Case studies
We only show two kinds of thing here: anonymised real measurements, and ourselves. Named case studies wait for written client permission โ until then, one entry is better than a fabricated second.
What you won't find here: No invented clients, no borrowed logo wall, no made-up โ+X% upliftโ. Every number below comes from real sampling in our production database. The brand has not authorised being named, so it is published anonymised โ we don't even state the category, since that alone would identify it.
One real quarterly re-measurement
Anonymised ยท real data35 prompts ยท 8 engines ยท Chinese and English ยท a 30-day window. The three numbers below come from one and the same sampling run.
โ One score split into two โ only then does it mean anything
An overall score of 2.12 looks bad. But 27 of the 35 prompts target the broad category โ 77% of the weight sits on a battlefield where this brand is simply absent. The overall score faithfully reflects that weighting: it doesn't say โthis brand is weakโ, it says โwe're mostly measuring where it hasn't turned up yetโ. Reporting the single number alone would lead a client to exactly the wrong conclusion.
โก Three weeks earlier we refused to draw this conclusion
In the previous report, the in-category mention rate read 50% โ on a sample of 2. We flagged it as โโ n=2, a lead rather than a conclusionโ and did not let the client act on it: one flipped cell would have made it 0% or 100%. This period the sample reached 94 cells at 97.9%, and only then did the lead become a conclusion.
When n is too small we say nothing; when it is large enough we speak. That is the reason to trust the rest of the numbers โ and the only thing we ask you to take on trust.
โข The 0.2% is solid
1,647 cells, 8 engines, both languages, 30 days โ mentioned 4 times. That isn't one platform's bias; it's across-the-board absence. This conclusion can be acted on.
The same ruler, pointed at ourselves first
Verify it yourselfUntil we have a named client endorsement, the most honest proof is to lay our own practice open. Each of the four below can be checked on the spot โ anything we couldn't let you check, we left out.
The scoring methodology is published in full
How scores are computed, how citations are judged, how thin samples are flagged โ auditable and reproducible.
Read the methodology โ8 engines, each scored separately
Only engines that actually work are labelled "Live" โ engines still being integrated never enter a score or a chart.
See the engine roster โIndustry benchmarks are public
Per-industry percentiles are published openly; industries with fewer than 5 sampled brands publish nothing โ a privacy floor enforced in the database, not by policy.
See the benchmarks โRun a free checkup yourself โ no signup
You don't have to take our word for it: enter your own brand name and get a snapshot in minutes.
Run a free checkup โWant to be the first named case study here? We don't pay for a name and we don't trade discounts for endorsements โ we'll ask you once, and only once your own numbers feel worth talking about.
Start with a free checkup โ