Method
How we measure
Every number we publish is a rate built from many repeated questions, not a screenshot. Here's how it works, and the rules we hold ourselves to.
Five steps
-
Start with how customers really ask
Before a study, we gather evidence of how customers word their questions: what they write in their own words, what they search for, and what the people who take their calls hear every day. Each need is then asked in a few different ways, from a short search like "best dentist in Belleville" to a full question with personal detail, because a small change in phrasing can change the answer.
A small pilot shows whether the wording matters before the questions are locked. If it does, that's a finding, and we report it. Every question we asked is published with the study, so you can judge them yourself.
-
Ask all six platforms
ChatGPT, Claude, Gemini and Perplexity, through each company's own developer access. We record which version answered and keep it the same through a study, so results stay comparable. Google's AI answers and map results come from real Google searches.
-
Ask over and over, spread out
Each question is asked many times, spread across several days and different times of day, so no single moment decides the result.
-
Keep every answer
Every answer is stored in full, along with the websites it cites. The businesses it names are matched against a list of the market, so a name counts whether it's written "Dr. Smith" or "Smith Family Dental".
-
Count honestly
A rate is the number of answers that named you, divided by the number of times we actually asked, counted from the data, never assumed. Google's AI answer is only counted when Google actually showed one.
What a rate means
"Named in 40% of answers" means that, of all the times we asked, 40% of the answers included you. Measured this way, a rate is accurate to within about 2–4 points. Comparing two rounds adds the margin of both, so a change of more than roughly 6–11 points, depending on the platform, is a real change worth knowing about.
named ÷ asked
Every rate, on each platform, for every business.
Rules we hold ourselves to
We measure. We don't predict.
We once built a score that tried to predict how visible a business would be from its website. Our own measurements showed it didn't work, so we dropped it. Now we only report what the AIs actually say.
We decide the test before we see the results.
Questions are locked before a study starts. For larger studies, we also write down how we'll judge the results in advance, so the data can't be bent to fit a story.
We label every change.
Each answer records which AI version gave it. When a platform changes, or we change a setting, new results are labelled separately and never quietly blended with old ones.
We check our own tools.
Automatic checks run on every batch, for example flagging any answer that was cut short. When we find a problem, we fix it, re-measure, and say so.
Nobody pays to be in a study.
We don't sell fixes, rankings or placements, and we never charge anyone to appear in our results.
Results stay private.
We publish patterns about how AI behaves. We don't publish which businesses were named, or how often. Each business can get its own results.
What we can't see
The answer any one person gets can differ from what we measure. Their location, their history with the app and their settings can all change it. We measure the standard answer, many times over, which is the fairest common picture there is. And because the AIs change often, every result is dated: it describes what the platforms said at the time we asked.
Put the method to work on your business.
Same questions your customers ask, same rigour, your market.