llms.txt: Does It Actually Help AI Visibility? (Evidence-Based Answer)

The short answer is no — not in the way it is usually sold. For Google Search and AI Overviews, the published evidence points the same way: the file is ignored, it shows no measurable link to being cited, and in practice it is almost never even requested. That does not make llms.txt a scam. It makes it a low-priority file that has been promoted far beyond what the evidence supports.

If you are deciding where to spend the next hour on AI visibility, spend it on the page itself, not on this file. The rest of this article shows why, with the evidence, so you can make that call on data rather than on a sales pitch.

What llms.txt actually is

llms.txt is a proposed convention, not a web standard. The idea is simple: a site places a plain-text file at its root, at /llms.txt, containing a structured summary of its content in a format designed to be easy for a large language model to read — page titles, short descriptions, and links, in markdown form.

The pitch behind it is easy to understand. Search engines have robots.txt and sitemaps; AI systems, the pitch goes, should have their own equivalent. Give the model a clean summary, and it will find, understand, and cite your content more easily.

That pitch has one problem: it assumes the systems that decide citations actually read the file. The evidence says, for the systems that matter most, they do not.

What Google says

Google states that Search ignores llms.txt. That is the decisive point for most sites, because Google Search and its AI Overviews are where the citation question is usually being asked. A file that the search engine explicitly ignores cannot be the reason a page is cited in its results — or the reason it is not.

This is worth stating plainly, because much of the llms.txt advice online is written as if the file were already part of how search works. It is not. It is a proposal, and the largest search engine has said it does not use it.

What the measurement studies found

Two published measurements, both from 2026, tested the claim against real data rather than theory.

First, SE Ranking analysed roughly 300,000 domains and found no citation correlation: sites with an llms.txt file were not cited more often than comparable sites without one. Correlation is not causation, and a single study does not settle a question on its own — but a null result across 300,000 domains is a strong signal that the file is not doing the work its promoters claim.

Second, an Ahrefs log study looked at whether the files were even being fetched. It found that 97% of published llms.txt files were never requested at all. A file that is never requested cannot influence what a system says about your site, however well written it is.

Taken together, the three pieces of evidence agree: the search engine ignores it, the citation data shows no effect, and the server logs show almost nobody asking for it. That is not a close call. It is a consistent answer from three different angles.

What the evidence does not prove

Honesty requires the limits as well as the conclusion.

The evidence is about Google Search and AI Overviews, and about citation in general. It does not prove that no individual tool, crawler, or assistant will ever read an llms.txt file. A specific product could document that it reads the file, and for that product the file would matter — but that is a tool-specific behaviour, not a search-ranking or citation mechanism, and it would need to be shown for that tool, not assumed for all of them.

The evidence also does not say the file is harmful. There is no published finding that adding llms.txt hurts rankings or citations. It is, by the same evidence, simply inert for the purpose most people add it for.

So the fair statement is this: llms.txt is cheap to add and almost certainly harmless, but there is no evidence it helps AI visibility in Google Search or AI Overviews. Treat it as a minor housekeeping file if you like — not as an AI visibility strategy, and never as a substitute for one.

What actually moves AI citations instead

The same body of research that dismisses llms.txt points clearly at what does work. The original generative-engine-optimisation research (Princeton et al., KDD 2024) found that quotations, statistics, and cited sources raised AI visibility by up to around 40%. Later 2026 guides converge on the same playbook: answer-first passages, specific figures, named sources, clear entity definitions, and coverage of the follow-up questions a searcher actually asks.

None of that lives in a separate file. It lives in the page itself — in the opening passage that answers the query directly, in figures a system can quote rather than paraphrase, in sources a reader can verify, and in supporting pages that cover the related questions so the topic is answered completely.

That is also why llms.txt is such a tempting distraction. It promises the benefit of being citable without the work of being quotable. The work is the page: a direct answer first, evidence attached to every claim, and a topic covered deeply enough that the page is the best source available, not merely a summary of one.

How to check your own site

You do not need to take the published studies on trust. Two checks, both free, tell you what is happening on your own site.

1. Check your server logs for llms.txt requests. Look for requests to /llms.txt over the last 30 days, and note which user agents made them. If the file has never been requested — the outcome the Ahrefs study found for 97% of published files — then no system is using it to learn about your site, and adding or improving it changes nothing.

2. Check your citations, not your file. Search the queries that matter to your business and note whether AI Overviews appear, whether they cite anyone, and whether they cite you. Then compare the cited pages with yours: do they answer first, cite specific figures, and cover the follow-up questions? That comparison, not the presence of a file, is where the gap usually is.

If both checks point the same way — no requests, no citations — the fix is the page, not the file.

The decision, in one paragraph

Add llms.txt if you want to: it costs minutes, and there is no evidence it harms anything. Do not pay anyone to create it, do not prioritise it above content work, and do not expect a citation change after adding it. If a vendor sells llms.txt as an AI visibility service, ask for the evidence — the published data says the file is ignored by Google Search, shows no citation correlation across roughly 300,000 domains, and is never requested in 97% of cases. The work that moves citations is older and less glamorous: clear answers, verifiable evidence, and complete topical coverage on the page itself.

If you have already added llms.txt and your citations have not moved, that is the expected outcome, not a sign that the file needs to be longer or more detailed. The next step is the same as it would have been without the file: make the page the source an AI system would want to quote.

Next step: Request a free written SEO audit of your site — a concise written report on what the data actually shows, whether an AI Overview or a ranking loss took your traffic, and what to fix first: Free SEO Audit You can also see how this diagnostic fits into a full growth engagement on the Services & Pricing page.