Sentiment

Which responses sentiment is computed on

Sentiment is computed at the whole answer level, not per individual mention. For each answer the AI model scores several parameters: overall emotional tone, brand relevance, factual confidence, helpfulness.

The metric set depends on the query type. Neutral, comparative and negative queries have different goals, so the scores differ too.

Filters

The same filters as in other analytics sections are available on the left: company, product, AI services, period. Two extra filters specific to sentiment: Query type (neutral, comparative, negative or all) and Sentiment (positive, neutral or negative only).

Four-card summary

Four indicators at the top give an instant overview:

• Overall score on a 0-100 scale with a progress bar. Green, yellow or red color gives an immediate read.
• Total answers for the period plus a split into positive, neutral, negative.
• Positivity: share of positive answers out of total.
• Needs attention: count of problem answers worth reviewing first.

Detailed metrics

A radar chart breaks the overall score down into specific parameters. Each parameter is scored separately and helps see where the brand looks strong and where it falls short.

Universal metrics (for all query types)

• Brand relevance: how relevant the answer is to your brand.
• Factual confidence: model's confidence that the facts in the answer match reality.
• Helpfulness: how useful the answer is for the reader.

For negative queries

• Issue severity: how serious the negativity in the answer is.
• Impact on brand: how heavily this answer can affect reputation.
• Actionability: whether something specific can be done about this answer.

For comparative queries

• Brand preference: which side AI takes in the comparison.
• Recommendation strength: how confidently it recommends or warns.
• Fairness: whether the answer is objective or biased.
• Evidence support: whether the position is backed by arguments and sources.

For neutral queries

• Informativeness: how substantive the answer is.
• Completeness: whether the answer covers the topic fully or only partially.

Sentiment distribution

A donut chart splitting answers into three categories: positive (green), neutral (gray), negative (red). One glance shows the prevailing tone of AI answers about your brand.

Sentiment dynamics

A line chart shows how the average score and tone distribution change day by day. Useful to catch moments when sentiment shifted sharply after a release, news or product change.

Top sources by sentiment

A table with AI services (ChatGPT, Gemini, Perplexity and others), their answer count and average score. Shows where your reputation is best and where there are gaps that need focused work.

Action plan

Below the widgets there's a What to do with this data? block. It's a list of specific recommendations split by priority: critical, important, medium and supportive. Hints are generated automatically based on the most noticeable gaps and strengths.

Answer lists

Three lists with concrete examples at the bottom. Each item opens the full answer text.

Problem answers

Answers with the lowest score or high impact on brand. The first thing to review: understand why AI talks like that and fix the company structure, synonyms or sources.

Best answers

Answers with the highest score and strong recommendation. They hint at the wording and angle that work best for you.

Flagged answers

Answers where the model flagged something unusual: suspicious sources, possible inaccuracies, controversial statements. Useful for manual risk checks.