Methods demonstration · September 9, 2026
How a 50% Mention Rate Can Hide Zero Discovery
A business can be mentioned in half of a mixed prompt panel while appearing in none of its unbranded buying answers. The pooled rate hides which question supplied the name.
The entire example
This is a constructed arithmetic example, not a live customer benchmark or a reproduction of VisiScan’s weighted scoring formula. Assume 40 unique, completed, classified answers with no missing requests.
| Prompt group | Valid answers | Business mentioned | Mention rate |
|---|---|---|---|
| Branded recognition | 20 | 20 | 100% |
| Unbranded buying | 20 | 0 | 0% |
| Pooled panel | 40 | 20 | 50% |
The pooled calculation is (20 + 0) / (20 + 20) = 50%. The discovery calculation is 0 / 20 = 0%. Both are correct arithmetic; only the second answers whether the business appeared without being named in the prompt.
Why the wording matters
“What does Example Dental offer?” provides the entity. “Which dentist offers emergency appointments in this city?” does not. A branded answer may help check factual accuracy, but it should not raise an unbranded discovery metric. A mention also needs a separate recommendation label: an answer can name a business while warning against it.
How to reproduce and avoid the mistake
- Label prompt intent before asking the engine.
- Keep one observation per prompt, engine and sample. Remove duplicate retries.
- Record actual coverage. Failed requests and unresolved entity matches remain visible as missing or unknown.
- Report branded recognition, unbranded mentions and unbranded recommendations separately.
- Retain prompts and original answers so another person can recompute the numbers.
Use the blank observation worksheet and the worked denominator guide for a real panel.
Data card and limitations
Selection: deliberately constructed counts. Subjects: none. Personal data: none. Collection window and models: not applicable. Exclusions: none in this example. Analysis: arithmetic proportions only. The example demonstrates a measurement failure mode; it does not estimate its prevalence, establish any brand’s performance, or validate an industry benchmark.
Version 1, September 9, 2026. Corrections can be reported through our contact page. Read the current product methodology before interpreting a VisiScan score, or start a free diagnostic.