No panel, no survey, no modelling. One engine, asked the same question several ways.
One engine, grounded in live web results, asked the same buying question several ways, with the brand names it returns recorded verbatim. No panel, no survey, no modelling.
For each question set we ask several phrasings: the category phrasing a shopper types, the situational phrasing they use when describing a problem, and the open phrasing they use when they do not know the product category exists. We record every brand named, the date, and the engine.
Vacancy is how often the engine names nobody at all. Contestability is how far the named brands are from having a leader. They are different questions and we keep them on separate scales, because blending them into one score hid the real finding the first time we tried it.
A question set whose capture came back empty is published as not scored, never as a zero. A failed capture and an empty category look identical in a number and are opposites in fact.
It is directional and not decision-grade. Prompt sampling reads an output distribution, and published work on ordered brand lists finds very low run-to-run consistency. It says an answer is unsettled. It does not say anyone is asking, and it is not evidence that a product will sell.
A brand appearing here is not endorsed by us and has not paid to be here. Nothing on this site can be bought into a list.
Current capture: probe-2026-09-06.json, engine gemini-2.5-flash (google_search grounded), two captures on file.