What "AI sentiment" measures
AI sentiment is the tone an AI assistant takes toward your brand in an answer — positive, neutral, or negative. It is NOT social-media sentiment; it is the tone of the AI answers themselves.
How it is graded
Every answer gets a fast heuristic read, and answers that are uncertain or borderline are re-checked by a calibrated LLM judge at temperature 0. When the judge disagrees with the heuristic or is low-confidence, the answer is held for human review rather than trusted blindly.
Directional vs. calibrated
We report your sentiment as directional until the judge's agreement with independent human review (Cohen's kappa) clears our bar over a large enough sample. Once it does, we surface it as calibrated with that agreement shown — the same way we show a confidence interval on accuracy. We would rather say "directional" than over-claim.
How you audit it
On AI Monitor → Audit how AI describes your tone (Pro and Custom plans), we show you the answers we flagged on tone. You confirm or correct each as positive, neutral, or negative. Your call corrects your own sentiment for that answer, and feeds a pooled quality signal — but it does not certify the measurement itself, which stays calibrated separately across the platform by trained reviewers on an unbiased sample. This keeps the "calibrated" claim honest: the party being measured never grades the instrument that measures them.
Why this matters
When AI describes you negatively, the fix is not to hide it — it is to counter it with evidence. Auditing tone tells us what is really negative so we can help you publish a corrective, evidence-backed answer and then re-measure whether it moved.