SpyderBot Documentation

Sentiment and brand portrayal

Sentiment describes the tone of substantive representation of a selected entity. It is measured separately for organic, aided, comparative, risk and other declared question groups. A result from a question that explicitly asks for risks must not be presented as spontaneous negative portrayal.

Each represented response–entity receives one overall label: Positive, Neutral, Mixed or Negative, or remains Unclassified. Mixed means meaningful positive and negative portrayal coexist. It does not mean the classifier is uncertain. An answer with no representation of the entity is not a neutral answer.

On this page

Sentiment Distribution · Sentiment Index · Net Sentiment · Sentiment Classification Coverage · Theme and Aspect Coverage · Portrayal Drift · Positioning Phrases, Recurring Claims and Support

Sentiment Distribution

Unit: Percent of classified representations

The balanced proportions of positive, neutral, mixed and negative portrayal among answers that represent the entity and have a usable sentiment classification.

How it is calculated

For each applicable measurement cell, the four label shares are calculated among classified representations and combined using conditional balanced aggregation. These four shares sum to 100% subject to rounding. Unclassified representations are outside this four-segment distribution and remain visible through classification coverage and supporting counts.

How to interpret it

Several mentions within one answer do not become several sentiment votes. A positive share is not the share of all prompts that recommend your brand.

What to do next

Open representative answers and check whether the tone concerns your brand, a named product, a founder or a quoted third party.

Illustrative example

In one fully observed, equally weighted cell with 6 positive, 2 neutral, 1 mixed and 1 negative classified representations, the shares are 60%, 20%, 10% and 10%.

Related concepts: Sentiment Index ↗ · Sentiment Classification Coverage ↗

Measurement scope and aggregation ↗ · Evidence and availability ↗

Sentiment Index

Unit: Index, 0–100

A compact summary of the four-label sentiment distribution in the displayed question group.

How it is calculated

Use the balanced distribution: Sentiment Index = 100 × (positive share + 0.5 × neutral share + 0.5 × mixed share), with shares expressed as fractions. Positive receives full credit, negative receives none, and neutral and mixed receive midpoint credit. The same distribution, question group and evidence support the index and its breakdown.

How to interpret it

The index is a descriptive encoding, not an AI confidence percentage, factual accuracy or customer satisfaction score. Neutral and mixed can produce the same index but describe different answers. Always read the distribution alongside it.

What to do next

Investigate the underlying positive, mixed and negative themes before choosing an action.

Illustrative example

For a distribution of 60% positive, 20% neutral, 10% mixed and 10% negative, the index is 100 × (0.60 + 0.5 × 0.20 + 0.5 × 0.10) = 75.

Related concepts: Sentiment Distribution ↗ · Net Sentiment ↗

Measurement scope and aggregation ↗ · Evidence and availability ↗

Net Sentiment

Unit: Index, −100 to +100

The balance of positive and negative portrayal in the classified sentiment distribution.

How it is calculated

Net Sentiment = positive percentage − negative percentage, using the same balanced distribution and question group. Neutral and mixed remain visible in the distribution but contribute neither positive nor negative net points.

How to interpret it

This is not the 0–100 Sentiment Index and is not Net Promoter Score. A net value of zero can mean neutral portrayal or an equal balance of strongly positive and negative answers.

What to do next

Inspect both tails of the distribution.

Illustrative example

60% positive minus 10% negative gives +50 net sentiment points.

Measurement scope and aggregation ↗ · Evidence and availability ↗

Sentiment Classification Coverage

Unit: Percent of represented answers

How much of the observed brand representation has a usable sentiment classification.

How it is calculated

Within represented answers, measure the classified share and apply conditional balanced aggregation. Unclassified representations stay in the coverage denominator. They do not silently become neutral sentiment.

How to interpret it

No representation means sentiment is not applicable. Representation with no usable classification means sentiment is unavailable; observed classification coverage can be 0%. Coverage is not classification accuracy.

What to do next

Treat incomplete classification as a limit on interpretation and review the evidence status.

Measurement scope and aggregation ↗ · Evidence and availability ↗

Theme and Aspect Coverage

Unit: Percent of represented answers

How often a specified topic or attribute appears in substantive representation of the entity.

How it is calculated

Count a theme at most once per response–entity, divide within the represented-answer scope, and apply conditional balanced aggregation. Multiple themes can coexist, so theme shares need not sum to 100%. The theme vocabulary and version define the comparison.

How to interpret it

Coverage describes discussion frequency, not whether the attribute is true or whether it caused a recommendation. Aspect sentiment, when available, uses its own explicitly labelled aspect-specific scope.

What to do next

Read the supporting passages before writing content that addresses a theme.

Related concepts: Sentiment Distribution ↗

Measurement scope and aggregation ↗ · Evidence and availability ↗

Portrayal Drift

Unit: Component changes

Observed changes in sentiment distribution or theme coverage between comparable measurement periods.

How it is calculated

Like-for-like sentiment components and themes are compared with the named baseline. Percentage-point differences describe the change in individual shares. The component breakdown shows which aspects changed; there is no universal blended drift percentage.

How to interpret it

A change can reflect AI variability, system changes or a scope change. It does not establish that a website edit caused the difference.

What to do next

Check comparability and repeated observations before escalating a portrayal issue.

Related concepts: Metric Change ↗ · Baseline and History Comparability ↗

Measurement scope and aggregation ↗ · Evidence and availability ↗

Positioning Phrases, Recurring Claims and Support

Unit: Distinct supporting responses, prompt/system counts or theme percent

Repeated ways AI describes an entity and the breadth of observed support for that portrayal.

How it is calculated

Exact phrase frequency counts distinct physical response–entity artifacts containing the attributed phrase, not repeated token occurrences. Semantically grouped themes use a declared theme vocabulary and conditional coverage among represented answers. Support badges count distinct Core Prompts and actual systems with supporting evidence, with eligible totals shown; variants and repeats do not inflate prompt counts. A recurring claim is the same attributed claim observed again, not a verified fact.

How to interpret it

A description is not automatically a strength or limitation. Quote attribution, negation and temporal context matter. “3 of 8 prompts; 2 of 4 systems” is evidence breadth, not a confidence score.

What to do next

Read representative evidence and verify material factual claims.

Related concepts: Theme and Aspect Coverage ↗ · Claim Verification Coverage ↗

Measurement scope and aggregation ↗ · Evidence and availability ↗