What is sentiment in AI answers?
Sentiment in AI answers is whether a brand mention is favorable, neutral, mixed, or negative. It is a reading of the wording in that response, not a score the model publishes. The same brand can be praised for one use case and cautioned for another in a single answer.
Why it matters for AI visibility
Visibility without regard to wording can celebrate a warning. A prompt that says your tool is unreliable is not a win just because your name appeared. Sentiment matters when you report share of voice or a list of mentions, because buyers act on the recommendation, not the name drop. Keep the judgment simple and repeatable: positive when the answer recommends you for the stated need, negative when it advises against you, mixed when it does both, and neutral when it only identifies you. Quote the sentence you classified. Do not average sentiment into a false-precision index, and do not assume a negative line came from a particular review unless the answer cites it. When the wording is wrong, fix the underlying fact on the pages and sources the engine can retrieve. Re-read the same prompt later. One kinder answer is not a trend.
Example
A prompt about alternatives says your product is a strong fit for agencies and a poor fit for enterprises that need a named compliance feature. Classifying the whole answer as positive would hide the caveat. Mark it mixed, and keep the sentence next to the label.
Related terms
Brand mentions in LLMs are what you are classifying. Share of voice counts those mentions and should not hide a negative cluster inside a healthy-looking percentage. Hallucination is a separate failure: a sentiment label does not tell you whether the claim is true.