Does llms.txt get me cited?

There is no evidence that it does. A study of 300,000 domains found no relationship between publishing the file and AI citation frequency, and Google says its Search does not use it. It is an hour of work and it is harmless; it is not the lever.

If Gemini cannot cite me, is it worth measuring?

Yes, for the other question. Being named in an answer moves a buyer whether or not a link comes with it, and model memory is a real measurement of what the engine believes about you. What it cannot honestly report is a citation rate, so ours reports none for it rather than a zero.

How long before any of this shows up?

Retrieval and identity repairs land as soon as the crawlers return, usually days to a few weeks, and a closing audit on the same questions shows them. Being named more often follows what the wider web publishes about you, which moves in months.

Can I just ask ChatGPT whether it knows my company?

You can, and the answer will be unreliable. A model's reply is not deterministic, so a single ask is one roll of a die. Counting mentions across repeated samples of the same question is what turns it into a number you can compare to next month's.

Answers

How do I get cited by ChatGPT, Claude, Perplexity and Gemini?

Be retrievable, be resolvable, and be described the same way by sources the engine already trusts. The mechanics differ per assistant: Perplexity searches on every question, ChatGPT and Claude search when they judge it necessary, and Gemini has no web grounding available through the routes an API reaches — so it answers from training data and cannot cite you at all.

The four do not behave alike

We measure all four on the same questions, and they differ before any optimisation enters the picture. Perplexity searches on every request and hands back the URLs it used. ChatGPT and Claude both have web search and both return citation URLs — but they decide per question whether to use it. Gemini, through the API routes available to anyone measuring this, has no search grounding at all: it answers from what it learned in training.

That last one has a consequence people skip past. If an engine never searches, nothing you publish this month can appear in its answer this month. You are optimising for the next training run, and for what the rest of the web says about you in the meantime.

In our own run — 60 grounded answers, 20 buyer questions across the three engines that can search — three of the answers came back with the engine declining to search at all despite having the tool available. A measurement that forces search on is answering "if it searched, would it name you", which is a different question from what a user sees.

What is actually in your control

Retrieval. An assistant fetches the page and reads what it receives. Anything that only appears after the browser executes JavaScript may not exist for it, and a crawler policy written against Googlebot says nothing about GPTBot, ClaudeBot or PerplexityBot, each of which reads robots.txt on its own terms.

Resolution. The engine has to be certain which company you are before it will put your name in a sentence. That is consistent naming, addresses and identifiers across your own pages and everyone else's — the boring half of the work, and the half that decides whether a retrieved page becomes a mention.

Quotability. Assistants lift claims, not paragraphs. A number, a definition, a range or a step survives extraction; a point that only lands after three paragraphs of set-up does not.

What is not in your control is the part that moves the number most: how consistently third parties describe you. An engine composing an answer prefers what it can corroborate.

Why the per-engine advice is mostly the same advice

Almost every guide written per assistant repeats one list with the logo changed, and that is closer to right than it looks: retrieval, identity and quotability decide all four, because all four are reading the same web.

The genuine differences are narrow and worth knowing. Whether the engine searches at all decides how fast anything you change can show up. Whether it returns its sources decides whether you can measure citation, or only mention. And the crawler that fetches your pages differs per vendor, which is the one place a per-engine action list is real.

Common questions

Does llms.txt get me cited?: There is no evidence that it does. A study of 300,000 domains found no relationship between publishing the file and AI citation frequency, and Google says its Search does not use it. It is an hour of work and it is harmless; it is not the lever.
If Gemini cannot cite me, is it worth measuring?: Yes, for the other question. Being named in an answer moves a buyer whether or not a link comes with it, and model memory is a real measurement of what the engine believes about you. What it cannot honestly report is a citation rate, so ours reports none for it rather than a zero.
How long before any of this shows up?: Retrieval and identity repairs land as soon as the crawlers return, usually days to a few weeks, and a closing audit on the same questions shows them. Being named more often follows what the wider web publishes about you, which moves in months.
Can I just ask ChatGPT whether it knows my company?: You can, and the answer will be unreliable. A model's reply is not deterministic, so a single ask is one roll of a die. Counting mentions across repeated samples of the same question is what turns it into a number you can compare to next month's.

Related5

Next step

Free AI Diagnostic

The free site check asks all four engines your buyers' questions and shows you, question by question, who was named, which page was cited and when it was measured.

Start the free diagnostic

Starts immediately in the browser.

Fee: Free
Length: 15 minutes

You keep the ranked list of candidates either way.

See what an engagement costs