets in one clean paragraph and a short list of links, your website probably cannot either. And it is an inexpensive option on a future in which agents acting for buyers read a site before a person ever does.
What we do not do is promise it to clients as a visibility tactic. The pages themselves earn citations: a direct answer near the top, consistent entity information everywhere the business appears, and crawler access that has been decided on purpose rather than by accident.
What to put in an llms.txt file
- One H1 with your exact company name, written the same way as on your Google Business Profile, LinkedIn page and invoices. Entity consistency is the one thing here that also helps everywhere else.
- A one-paragraph summary in a blockquote: what you do, for whom, where, and in which languages. Plain statements, no slogans, nothing you would not stand behind in a contract.
- Short sections of links to the pages that actually describe the business: services, pricing or process, case studies, contact. Give each link one sentence explaining what the reader will find. Twenty well-chosen links beat two thousand.
- An Optional section for material that is useful but secondary, such as older articles or archive pages, so a model short on space knows what it can skip.
- Generate it in the same build step as your sitemap, serve it as plain text at /llms.txt, and treat a broken link in it as a broken link on the site. A stale map is worse than no map.
llms.txt: common questions
- Does llms.txt improve SEO or Google rankings? — No. Google has said it does not use the file, and Google Search and its AI features rank and cite the normal pages of your site. Publishing llms.txt neither helps nor hurts your position in Google.
- Will an llms.txt file get my business cited by ChatGPT? — There is no evidence that it does today. A large 2025 study found no link between having the file and being cited by AI assistants. Citations come from clear, well-structured pages that crawlers are allowed to reach.
- Is llms.txt the same as robots.txt? — No. robots.txt tells crawlers what they may access; llms.txt only suggests what is worth reading. If robots.txt or your CDN blocks an AI crawler, an llms.txt file changes nothing.
- Should a small business bother creating one? — It is reasonable if it takes under an hour and is generated automatically. It is not reasonable to pay for it as a standalone service or to prioritise it over fixing the pages it points to.
Keep reading
Build something
worth trusting.
Tell us what's slowing your business down. We'll show you the system that fixes it — and how fast.