01
Questions that name you are counted apart.
A question with your name in it is answered with your name almost every time. Your headline figures use only the questions that do not mention you, and the screen says which set a number comes from.
Method
How we measure, and what we refuse to report
Questions that already name your brand make any mention rate look good. Here is how we keep them apart, what we count, and where our numbers stop.
One scan, two very different numbers
Say your brand is Postprism. When a question has the word Postprism in it, assistants name Postprism in 18 of 20 answers. When it does not, they name it in just 2 of 40.
Added together, that is 20 of 60, a third of all answers, and you would look well known. Your headline visibility uses the second number, because it shows how often your brand comes up when the question did not name you.
The first number is still useful. Answers to questions that name you show whether assistants describe you correctly and which of your pages they link to. So you get both numbers, one beside the other. They are never added up into one.
Illustrative.
What gets counted
Which answers go into a figure, and what each one adds.
02
Every number shows what it is based on.
We show both the count and the total, such as "named in 3 of 40 answers", so you know exactly what a figure means. Only successful answers are counted. Numbers shown together use the same total, making them easy to compare.
03
Each answer counts once per brand.
A long answer that names a competitor three times still counts once for them, so repetition cannot make a rival look bigger than they are. Answers that name no brand at all still count in the total, so your share is never inflated. We look for the brand names you confirm at setup, so a nickname or misspelling is not counted.
04
See whether a rival wins with its own pages or someone else's.
When an answer links to a competitor's own website, it counts for them. When it links to a review site that writes about them, it counts for the review site. So you can tell whether a rival is winning with their own pages or with what others say about them. Every other link still appears in your list of sources, and being named and being linked to are always shown as two different things.
What you can check yourself
Every figure leads back to the answers it came from.
05
Every answer is kept word for word.
The full answer is stored as it came back, with the pages it linked to. Some screens add a model's reading of it, such as why you lost, but that reading is kept apart and never replaces the answer. Every figure leads back to the answers it was counted from, so you can read what an assistant actually wrote.
06
A failed question is shown, not hidden.
Sometimes an assistant returns an error instead of an answer. That question stays in your list of answers, marked as failed. It is left out of the total, so a failure can neither raise nor lower a rate.
What a model reads for you
Some findings are a reading of the answer, not a count. They are shown that way.
07
Why you lost is a model's reading, not a fact.
Who was named and which pages were linked come straight from the answer. Why they won does not. A model reads the answer and suggests how the winner got there, such as a review roundup, a comparison page or a directory listing. That is an interpretation, not the assistant's own reasoning, and it does not promise that a new page would change the answer. The screen says a model wrote it and shows it beside the answer. A loss the model cannot place is left as "Not sorted yet".
08
What an answer says about you is read by a model too.
Whether an answer speaks for or against your brand is a model's reading, sentence by sentence. Each reading is shown with the exact sentence it came from, so you can judge it yourself.
What a change means
Answers move on their own. A change has to be bigger than that.
09
A change has to beat the usual swing.
An assistant asked the same question twice can answer differently, even when nothing changed. The swing check measures that. It takes up to 20 of your questions that do not name you and asks each one 3 more times on every assistant. A question counts as unsteady when your brand was named in some of those answers and not in others. Changes in wording or links do not count. From the share of unsteady questions, it takes the highest share the sample cannot rule out, at 95 per cent confidence, and the jumpiest assistant sets the bar. Your mention rate has to move by more than that before a change is shown as "Bigger than normal variation". A smaller move is shown as "Too small to tell". With fewer than 10 sampled questions, no verdict is given at all.
10
What the swing check cannot tell you.
It judges whether you were named, not changes in wording or links. Its repeats are asked within one scan, so the real swing from week to week is probably larger. The bar is a cautious rule of thumb, not a statistical test of the change itself, so a change that clears it is worth a look rather than proof. It is off until you turn it on in Settings, because it adds calls to every scan.
11
A trend needs two scans.
With one scan there is no trend, and the chart says so rather than drawing a flat line. Counts are drawn from zero, and a rate chart never zooms in so far that a small change looks large.
What we leave out, and why
What we do not report, on purpose
- A ranking. An assistant naming you first is not the same as preferring you.
- Search volume. We do not measure how often anybody types a question.
- A single visibility score that blends everything into one number.
- Traffic. We cannot see who visits your site after reading an answer.
- A comparison against a figure we did not measure.
What a scan cannot tell you
- Your questions decide what is measured. A scan covers the questions written for your site and approved by you. It does not cover every question buyers ask, or how often each one is asked.
- A scan shows that something changed, not why. A change after you publish a page may have another cause.
- Answers change from one week to the next, even when nothing else did. A single scan is a reading, not a verdict.
- We ask each provider directly. A person signed into the app can get a different answer.
- Every figure comes from your own scans of your own site. We do not publish a benchmark across other companies.
See the answers behind your numbers
Run a free scan to see where your brand appears, which competitors are named instead, and the AI answers behind every result.