The method
The Five Chats: how to check if ChatGPT recommends your business
Published · Updated
Ask it. Write down five questions your customers put to an assistant, open a fresh chat for each, ask all five on ChatGPT, Claude, Gemini and Perplexity, and count how many of the twenty answers say your name. That is the whole method. It takes fifteen minutes, it costs nothing, and the number you get is the one every other claim in this field should be measured against. We publish it in full so you can run it without us, and so you can check anyone who tells you what an assistant "says".
1. Write the five questions
Use your customer's words, not yours. Your customer does not type "leading dermatology practice"; they type "best clinic in Amsterdam for filler that looks natural". Two of your five should be generic ("best hair transplant clinic Amsterdam") and three should be situational: who the person is, where they are, what is wrong. "I live in Amsterdam Noord and want FUE with minimal scarring, where should I go" is a situational question. The generic ones tell you who wins the beauty contest. The situational ones tell you where a smaller practice can be named, and those are the ones worth owning.
Do not put your own name in any question. You are measuring whether the assistant brings you up, not whether it knows you exist.
2. Open a fresh chat every time
Twenty questions, twenty chats. Assistants carry context from one message to the next inside a conversation, so a second question in the same chat is answered by a model that already knows what you asked first. Turn memory off where the setting exists. Log out if you can; a logged-in account carries history the model may use.
One caveat we found the hard way: your location leaks in. On 15 September 2026 we asked ChatGPT "who can help me get my company into AI answers" from a machine in Turkey, logged out, no history, and it named two Turkish agencies. The same question from Amsterdam gives a different list. Note where you ran from, and if your customers are in Amsterdam, run it from Amsterdam.
3. Ask all four assistants
ChatGPT, Claude, Gemini, Perplexity. They do not agree with each other. Asked "AI visibility consultant Amsterdam" on the same afternoon, ChatGPT named five businesses and Perplexity named sixteen, and one name appeared on both lists. A practice named by one assistant and skipped by the other three has a different problem from one named by none, and you cannot see the difference from a single chat.
4. Count the names, and keep the sources
For each answer, write down the date, the question as you typed it, the assistant, the full reply, and every business it named. Then look for the sources. Perplexity shows them under a "Links" tab; ChatGPT has a "Sources" button at the end of the answer; Gemini and Claude show them inline when they searched. Copy those too.
The count is the measurement. The sources are the map. If a competitor is named and you are not, the page that put them there tends to be in that list, and that page is where the work starts.
5. Read the twenty answers as one
Tally it: how many of twenty answers name you, how many name each competitor, which sources come up more than once. Three patterns tell you most of what you need.
Named nowhere. The assistants have no material on you. See the four reasons ChatGPT leaves a business out.
Named on generic questions, skipped on situational ones. The model knows you exist but cannot match you to a specific need. Your pages do not say who you are for.
Named by one assistant, not the others. That assistant found a source the others do not read. Find it, and put the same facts where the others look.
What the number is for
Run it once and you have a baseline. Run it again in a month and you have movement, which is the only thing worth reporting. Our own first run, on 15 September 2026, put Van Found at zero of ten answers on the two assistants we could reach without an account. We logged it, dated it, and we will publish the next one.
The method has a weakness you should know about: it is a sample. Twenty answers on one afternoon is not the model's view of you, it is the model's view on that afternoon, from that location, with those words. Run it monthly, keep the questions the same, and read the trend rather than any single result.
If you would rather not spend the afternoon
Send us your website. We run your five questions across all four assistants, log every answer, and send you the names and the sources within a few working days, free. The report is the method above, done for you, and you keep it either way.
Van Found is an Amsterdam consultancy that gets clinics, law firms and other professional practices named when ChatGPT, Claude, Gemini and Perplexity answer. One client per category, at a time.