Before any of this becomes a dashboard, do it by hand once. Pick the single question you would most like to win, ask it, and read what comes back slowly.
Use a fresh session with no history. Your own account has months of context about who you are and what you sell, and a model that already knows you is not a fair test of whether it would mention you to a stranger.
Write down every brand named, in order. The order matters more than the count — first mention carries most of the weight, and the brands after the third are decoration.
If you are not in the list, resist the urge to explain why. Just note it.
If you are mentioned, find the exact words around your name. That clause is your positioning as it currently stands, compressed to about eight words by something that has read a great deal about you.
Check what got cited. Then check whether those pages are yours, and whether they are the pages you would have picked. This is where most of the surprises live.
Rephrase the question slightly — same intent, different wording — and watch how much moves. That variance is the single most useful thing to understand early, because it sets your expectations for everything that follows. A brand that appears in one phrasing and vanishes in the next is not being tracked, it is being sampled.
Once you have felt that by hand, the case for tracking prompts repeatedly rather than checking once makes itself.
The same audit run against your closest rival is worth the extra ten minutes, and it is more informative than running it against yourself a second time.
You are reading for contrast. Which questions name them and not you. Which adjectives attach to them. Whether the pages cited for them are product pages or third-party writing, because that tells you which half of the system is carrying them.
Most usefully, it shows you what a good result looks like in your category. It is difficult to judge whether your own answer is weak without something to hold it against, and a competitor who is doing well gives you a target that is demonstrably achievable rather than theoretical.
Before you run any of it, write a sentence predicting what the answer will say. Then compare.
The gap between what you expected and what came back is the actual finding, and it disappears the moment you read the real answer — hindsight closes it within about a minute. Recording the prediction first is the only way to keep it.
Contents
How to put this in a monthly report without over-claiming, and what to say when the numbers move for no reason.
Read →Everyone starts by tracking the prompts they wish customers asked. Here's how to find the ones they actually do.
Read →Not your homepage. A working theory of the sources that shape what AI answers say about a brand, and what that means for where you spend effort.
Read →Tell us your domain and we'll send you a visibility report — free, no card.