What Is a Sentiment Score?
What Is a Sentiment Score?
Definition: A sentiment score is a number or label that rates how positive or negative a piece of text is. In AI search, it rates how an answer frames your brand: positive, neutral or negative. Tools calculate it differently, so scores aren’t comparable across tools, but the trend shows which way AI’s framing is moving.
Sentiment scores started in customer service and social listening, where they rate calls, reviews and posts. The same idea now applies to AI answers: instead of rating what customers say, you rate what an AI engine says about you when a buyer asks.
How Is a Sentiment Score Calculated?
Most methods use natural language processing in three steps:
- Break the text into units, such as sentences, phrases or turns in a conversation.
- Classify each unit as positive, negative or neutral.
- Combine the results into a score, a percentage or an average.
There is no single standard for step 3. Three examples from vendor documentation:
| Source | How it scores |
|---|---|
| Amazon Connect | Each speaker turn is classed positive, negative or neutral. Each period of the call gets a score from −5 to +5, and the overall score is the average |
| NiCE | A 0 to 100 scale, where 100 is the most positive outcome and 0 the least |
| AirOps (AI answers) | (Positive mentions + half of neutral mentions) ÷ total brand mentions, as a 0 to 100 score. 60 and above is labeled positive and below 40 negative |
A “good” score on one scale means nothing on another. Compare scores only within the same tool, prompt set and period.
What Is an AI Sentiment Score?
An AI sentiment score rates the tone of what AI engines say about your brand across a set of prompts. Two things are usually scored:
- The answer itself: does the engine describe you positively, neutrally or negatively, and does it recommend you?
- The sources behind it: the pages the engine cites, and how they describe you.
The second matters because AI answers are built from the pages the engine finds. If those pages carry an old complaint or a wrong fact, the answer tends to repeat it.
Sentiment is separate from visibility. A brand can be mentioned often and framed poorly, or mentioned rarely and framed well. Score it on its own line, next to mentions and citations. Factual errors need their own flag, because you fix them differently from opinions.
What Does Framing Look Like in Real AI Answers?
Wellows ran its own AI Brand Report on wellows.com on 20 October 2026, asking ChatGPT in the US six buyer questions six times each (36 answers):
| What the answers did | Count | Share |
|---|---|---|
| Raised a feature gap as an objection | 22 of 36 answers | About 61% |
| Recommended Wellows outright in a verdict | 0 of 27 verdicts | None |
| Cited pages on wellows.com | About a third of citations | 33% |
These are counts of how answers framed the brand, not a sentiment score. They show what a score is built from. Each answer is read, tagged, and turned into a share.
Source: Wellows AI Brand Report, ChatGPT, United States, 20 October 2026. One engine, one country, one day. Answers move, so treat it as a sample.
How Do You Measure an AI Sentiment Score?
- Write a fixed prompt set. Include brand, review, comparison, alternatives and category prompts.
- Run each prompt several times on each engine, from the same country, and save the verbatim answers and cited URLs. Answers vary run to run, so one run proves little.
- Score each answer that mentions you as positive, neutral, negative or mixed, using one written rubric.
- Flag factual errors separately from opinion.
- Calculate a rate per engine: negative rate % = answers that frame you negatively ÷ answers that mention you × 100.
- Track the trend, not a single reading.
For fixing what you find, see how to fix negative brand sentiment in AI.
Does Sentiment Differ by AI Engine?
It can, because engines cite different pages. In a Wellows study of 22.7 million citations (January to June 2026), 79.6% of cited websites appeared on only one of the five engines. If the sources differ, the framing can too, so score each engine separately instead of blending them. Read the full study.
How Does Wellows Handle Sentiment?
Wellows scans ChatGPT, Gemini, Perplexity, Google AI Overviews and Google AI Mode every day. LLM Citations codes each citation for sentiment and shows the complete answer with how your brand is framed, plus the source URLs behind it. On the AI Visibility Platform, sentiment is read from each cited page and shown with each prompt, so a negative source is something you can find and deal with.
Conclusion:
A sentiment score turns the tone of text into a number. For AI search, the useful version rates how answers frame your brand, engine by engine, with the source pages behind each result. Different tools score differently, so compare within one tool and watch the trend. See how AI describes your brand today with the free AI Brand Report.
FAQs
What is a sentiment score?
What is a good sentiment score?
How is sentiment scored in AI answers?
How can I improve my brand's sentiment score in AI search?
What is a good AI visibility score?
Summarise this article
Share this article
About the Author
I'm a Content Strategist at Wellows, an AI visibility platform that helps brands track how they're cited across ChatGPT, Gemini, Perplexity, and Google's AI features. I create the… More about me