← DocumentationReading the results
Stability class: always, contested, never
The per-prompt classification that tells you where the work actually is.
Each prompt, per engine, is classified from its own samples into one of three states.
| Class | Meaning | What to do about it |
|---|---|---|
| Always | The brand appears in effectively every sample | Defend it. Watch for the model update that takes it away. |
| Contested | The brand appears in some samples and not others | This is where effort converts. A contested prompt is one where the engine is choosing between you and someone else. |
| Never | The brand appears in no sample | Either a genuine gap or a mis-chosen prompt. Check the answer before spending on it. |
Why it matters more than the headline number
Aggregate share of voice moves slowly and tells a client where they stand. Stability class tells them where to spend. Roughly three-quarters of prompt–engine cells are near-deterministic, so the contested minority is the entire addressable surface — and it is invisible in any tool that reports a single averaged score.
→Next step
See it against your own client list
A working demo runs your prompts, in your market, on live engines — not a sandbox with seeded data. Bring one client brand and three competitors.