Sentiment

Sentiment scores the tone of the language surrounding a brand mention: is the engine recommending you, listing you neutrally, or hedging about you.

Off by default

Classifying tone is a second inference pass over every captured answer, and it is a real cost line rather than a rounding error. It is switched on per brand, when someone has a reason to want it, rather than billed to everyone by default.

It is the most stable thing we measure

Counter-intuitively, tone is *more* consistent across models and repeats than presence is. Whether you get mentioned fluctuates; how you get described, once mentioned, is fairly stable. That makes sentiment a better early-warning signal than a headline metric — a tone shift is usually real, where a visibility wobble is usually sampling.

Reading it honestly

• Sentiment is only defined on runs where the brand was mentioned. A brand mentioned twice with glowing tone has a wonderful sentiment score and a visibility problem.
• Neutral is the modal outcome and is not a failure — most AI answers list vendors rather than praising them.
• A negative shift with a stable source set usually means a model update. A negative shift with a changed source set usually means something got published.