Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.getcited.in/llms.txt

Use this file to discover all available pages before exploring further.

Sentiment measures how AI platforms characterize a brand when they mention it — whether the mention is positive, neutral, negative, or mixed in tone. Unlike mention rate or share of voice, sentiment is a qualitative signal that captures what the AI is saying about you, not just whether it is saying anything.

​
Why it matters

A high mention rate with negative sentiment is worse than a lower mention rate with positive sentiment. If the AI consistently describes your brand as “controversial,” “overpriced,” or “declining in quality,” every mention is working against you. A prospect asking ChatGPT “is Brand X worth buying” will receive a characterization that shapes their purchase intent before they ever visit your site. Sentiment also interacts with authority. When AI platforms cite editorial sources that are critical of your brand, the resulting answers inherit that criticism. This makes sentiment particularly sensitive to press coverage, review aggregators, and competitor comparison articles — the same sources that drive citation rate and average position.

​
Formula

Cited computes sentiment as a categorical classification per response, then aggregates: 
sentiment_score = (positive_mentions - negative_mentions) / total_mentions
Inputs:
  • positive_mentions — count of responses where the AI characterized the brand favorably
  • negative_mentions — count of responses where the AI characterized the brand unfavorably
  • total_mentions — all responses mentioning the brand, including neutral and mixed
The score ranges from -1.0 (always negative) to +1.0 (always positive). A score of 0 indicates neutral or balanced coverage.

​
Example calculation

A skincare brand is mentioned in 40 responses. Of those, 28 are positive (“highly rated,” “effective for sensitive skin”), 3 are negative (“expensive relative to alternatives,” “mixed reviews”), and 9 are neutral (“offers a range of products for combination skin”). 
sentiment_score = (28 - 3) / 40 = 0.625
A score of 0.625 indicates strongly positive sentiment — most mentions characterize the brand favorably, with minimal negative coverage.

​
Sentiment categories

Cited classifies each response into one of four categories:
  • Positive — the AI describes the brand in clearly favorable terms (effective, well-rated, reliable, innovative)
  • Neutral — the AI mentions the brand without evaluative language (offers, provides, available in)
  • Negative — the AI describes the brand in clearly unfavorable terms (controversial, overpriced, declining, problematic)
  • Mixed — the AI presents both favorable and unfavorable characterizations in the same response
Mixed sentiment is common for brands with polarizing reviews or ongoing competitive positioning — the AI often surfaces both perspectives in a single answer.

​
Empirical benchmarks not yet available

Cited does not currently publish percentile benchmarks for sentiment across the Cited Index. This is a deliberate decision driven by data quality. The problem. In the March 2026 Cited Index edition, 86.5% of brand-category observations were classified as “mixed” sentiment. Only 13.2% were classified as “positive” and 0.4% as “negative.” This distribution does not support meaningful percentile benchmarks — when 9 out of 10 brands land in the same category, percentiles convey no useful information. Why this happens. The Cited Index uses non-branded category queries, which elicit balanced, multi-brand responses. When an AI lists “the best travel bags in India,” it rarely characterizes any single brand with strong positive or negative language — it provides a menu. Sentiment on this kind of data collapses toward “mixed” for almost everyone. When benchmarks will be added. Sentiment benchmarks will be published in a future Cited Index edition once the classifier produces more signal variation — likely by adjusting to a numeric rather than categorical model, or by weighting query types that elicit stronger characterizations. See the benchmarks methodology for the full list of metrics with and without current benchmarks.

​
What changes this metric

  • Editorial coverage tone — if recent articles about your brand skew positive or negative, AI responses inherit that tone
  • Review aggregator sentiment — AI platforms draw from review sites; shifts in review scores flow through to sentiment over weeks
  • Competitive narrative — comparison articles that cast your brand as “the budget option” or “the premium choice” shape how AI describes you
  • Crisis events — product recalls, PR incidents, and regulatory actions produce measurable sentiment shifts within days of appearing in training data
  • Customer testimonials — prominent case studies and customer stories that LLMs index push sentiment in a positive direction

​
How Cited measures it

Cited parses each AI response for brand mentions and passes the surrounding context to a sentiment classifier that returns one of: positive, neutral, negative, or mixed. The classifier is tuned to consumer brand evaluation language and is calibrated against manually labeled samples. Sentiment is tracked per response, per platform, and per query. The aggregate score for a brand is the proportion of positive minus negative mentions over the total, computed daily for dashboard brands and monthly for Cited Index brands.

​
Frequently asked questions

The Cited Index classifier currently produces ‘mixed’ sentiment for 86.5% of brand-category observations because non-branded category queries tend to elicit balanced, menu-style responses from AI platforms. Percentile benchmarks require meaningful distribution — when nearly every brand lands in the same bucket, percentiles don’t convey useful information. Benchmarks will be added when classifier signal variation improves.
Not necessarily. For most brands in most categories, mixed is the realistic result — AI platforms describe your brand in neutral or balanced terms alongside competitors. Mixed only becomes a concern if you expected strong positive sentiment and are not seeing it. True negative sentiment (the AI explicitly describing your brand unfavorably) is rare but serious when it happens.
Sentiment is driven primarily by the tone of sources the AI is citing. Earn editorial coverage in reputable sources that describe your brand positively, ensure your own website and product pages use clear, confident language about your strengths, and monitor review aggregators for shifts that could flow through to AI responses over time.
Sentiment is the slowest-moving metric Cited tracks. Training data updates for most LLMs happen on multi-week or multi-month cycles, so sentiment shifts lag behind the actual events that caused them. Crisis events (recalls, controversies) show up in sentiment within 2-4 weeks; gradual positioning shifts take 2-3 months to propagate.
Yes. Perplexity tends to produce more evaluative language because it surfaces citations directly, inheriting the tone of the source articles. ChatGPT and Claude tend toward neutral or mixed sentiment because they summarize rather than quote. Gemini is variable, depending on which grounding sources it selects. Platform-specific sentiment is tracked per brand in the Cited dashboard.