Define the measurement decision
Separate polarity from issue taxonomy. One answers how the brand is framed; the other explains what the answer is about.
Measurement method
| Step | Action |
|---|---|
| 1 | Define label rules and examples before scoring snapshots |
| 2 | Keep Neutral as a valid state rather than treating it as missing or non-negative |
| 3 | Add issue tags only when the answer text supports them |
| 4 | Review borderline and mixed answers with the original prompt and context |
| 5 | Report Sentiment by platform and Prompt Group alongside sample size |
Worked calculation or observation
An answer may recommend a product for small teams while warning that enterprise controls are unclear. A single positive label loses the limitation; a polarity plus issue tag preserves both signals.
Evidence to retain for AI search Sentiment
Keep these fields with the decision:
- metric name
- numerator
- denominator
- scope
- platform
- Prompt Group
- success state
- observation date
- formula version
Limits and UnderAI's role in AI search Sentiment
Sentiment classification is an interpretation of observed text, not a measurement of customer emotion or brand reputation across the whole market.
UnderAI exposes a per-snapshot Positive, Neutral, or Negative Sentiment state and the underlying answer evidence. Public copy should use the verified Sentiment field consistently.