Glossary
Sentiment scoring
Volume without tone is misleading—pair with Share of Voice.
Use it to catch backlashes before they show up in classic social metrics.
Definition
Sentiment scoring in Getllmspy classifies how a model talks about your brand inside a specific answer: positive (endorsement, recommendation), neutral (factual listing), or negative (risk, criticism, warning). It is scoped to the mention context, not global brand NPS. Two answers can differ sharply if one is a buyer guide and the other is a troubleshooting thread.
How it's computed
Each answer in a run is segmented around your brand string and competitor co-mentions. Classifiers map language cues to labels, then results roll up per model and per prompt pack. Fanout queries reveal whether sentiment flips when the prompt angle changes.
How to read it
Watch deltas week over week. A rising negative share with flat mentions is an early warning—often tied to news, reviews, or a factual error you should fix with grounded content and citations.
Sentiment scoring vs LLM-Score
LLM-Score folds sentiment with presence and correctness. Sentiment alone can look good if you are rarely mentioned—always read the trio together.
When to use
- PR and comms war rooms during incidents.
- Product launches where early reviews skew models.
- Executive reporting when you need plain-language tone trends.