Sentiment — AI brand perception in depth

The Sentiment page ("AI Brand Perception") goes beyond the single score on Overview: it shows why your sentiment is what it is — which aspects drag it, which providers differ, and what the negative answers actually say. Scores come from Sentiment Model v2, a confidence-weighted, prompt-level aggregation. Sentiment is only ever measured on answers that mention your brand.

The top row

• Sentiment (30d) — 0–100 with a Positive / Neutral / Negative pill.
• Recommendation Rate (30d) — how often answers actively recommend you (marked "low data" when sparse).
• Confidence — how much evaluation data backs the score (evaluation count, average confidence, spread). A high score on low confidence is a weaker claim than a mid score on high confidence.

Aspect breakdown

Sentiment is decomposed by aspect — pricing, trust, quality, support, and so on — each with a signed contribution and the polarity-tagged terms that drove it. This answers "what exactly is dragging my sentiment down": often one aspect (say, pricing uncertainty) is responsible while everything else is fine.

Lowest Sentiment Responses

The actual worst answers, sorted ascending, each with its vs-competitor outcome. Read a few before reacting to any score movement — the raw text is the ground truth, and it frequently reveals a single recurring claim you can correct at the source.

Competitor Impact

What share of your answers mention competitors, and how your sentiment differs with vs. without competitors present. A big gap usually means comparison-style answers frame you unfavorably — a positioning problem, not a reputation problem.

Trend and magnitude

• Daily trend of the 0–100 score.
• Magnitude breakdown — Strong/Weak Positive, Neutral, Weak/Strong Negative — distinguishes "mildly nice everywhere" from "polarized".

Sentiment by Platform

One row per provider with mention count and score. Providers differ systematically; a drop isolated to one platform is a platform story and should be investigated in that provider's answers.

Negative-reason tags

Negative responses are tagged by cause — competitor-favoring language, trust/risk wording, pricing uncertainty, regulatory/compliance mentions — and by intent (Discovery, Comparison, Support, Transactional). The tags turn "sentiment fell" into "comparison answers started citing pricing uncertainty," which is actionable.

Relationship to Prompt Lab's sentiment column

Prompt Lab shows a per-prompt label from the same underlying evaluations (prompts where your brand never appears show neutral — absence is a visibility issue, not a sentiment one). This page aggregates project-wide and adds the diagnostic layers.