Does llms.txt Work? Google Says Not for Search, and the Data Agrees

Not for Google Search. Google’s own documentation says Search ignores llms.txt files entirely, and that publishing one “will neither harm nor help your site’s visibility or rankings.” A study of 137,210 domains found that 97% of the llms.txt files being published received zero requests in a month. I listed llms.txt as a technical deliverable on this site until recently. The evidence turned, so I stopped.

This is a slightly uncomfortable post to write, which is usually a sign it is worth writing. The proposed convention was reasonable, plenty of credible people recommended it, and I was among them. Then the data arrived.

What did Google actually say?

On 15 June 2026 Google added a note to its AI optimization guide clarifying its position. The documentation now states, in a section explicitly labelled mythbusting:

“You don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn’t use them.”

And on maintaining one anyway: “Doing so will neither harm nor help your site’s visibility or rankings in Google Search, as Google Search ignores them.”

That is about as unambiguous as search documentation gets. It is also narrowly scoped, and the scope matters: this is a statement about Google Search, including its generative features. It is not a statement about ChatGPT, Claude, Perplexity or the coding agents that do sometimes read these files. Anyone telling you llms.txt is universally worthless is overclaiming in the opposite direction.

What does the usage data show?

Ahrefs published a study of 137,210 domains on 15 June 2026. Of those, 28% had published an llms.txt file. Of the files published, 97% received zero requests during May 2026. Nothing fetched them at all.

Among the small remainder that were fetched, AI retrieval bots accounted for just 1.1% of requests. The heaviest requesters were SEO audit tools checking whether the file existed, which is a fittingly circular result: the main consumer of llms.txt is software verifying that someone followed the advice to create llms.txt.

Two honest caveats on that study. Ahrefs notes its own customer base skews technical and SEO-aware, so 28% adoption should be read as an upper bound rather than a typical rate. And a fetch is not the same as a read, so even the small positive numbers are ceilings rather than confirmed usage.

Why did this get so popular?

Because it was plausible, cheap and legible, and those three things together will carry an idea a long way without evidence.

The pitch made intuitive sense. Models need to understand your site; robots.txt exists for crawlers; therefore a file explaining your site to language models should help. It cost almost nothing to implement. It gave agencies something concrete to put on a scope of work at a moment when clients were anxious about AI search and wanted a technical answer. And there was no easy way for a buyer to test whether it worked.

That combination describes a fair amount of early GEO advice, which is the more useful lesson here. When a new discipline arrives, the tactics that spread fastest are the ones that are easiest to explain and sell, not the ones with the strongest evidence. It takes a year or two for the data to catch up, and some of what everyone agreed on turns out to be decoration. I would rather be the person who tells you which parts, including when I got it wrong, than the person still selling the decoration.

What I changed on this site

I have removed llms.txt from the technical deliverables I list, because charging for something on the basis that it drives citations, when the only primary source on the question says Google Search ignores it, is not defensible.

The file itself is still there. Google is explicit that it causes no harm, some agents do read these files, and the convention may yet find a use in agent workflows. Keeping a file that costs nothing is different from selling it as visibility work.

What replaces it in the scope is the work that the evidence actually supports: a resolvable knowledge-graph entity so a model can identify you, corroboration from third-party sources the engines already trust, substance structured so a specific claim can be lifted and attributed, and verified crawler access so any of it can be read at all. That last one is worth checking this quarter in particular, given Cloudflare’s September default changes.

None of that is as tidy as adding a text file. That is exactly why the text file was so appealing, and it is why most DIY GEO stalls: the real levers are interdependent, slower, and harder to put on an invoice.

If you want to know which of the tactics you are currently paying for hold up, that is what the audit is for.

Frequently asked questions

Does llms.txt actually work? Not for Google Search, which states plainly that it ignores these files and that they neither help nor harm rankings. Across 137,210 domains, 97% of published llms.txt files received zero requests in May 2026. No published evidence shows it driving citations in any major AI search product.

Should I delete my llms.txt file? You don’t have to. It causes no harm and some tools read it. But stop counting it as AI-visibility work, and be sceptical of anyone billing it as a citation tactic.

Why did llms.txt become so popular then? It was plausible, cheap and easy to sell at a moment when everyone wanted a concrete technical answer. It spread on plausibility rather than evidence.

What actually earns AI citations instead? A resolvable entity, third-party corroboration, quotable substance that can be lifted cleanly, and crawler access. Less tidy than a file, which is the whole problem.