ChatGPT SEO: how to find out whether ChatGPT mentions you, and how to change it

The question you are actually asking

You typed your name into ChatGPT, or asked it "which supplier of X would you recommend". Two cases:

• It cites you. You do not know whether that is stable, or why.
• It does not cite you. You do not know who it names instead, or where that name came from.

Either way, you do not have the answer. This page explains how to get it, and what moves it.

How ChatGPT picks a name

ChatGPT with web search enabled does not answer from memory. It turns your question into several queries, opens three to eight pages, and writes a synthesis from what it read there.

The consequence is mechanical: if none of the pages it opened names you, you cannot appear in the answer. The quality of your site does not enter the equation at that moment — it simply is not open.

What our public measurements show is that the pages opened are almost never brand sites. On a question about campsites in Normandy, the five sources ChatGPT consulted were five tourist boards: not a single property read directly. The most-cited name in that whole market is not a campsite, it is the regional tourist board. The full study.

Why testing by hand proves nothing

This is where nearly everyone goes wrong, and it is verifiable in two minutes.

A language model is not deterministic: ask the same question three times and you get three answers, sometimes naming different brands. Your single test therefore carries the weight of a one-person survey. You can be absent from the answer you saw and present in one answer out of three — or the reverse, which is worse, because you think you are fine.

And if your question contains your name, the result is worth nothing at all: ChatGPT will talk about you because you asked it to. The only useful question is the one asked by a buyer who has never heard of you.

The problem in detail, with figures.

The three levers that weigh

1. Being named on the pages ChatGPT opens. Trade directories, comparison sites, review sites, specialist press, forums, local bodies. This is the dominant lever, and it sits almost entirely off your site. You still need to know which ones: they change completely from one market to the next — forums in one sector, institutions in another, merchants in a third.

2. Being identifiable through a precise constraint. Vague question, vague answer: directories and large chains. Precise question, precise names. Our measurements show independents named as soon as the question tightens — a particular facility, proximity to a place, a usage constraint. That is where the room is.

3. Writing quotable facts. ChatGPT cites what it can attribute: a figure, a date, an area, a compatibility. It does not cite "impeccable service". Every superlative on your site is a sentence the AI cannot reuse.

What does not weigh, or barely: volume of content, repeated keywords, and "GEO optimisations" sold as markup to add. The technical layer only matters when it blocks — a robots.txt forbidding GPTBot, content loaded only in JavaScript.

Checking properly, once

A measurement that stands up needs four things: several questions, several assistants, several passes, and no mention of your name in the questions.

That is what our GEO audit does: 40 questions written for your market, put to ChatGPT, Perplexity and Claude, three times each — 360 measurements. The report gives your citation rate, your recommendation rate, the competitors named in your place, and the sources actually read in your market.

€149 excl. VAT, report within 48 hours.

Frequently asked

Does ChatGPT read my site? It can, if GPTBot and OAI-SearchBot are not blocked by your robots.txt. But in our measurements it mostly reads something else. Allowing the crawler is necessary, not sufficient.

Do you have to pay OpenAI to be cited? No. There is no advertising slot inside a ChatGPT answer today. The place is earned on the sources.

What about Perplexity, Claude, Gemini? The mechanism is the same — read a few pages, synthesise, name two or three — but the sources retained differ from one assistant to another. That is why we query three: being cited by one says nothing about the other two.

How long before a change shows? Assistants re-read the web continuously, without the indexing delay of a classic engine. A new presence on a read source can appear within weeks — and disappear just as fast if the source changes.