Why asking ChatGPT about your brand is not a measurement
Asking an AI assistant about your own brand returns one polite answer to a question no buyer asks. To know whether AI recommends you, measure unbranded buyer questions across several engines and count how often you are named.
The test most people run
You open ChatGPT and type “what is Acme CRM?”. It replies with a tidy, accurate summary. It feels like good news. It tells you the model has heard of you and nothing more, because you supplied the name.
The question a buyer asks
A buyer does not know your name yet. They ask “best CRM for small agencies” and receive a shortlist. Either you are on it or a competitor is. That answer shapes the shortlist, and it is the one worth measuring. This kind of question is called a buyer question.
Three reasons one answer is not enough
- Wording changes the answer. “Best CRM for agencies” and “CRM for a five-person studio” can produce different shortlists.
- Engines disagree. ChatGPT, Gemini and Claude draw on different sources and name different brands.
- Answers move. The same question can return a different answer next week.
What to measure instead
Write a set of buyer questions for your market. Ask each one to each engine without naming yourself. Count the answers that name you. The result is a mention rate per engine, plus a list of the competitors named in your place and the sources the engines cite.
You can do this by hand with a spreadsheet. It is slow, and it needs repeating to show a trend. Mentionary automates the same method: it writes the questions from your site, asks ChatGPT, Gemini and Claude each one, and reports the rate.
Doing it from inside your assistant
With the Mentionary connector added to Claude or ChatGPT, you can ask for the measurement in plain language and have the assistant turn the numbers into a board report, a competitor view or a list of fixes. The first analysis is on the Free plan.
Related questions
Do not ask about your brand by name. Write the questions a buyer would ask about your category, ask each one without naming yourself, and count how many answers include you. Repeat on Gemini and Claude, because the engines differ.
Enough that one answer does not swing the result. Ten questions give a rough first reading. Twenty-five or more per engine give a rate you can track over time.
Not from its own knowledge. A model has no record of what it told other people. It can only report visibility if a tool runs the questions and counts the answers, which is what the Mentionary connector does.
Your buyers are already asking AI. Find out who it tells them to pick.
Add Mentionary to Claude and ask one question. Your first analysis across ChatGPT, Gemini and Claude is free, inside your assistant or in the web app.
Free plan: 10 buyer questions, 3 engines, 2 competitors. No card.