How MentionBerry counts
One answer from ChatGPT tells you very little. We ask the same question many times, in a clean session, and show how often your brand was named and how sure that number is.
Why one answer isn't enough
ChatGPT and Google's AI Mode (the chat-style search mode in Google) write a new answer every time you ask, even when you type exactly the same words. A brand can be named in one answer and missing from the next. So a single check, by you or by a tool, is a coin toss you only see once.
We report how often a brand was named across repeat runs of the same question: "named in 7 of 10 runs". That is a share, and it comes with a range.
The range under every number
Ten runs is a small sample. If a brand was named in 7 of 10 runs, the share you would see over many more runs is probably somewhere between about 4 and 9 in 10. With 20 runs (14 of 20 named) the range narrows to about 5 to 9 in 10; with 100 runs (70 of 100), about 6 to 8 in 10. We show that range next to every share, so a change from one week to the next is only called a change when it is bigger than the range. (The range is a 95% Wilson interval, a standard way to put limits on a share from a small sample.)
What we record from each answer
For every run we keep the full answer and note:
- Whether your brand was named, and where in the answer.
- Which other brands were named, so you see who is named instead of you.
- Whether the assistant searched the web for that answer, and which pages it linked to.
- Whether a linked page backs the mention. If the assistant named you but linked no page about you, the name came from what the model already knew (its training), not from a search. That tells you where to work: your pages, the pages that list you, or your reputation in general.
Clean sessions
Answers change with the person asking: ChatGPT can use a person's saved memory and past chats. We ask in a clean session (in ChatGPT, a Temporary Chat set to Unpersonalized, which uses no memory or custom instructions), from a United States connection, so the answer reflects what a new visitor would get rather than what one account has taught it.
Questions written the way buyers ask
The questions are written the way people ask for a recommendation ("What's the best CRM for a small agency?"), not the way a brand describes itself. Where we can, we start from phrasings real people search for.
What we don't do
- No single "AI visibility score" that blends everything into one number. You get the share, the range and the answers behind them.
- No company pays to be counted, ranked or left out.
Where MentionBerry is today
MentionBerry is being built. The guides on this site explain how AI assistants choose what to name, with every figure linked to its source. The first public counts will appear here with the exact questions, the dates and the number of runs.