Skip to main content

Sentiment Score

Understand how your brand is framed in generative answers.

Written by Apolline Vanneste
Updated over 2 months ago

🎯 What the Sentiment Score captures

In generative search, being mentioned is not enough.

LLMs synthesise multiple sources into a single answer, which can result in mixed narratives: a brand may be praised by one source and criticised by another — sometimes within the same prompt. This effect is especially important at lower-funnel queries, where tone directly influences trust and decision-making.

The Sentiment Score captures how a brand is portrayed within:

  • LLM-generated answers

  • and the sources used to construct those answers

Each detected brand mention is analysed across positive, neutral and negative dimensions, with a score assigned to each.
These classifications are aggregated into a normalized score ranging from -100 to +100, providing a high-level view of brand framing in generative search.


🧮 How the Sentiment Score is calculated

The Sentiment Score is built from individual brand mentions.

For each prompt:

  • brand mentions detected in sources and generated answers are analysed individually

  • each mention receives a distribution of positive, neutral and negative sentiment scores, using LLM-based analysis

Sentiment is then aggregated progressively across different levels (prompt, source, thematic, topic, period), always starting from the scored mention, to avoid bias introduced by top-level aggregation.

The global score ranges from -100 to +100:

  • negative sentiment is subtracted from positive sentiment

  • neutral mentions contribute to a balanced score around zero

Note: The exact implementation may evolve over time as models and APIs change.


🔎 How to read the Sentiment Score

The Sentiment Score should be read as a directional signal, not an absolute judgment.

  • A positive score suggests the brand is generally framed favourably

  • A neutral score often reflects factual or descriptive mentions

  • A negative score indicates recurring critical or unfavourable framing

Sentiment is context-dependent: the same brand can be framed differently depending on the prompt, the source, the engine or the stage of the user journey. This is why the score is most meaningful when analysed across prompts, topics, sources and time periods.


🔗 How this KPI fits with other signals

The Sentiment Score is designed to be read alongside other KPIs, such as Presence Rate and Visibility Score, to understand not only where a brand appears, but how prominently and how it is framed within generative answers.


To go further

Did this answer your question?