{"id":2821,"date":"2026-09-30T03:23:47","date_gmt":"2026-09-30T03:23:47","guid":{"rendered":"https:\/\/maxaeo.ai\/blog\/llm-sentiment-score-formula\/"},"modified":"2026-09-30T03:23:47","modified_gmt":"2026-09-30T03:23:47","slug":"llm-sentiment-score-formula","status":"publish","type":"post","link":"https:\/\/maxaeo.ai\/blog\/llm-sentiment-score-formula\/","title":{"rendered":"LLM Sentiment Score Formula: A Reproducible Brand Measurement Method"},"content":{"rendered":"

By maxaeo.ai \uff5c Published 2026-09-30 \uff5c Updated 2026-09-30<\/em><\/p>\n

An LLM sentiment score formula<\/strong> should measure how favorably AI-generated answers portray a brand\u2014not merely count positive and negative words. A defensible method scores brand-specific statements, preserves their context, accounts for classification confidence, and reports coverage separately. The result is an auditable metric from -100 to +100 that can be compared over time.<\/p>\n

What Is an LLM Sentiment Score?<\/h2>\n

An LLM sentiment score is a normalized measure of how positively, neutrally, or negatively a brand is presented within a defined sample of AI answers. It evaluates the framing around detected brand mentions. It does not measure visibility, citation frequency, factual accuracy, or the underlying model\u2019s private opinion.<\/p>\n

Use five anchored labels for every eligible brand mention:<\/p>\n

\n\n\n\n\n\n\n\n\n\n
Label<\/th>\nRaw score<\/th>\nNormalized polarity<\/th>\nInterpretation<\/th>\n<\/tr>\n<\/thead>\n
Strongly positive<\/td>\n+2<\/td>\n+1.0<\/td>\nExplicitly recommended or praised<\/td>\n<\/tr>\n
Positive<\/td>\n+1<\/td>\n+0.5<\/td>\nFavorably described without strong endorsement<\/td>\n<\/tr>\n
Neutral<\/td>\n0<\/td>\n0<\/td>\nFactual mention with no clear evaluative framing<\/td>\n<\/tr>\n
Negative<\/td>\n-1<\/td>\n-0.5<\/td>\nMaterial limitation or unfavorable comparison<\/td>\n<\/tr>\n
Strongly negative<\/td>\n-2<\/td>\n-1.0<\/td>\nExplicit warning, rejection, or serious criticism<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n

A statement such as \u201cBrand X is suitable for small teams but lacks enterprise controls\u201d may require a neutral or negative label depending on the prompt\u2019s intended audience. That is why the complete answer and buyer context must remain attached to every score.<\/p>\n

\"LLM<\/figure>\n

What Formula Should You Use?<\/h2>\n

The recommended formula is a confidence-weighted mean of normalized polarity scores:<\/p>\n

LLM Sentiment Score =\n100 \u00d7 \u03a3(w\u1d62 \u00d7 c\u1d62 \u00d7 p\u1d62) \u00f7 \u03a3(w\u1d62 \u00d7 c\u1d62)\n<\/code><\/pre>\n
Where:<\/p>\n
\n\n\n\n\n\n\n\n\n
Variable<\/th>\nMeaning<\/th>\n<\/tr>\n<\/thead>\n
p\u1d62<\/code><\/td>\nNormalized polarity: -1, -0.5, 0, +0.5, or +1<\/td>\n<\/tr>\n
c\u1d62<\/code><\/td>\nClassification confidence or adjudication agreement from 0 to 1<\/td>\n<\/tr>\n
w\u1d62<\/code><\/td>\nPredetermined prompt, engine, market, or buyer-intent weight<\/td>\n<\/tr>\n
i<\/code><\/td>\nOne classified brand-bearing observation<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n
The output ranges from -100 to +100<\/strong>. A positive score means favorable framing outweighed unfavorable framing; zero indicates balanced or predominantly neutral treatment.<\/p>\n
Use equal weights of w\u1d62 = 1<\/code> unless there is a documented business reason to prioritize certain prompts. Weights must be established before reviewing the results, or analysts may unintentionally amplify favorable observations.<\/p>\n
How Do You Calculate the Score Step by Step?<\/h2>\n
A reliable calculation begins with controlled data collection rather than the arithmetic itself.<\/p>\n
    \n
  1. Define the prompt cohort.<\/strong> Include category discovery, alternatives, comparisons, use cases, objections, and purchase-intent questions.<\/li>\n
  2. Fix the test conditions.<\/strong> Record the prompt version, engine, language, country, model mode, date, and retrieval status.<\/li>\n
  3. Capture complete answers.<\/strong> Store the original response, cited sources, brand aliases, recommendation position, and timestamp.<\/li>\n
  4. Extract brand-specific evidence.<\/strong> Identify the relevant statement while retaining enough surrounding text to preserve caveats.<\/li>\n
  5. Apply the five-point rubric.<\/strong> Require a polarity label and a short evidence-based rationale.<\/li>\n
  6. Resolve uncertain labels.<\/strong> Use repeated judge runs, a second classifier, or human review for disagreements.<\/li>\n
  7. Calculate and segment.<\/strong> Report the aggregate score alongside engine-, intent-, market-, and competitor-level results.<\/li>\n<\/ol>\n
    Do not count an absent brand as neutral. Absence belongs in a visibility score<\/a>, while sentiment applies only when enough brand-specific language exists to classify.<\/p>\n
    Worked Example: 12 AI Answer Observations<\/h2>\n
    To test the formula\u2019s interpretability, consider an original synthetic dataset containing 12 brand-bearing answers. Eleven could be classified, while one lacked enough context.<\/p>\n
    \n\n\n\n\n\n\n\n\n\n\n
    Polarity<\/th>\nCount<\/th>\nConfidence<\/th>\nWeighted polarity contribution<\/th>\n<\/tr>\n<\/thead>\n
    +1.0<\/td>\n2<\/td>\n0.90<\/td>\n+1.800<\/td>\n<\/tr>\n
    +0.5<\/td>\n3<\/td>\n0.85<\/td>\n+1.275<\/td>\n<\/tr>\n
    0<\/td>\n3<\/td>\n0.90<\/td>\n0<\/td>\n<\/tr>\n
    -0.5<\/td>\n2<\/td>\n0.80<\/td>\n-0.800<\/td>\n<\/tr>\n
    -1.0<\/td>\n1<\/td>\n0.75<\/td>\n-0.750<\/td>\n<\/tr>\n
    Unclassified<\/td>\n1<\/td>\n\u2014<\/td>\nExcluded<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n
    The weighted numerator is 1.525<\/code>, and the total confidence weight is 9.40<\/code>:<\/p>\n
    LLM Sentiment Score = 100 \u00d7 1.525 \u00f7 9.40 = +16.2\nClassification Coverage = 11 \u00f7 12 \u00d7 100 = 91.7%\n<\/code><\/pre>\n
    A basic net-sentiment calculation would produce (5 positive \u2212 3 negative) \u00f7 11 = +18.2%<\/code>. The confidence-weighted result is lower because uncertain labels have less influence. Reporting +16.2 with 91.7% coverage<\/strong> is more informative than presenting either number alone.<\/p>\n
    \"Example<\/figure>\n
    Which Metrics Should Remain Separate?<\/h2>\n
    Sentiment should not become an opaque composite of every AI search KPI. Keep these measurements separate:<\/p>\n