Key takeaways
- llms.txt is a simple Markdown file that points AI at your key pages. It is cheap to add, but in 2026 no major AI engine confirms it reads one.
- It is a curated list of your best pages, written in Markdown and saved at /llms.txt. It was proposed in 2024 and takes minutes to generate.
- The honest status: Google says no AI system currently uses it, and a 2026 study found 97% of these files were never fetched.
- It is not robots.txt and not schema. It controls nothing and cannot force a citation, so it does not replace your fundamentals.
- For most UK SMBs it is optional. Add it if you like, but spend real effort on technical SEO, content and authority.
Every few months a new file is held up as the thing that gets you into AI answers. Right now it is llms.txt. The pitch is tidy: add one small file, and ChatGPT, Claude and the rest will understand your site and quote you more often. It is worth knowing what is actually true here, because the gap between the claim and the evidence is wide.
llms.txt is a plain Markdown file you place at the root of your domain, at /llms.txt, that lists your most important pages so AI tools can find and read them without wading through menus, ads and page furniture. It was proposed in 2024 and it is cheap to create. But as of 2026, no major AI engine has confirmed it reads one, so for most UK businesses it is a nice-to-have, not a priority. Here is the straight version, with the fundamentals that actually move the needle kept firmly in view.
What is llms.txt?
llms.txt is a curated, machine-readable map of your website, written in Markdown and saved at yourdomain.co.uk/llms.txt. It was proposed by Jeremy Howard of Answer.AI in September 2024, and the specification lives at llmstxt.org.
The idea is reasonable enough. A normal web page is built for people, so it is wrapped in navigation, cookie prompts, related-article blocks and other furniture. A language model trying to understand your site has to strip all that away first. An llms.txt file skips the mess: it gives a short description of what you do, then a clean list of links to the pages that matter, each with a one-line note. The original proposal frames it as a way to hand an AI “the most pertinent and LLM-friendly context” instead of leaving it to guess.
A basic file has an H1 with your name, a blockquote summary, then H2 sections listing your key pages. That is it. Some sites also publish a fuller variant, llms-full.txt, that puts the actual page content into one file rather than only linking to it. You can write one by hand in twenty minutes, and plenty of tools and content management plugins will generate one for you.
What it claims to do
The promise is that llms.txt helps AI tools understand your site faster and describe it more accurately. By pointing models straight at your best pages, the theory goes, you reduce the chance they miss something, misread it, or waste their limited context window on clutter.
On paper it is sensible. Clean, well-organised information is easier for a machine to use, and the proposal has had real traction on the publishing side. Thousands of sites now have one, and some AI companies publish llms.txt files for their own documentation. The catch is the difference between publishing a file and anything actually reading it.
Does anything actually read it?
This is where honesty matters: as of 2026, there is no solid evidence that the major AI engines read llms.txt at all. The proposal is popular. The usage is not.
Google has been blunt. In June 2025, Google’s John Mueller wrote that “no AI system currently uses llms.txt”, adding that it is obvious from server logs: the chatbots people want traffic from fetch your actual pages, not your llms.txt file. He compared it to the old keywords meta tag, a signal that sounded useful and did nothing.
The server data backs him up. An Ahrefs study of 137,000 domains found that while many sites publish the file, 97% of valid llms.txt files received zero requests in May 2026. Nobody, human or machine, was fetching them. No major provider, whether OpenAI, Google, Anthropic or Meta, has publicly committed to reading one in production.
Two things can be true at once. Some AI labs publish an llms.txt for their own docs, which is good housekeeping. That is not the same as their consumer chatbots reading yours when they answer a question about your business. Do not let the first be sold to you as proof of the second.
llms.txt vs robots.txt vs sitemap
llms.txt is often lumped in with robots.txt and your sitemap, but the three do completely different jobs. Only one of them controls anything, and it is not llms.txt.
| llms.txt | robots.txt | XML sitemap | |
|---|---|---|---|
| Purpose | Point AI at your most useful pages, in clean Markdown | Tell crawlers which URLs they may or may not fetch | List your URLs so engines can discover and index them |
| Format | Markdown | Plain-text directives | XML |
| Controls access? | No | Yes, by user-agent | No |
| Mainly read by | AI tools, in theory | Search and AI crawlers | Search engine crawlers |
| Status | Community proposal, 2024 | IETF standard, RFC 9309 | Long-established, universally supported |
| Widely acted on? | No clear evidence | Yes | Yes |
The practical point: robots.txt and a sitemap are settled, understood and genuinely used, so keep both correct. llms.txt is a suggestion the engines have not agreed to follow. It cannot block a crawler, it cannot force a citation, and it is not a substitute for either of the other two.
It is also not schema markup. Schema is structured data embedded inside your pages to label what they mean. llms.txt is a separate file that points at pages. Different mechanism, different place, different job.
How it fits with AEO and the fundamentals
llms.txt sits right at the edge of answer engine optimisation, and the edge is where it belongs until the evidence changes. The things that actually get you cited by AI are not in dispute, and none of them is a text file.
AI engines quote sources they can crawl, understand and trust. That means a sound technical foundation, content that answers real questions clearly, accurate schema, and genuine authority across the web. An llms.txt file does none of that heavy lifting. If your site is slow, thin or hard to crawl, a neat Markdown index will not rescue it. If your fundamentals are strong, the engines are already reading your pages directly.
So do you actually need one?
For most UK SMBs, no, you do not need an llms.txt file, and it should not be near the top of your list. If you want one, it is low-cost and carries no SEO risk: it will not harm your rankings, and if the engines do start using these files, you are already set up. Treat it as a ten-minute tidy-up, not a growth project.
Spend the real effort where the returns are proven. Get the technical basics right, publish content that answers the questions your customers actually ask, keep your schema honest, and build the authority that makes an AI confident enough to name you. Do that, and whether or not you ever add an llms.txt file stops mattering very much.
If you would like a straight read on where your AI-search visibility actually stands, and what is worth doing first, book a free growth review or get in touch. Plain answers, priorities in order, no hype.