Method

How we measure

Most tools give you one visibility number. That number moves between runs, so on its own it can mislead you. We show how much it can move.

1. Step 1

   We choose the questions with you

   We write about 10 questions a buyer in your market would ask an AI assistant. You review them. A fixed template is only the starting point, because the questions decide the result.

2. Step 2

   We ask each question many times

   AI answers change from one request to the next. So we ask every question 20 times or more and count how often each brand is named.

3. Step 3

   We report a range, not one number

   Every rate comes with a 95% margin of error. If a brand shows up in none of 5 answers, all we can say is that it is below about 43%. With 20 answers that limit drops to about 16%. We show the range so you can see how much to trust it.

4. Step 4

   We only call a difference real when the ranges do not overlap

   If a competitor looks ahead but the ranges overlap, the report says the ranking is not proven. It does not invent a winner.

What this does not tell you

• We test the model API. This is not the same as the ChatGPT website, which can search the web and personalise answers.
• A brand is counted when its name or domain appears in the answer as plain text. Brands with a common word as their name need a custom rule.
• Results describe one engine on one day. They can shift when the model changes.
• Our overall range treats your questions as a sample of your market. It is only as good as the questions we choose together.