How it works
Why we count, and how a study runs
Asking an assistant yourself shows you one answer on one day. A study runs at least a thousand simulated buyer conversations, reads every mention and checks the sources, then hands you the fixes in order.
The obvious question
Why not just ask the engines yourself?
-
One answer is one draw.
Aleve was named in 55 of our 120 headache answers. Ask once and it is close to a coin toss whether you see it. Across 600 people, the same question returned the same brand list fewer than once in a hundred times (SparkToro and Gumshoe, 2,961 runs).
-
The engines read other sites first.
None of the twelve sources cited most in the headache census is a maker’s or a retailer’s own site. goodrx.com was cited in 34 of 120 answers, mayoclinic.org in 13. That is where a fix list starts.
-
Counting is reading.
Store brands were mentioned 39 times. We read every mention: 38 were a place to buy or one example among several, 1 was a product. A keyword count would have reported 39.
How a study runs
Three steps, then the same count again
-
Measure
At least 1,000 simulated buyer conversations a study, built from 50 to 100 buyer scenarios, each simulated across 20 buyer types and free and paid versions of the leading AI assistants. Every brand’s share is printed with the count it came from.
-
Find the sources
Every domain the engines cite gets three checks: popularity rank, whether its homepage fits the topic, and signs of a recent ownership change. A domain that fails is marked flagged, in words.
-
Fix, in order
At the Reveal you get a signed Shelf Report and a fix list graded by evidence. The Re-check at day 90 runs the same buyer scenarios again and shows whether the number moved; the Category and Market Studies include it.
In every health category
We measure and source. We do not write, edit or suggest any statement about what a product does; that stays as your label states it.
The fix list covers where you are missing and what stops an assistant reading your pages, never what your product says it does.
Questions
Before you ask
Do you write health claims for us?
No. We measure and source. We do not write, edit or suggest any statement about what a product does; that stays as your label states it. The fix list covers where you are missing and what stops an assistant reading your pages, never what your product says it does.
Why pay to be measured?
You are not paying for the count. You are paying for the fix list it produces: which sources AI trusts in your category you are missing from, what on your own site stops an assistant reading it, and what to do first. The count shows the list was needed and, at the Re-check, whether the work moved the number.
Is a brand in your census a recommendation?
No. A census reports which names the engines gave, how often, and which sources they cited. The names are measurement results, not recommendations.
Which assistants do you measure?
Free and paid versions of the leading AI assistants. Claude, ChatGPT and Gemini are live; the wider every-model study is under way. Every census page names the assistant behind each number, and the method page lists them all.
Do you promise a ranking?
No promised rankings, no bought reviews, no hidden text, no incentivised posts, with no guarantee of results, stated in the statement of work.
Start with three buyer scenarios
We send back how often AI named each brand, which sites it cited and one source worth a closer look.