How we measure, so you can check us.

24 questions your buyers would type, in four layers: brand-direct, head category, long-tail category, problem-phrased. Frozen before the baseline. Exact wordings stay sealed until the engagement ends, so nobody optimizes for the test.

4 layers · 24 questions · sealed

Each question runs five times per engine per round. The same prompt can name you at noon and forget you at midnight. A rate is the share of answers that mention you across those samples. We call it rSoV, resampled share of voice.

rSoV = mentioning answers ÷ all answers, n = 5 per question

Four competitor brands we do no work for run in every round: two category leaders, two small tools. If the whole platform shifts, the controls shift too, and that drift is subtracted from your delta instead of being reported as progress.

your delta − control delta = what the work did

Every answer is judged on three questions: were you mentioned, was your domain cited, and was the company described actually you rather than a name twin or an old version of you. Raw answers are archived so any verdict can be re-checked.

mentioned · cited · identity correct

Each round ships mention rates, citation rates, identity-confusion rates, the sources engines cited, and a log of what we changed and when. If a number refuses to move, it is reported refusing to move.

see it applied to us → Lab