Do I Need an llms.txt File?

llms.txt is a plain-text file, proposed in 2024, that gives AI tools a short summary of a site: what the business does, who runs it, and a list of key pages with one-line descriptions, without the navigation, ads and scripts a normal HTML page carries. It sits at the root of the domain next to robots.txt and sitemap.xml. A longer companion, llms-full.txt, dumps full page text into one file.

It is not web infrastructure in the way robots.txt or sitemap.xml are. Ahrefs puts it plainly: despite the filename, it is not a robots.txt-style directive; it controls nothing and blocks nothing.

Google's Search Central page AI features and your website (last updated 10 December 2025, read 15 October 2026) is direct:

Table · scrolls sideways

The page does not name llms.txt, but llms.txt is an AI text file. Earlier, in April 2025, Google's John Mueller wrote that, as far as he knew, none of the AI services had said they use it, and that "it's comparable to the keywords meta tag" (Search Engine Journal): a field site owners fill in out of hope.

This is where 2026 brought real evidence. Ahrefs examined all 137,210 domains in its Web Analytics product that received traffic in May 2026, checking each root for a working llms.txt and classifying the requests by user agent. Its findings, published 15 June 2026:

• 28% of those domains published an llms.txt file. Ahrefs notes its customers skew technical and SEO-aware, so that is an upper bound for the wider web.
• 97% of those files received zero traffic in May 2026.
• 96% of the requests that did arrive came from bots. SEO audit tools were the largest group at 21.7%; named AI bots made up 19.5%, with GPTBot first and Claude-Code second.
• "Fetched" does not mean "read", so even those figures are ceilings.

PPC Land's write-up pairs the Ahrefs result with Originality.ai's tracking, which counted llms.txt files rising from 4,088 in June 2025 to 36,120 by May 2026. Adoption is climbing much faster than anyone is reading the files. Ahrefs' own conclusion is that "the cons outweigh the pros right now".

We'd rather say this plainly than leave it implied. Our own site and the Tokas Lex law-firm site we built both serve an llms.txt. Our Ontario Practice Website Index records which Ontario law, clinic and accounting sites publish one. Here is why, and what we don't claim.

1. It costs almost nothing on our builds. The file is generated from the same route list as the sitemap, so it updates itself and can't point at pages that no longer exist. If it had to be maintained by hand, we would weigh it differently.
2. A few tools do fetch it. Ahrefs saw GPTBot and Claude-Code among the fetchers. That is a small audience, but a real one for the price of a generated file.
3. It doubles as a facts check. Writing a one-paragraph description of a business, with its name, location and services stated exactly as elsewhere on the site, catches inconsistencies we would want to fix anyway.

What we don't claim: that it improves Google rankings, earns AI citations, or appears in any measurable result. If a client would rather not have one, nothing else on the site depends on it. Ahrefs also suggests treating the file like code, because whatever is in it may be fed to an AI tool; ours contains only public facts and links.

Keep it short and accurate:

1. A one-paragraph plain description of the business at the top, not ad copy.
2. Contact and location details written exactly as they appear elsewhere on the site.
3. A few headed sections (services, industries, work, guides), each with one-line links.
4. Generated at build time from the same source as the sitemap, so the two never disagree.

The common mistakes are the ones in the example above: ad copy instead of a description, links to pages that were renamed years ago, and business facts that don't match the rest of the site.

For most small businesses, no. llms-full.txt is a second copy of all your content that can silently drift out of date, and the Ahrefs study measured only the index file, so there is no evidence of a larger audience for the full version. The summary file does the useful part: telling a tool where to look. The tool can fetch the real page from there.