> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mavel.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Sentiment

> A 0-100 index of how AI answers characterize your brand when they mention it, built from LLM scoring of each mention span.

Sentiment measures how favorably an AI answer describes your brand when it mentions you. It is expressed as an index from 0 to 100:

* **0**: maximally negative framing across all mentions
* **50**: neutral or purely factual mentions
* **100**: maximally positive framing

Most brands score between 40 and 80. The number by itself matters less than its trend and how it compares against competitors tracking the same prompts.

## How the score is built

For each captured answer that mentions your brand, Mavel extracts a text window around the mention and sends it to a language model classifier. The classifier returns a score in the range -1 to +1 for that mention span, where -1 is strongly negative ("avoid", "has issues", "outdated") and +1 is strongly positive ("industry leader", "we recommend", "best-in-class").

Those raw scores are stored per mention. The displayed sentiment index converts the average raw score to the 0-100 range: `index = round((average_score + 1) × 50)`. A raw average of 0 (neutral) maps to 50. A raw average of +0.6 maps to 80. A raw average of -0.5 maps to 25.

The final displayed value is the average across all scored mention spans in the selected window and filter scope.

## What moves it

* **The source content AI engines read**: if influential pages in your category describe you neutrally or negatively, that language tends to appear in AI answers. The [Sources](/results/sources) page shows which domains are cited; those are the pages shaping your sentiment.
* **Competitor framing**: answers that position you as an alternative to a leading brand sometimes use comparative language that scores lower than direct recommendations.
* **Answer structure**: some AI surfaces tend toward factual, neutral answers regardless of brand; others more freely express preference. Surface-level differences are expected and visible when you filter by surface.

## How to use it

<Steps>
  <Step title="Compare against competitors">
    A sentiment index of 60 means little in isolation. If your competitors score 75-80 on the same topic, your 60 is a gap: AI engines are describing them more favorably for those prompts. The Top Brands table shows sentiment for every tracked brand side by side.
  </Step>

  <Step title="Watch the trend">
    Sentiment shifts are often the earliest signal of a narrative change. A new critical review or comparison article entering an AI engine's source set can lower your sentiment before it affects your visibility. Check the period-over-period delta.
  </Step>

  <Step title="Filter by topic to isolate the problem">
    A low sentiment on one topic (e.g. "enterprise pricing") while you score well on another (e.g. "ease of use") tells you the narrative problem is specific. Fixing it requires addressing the source content that AI engines read for that topic's prompts.
  </Step>

  <Step title="Cross with Sources">
    Open [Sources](/results/sources) and filter to the same topic. The domains being cited for those prompts are the pages whose language is driving the sentiment score. Those are the pages to monitor, respond to, or counter with better content.
  </Step>
</Steps>

## How to interpret it

**Index 65-80 (green zone)**: predominantly positive framing. Most recommendations mention your brand in a favorable or neutral context.

**Index 40-65 (amber zone)**: mixed framing. The answers mention you but include qualifications, comparisons, or neutral-to-negative language. A source-content review is usually worthwhile.

**Index below 40 (concerning)**: predominantly negative or strongly cautionary framing. This is rare but indicates that AI engines are frequently citing negative sources or positioning your brand as one to avoid.

<Note>
  Sentiment is null (shown as an em dash) when no mentions with scored spans exist for the selected window and filter. Visibility can be high while sentiment is null if the mentions in the window were captured before the scoring pipeline was active, or if the surface does not yet return scorable mention spans.
</Note>

## Relationship to the other metrics

A brand with high [Visibility](/results/visibility) and low sentiment has a different problem than one with low visibility and high sentiment. The first needs content strategy to address the narrative; the second needs content strategy to get mentioned at all. Read sentiment alongside visibility in the same filter scope: one without the other is incomplete.