How to check whether your brand appears in ChatGPT, step by step

Why checking on your own laptop tells you the wrong thing

Most people run this test once, in their everyday ChatGPT account, by asking about their own brand. The answer looks fine, and it means almost nothing. Two things distort it.

The first is personalisation. ChatGPT keeps memories of what you have told it and reads your custom instructions, so an account that has discussed your company for months will favour your company. OpenAI's own help page on Temporary Chat says a temporary chat does not read or create memories, but that it still follows your custom instructions if they are on. So the clean setup is both: temporary chat, custom instructions off. The same logic applies to a Google account that has visited your site a hundred times, which is why we test in a private window.

The second is the question itself. Ask "is [brand] any good?" and the engine will find you, because your name is in the prompt. The number that matters is whether you are named when the buyer does not know you exist. That is the unbranded number, and it is the only one this method produces.

Step 1: write ten questions that do not contain your name

Take the questions a buyer asks before they have a shortlist, and spread them across four moments. For a company that sells language training to businesses, that looks like this:

• Category discovery. "Best business English training for a sales team of 20 in Lyon."
• Comparison. "Online language courses or in-person classes for employees, which works better?"
• Recommendation. "Which provider should a mid-sized company use for corporate language training in France?"
• Problem. "How do I get my team to B2 English in six months?"

Two or three per moment is ten. Write them the way the buyer would type them, in the buyer's language, and keep your brand and your product names out of every one. If you already track prompts in a tool, take ten unbranded ones from that list. The full method for building the set is in an earlier post.

Step 2: set up a clean session on each engine

ChatGPT is in the title, but a buyer in Europe now meets four engines, and each has its own clean-session recipe. Google switched on AI Overviews and AI Mode in France on 22 July 2026, later than most of the EU (see Blog du Modérateur's report), so French teams in particular are looking at something new.

Ask in the language your buyer uses. A French buyer asks in French. If you serve several markets, run the set once per language, because the engines do not treat them the same way and the sources they cite differ.

Step 3: ask, and record four things

Paste each prompt, wait for the full answer, and fill four columns before you read anything else.

• Named. Does the answer mention your brand at all?
• Linked. Is one of your pages among the sources?
• Position. Do you appear in the first third of the answer, or at the end of a long list?
• Who else. Every other brand named, and the domains the engine cites.

The last column is the one most people skip, and it is where the work comes from. The sources an engine cites for your category are the pages you need to be on, and most of them are not the brands' own sites. Screenshot every answer as well. Engines change, and you will want the before.

Step 4: repeat three times, same prompts, same engines

Ask the same question twice and ChatGPT may name different brands. One answer is a sample. Three answers to the same prompt, spread over a week, are a pattern. Ten prompts, four engines, three runs is 120 answers, which is an afternoon spread across a week, and it is the smallest set we would draw a conclusion from.

What the sheet tells you

Three numbers and one sentence. Per engine: the share of unbranded prompts where you are named, the share where you are linked, and the brand named most often when you are not. The sentence is the single prompt where you are invisible and a competitor is the default answer, with the source the engine cited for it. That prompt is where you start.

Expect the sources to surprise you. In an October 2025 Ahrefs study of the 1,000 pages ChatGPT cited most, 28% had no organic search visibility at all, and nearly 90% of the pages with a detectable update date had been updated that year. Ranking on Google helps, but it is neither required nor enough. The engines also reach for recent pages, and for pages that never ranked.

What this means for you

Do the first pass this week. It costs an hour and no tool. Then fix one thing: the page that should have answered the prompt where you were invisible, written so that its first paragraph answers the question directly. Rerun the set next month and see whether the sheet moves.

If you would rather have it done properly, across markets and languages, with the fixes shipped rather than listed, book a consultation and bring the sheet. It is the best brief a strategist can get.