What Is a Sentiment Score? Formula, Scale & How to Track It

Quick Summary

A sentiment score is a number that shows how positive, negative or neutral customer feedback is. It’s a fast way to track how customers feel at scale, but easy to misread, because every tool uses its own scale and formula:

• Single comments are usually scored from -1 to +1, with 0 as neutral.
• Net sentiment rolls thousands of comments into one number from -100 to +100.
• Aspect-level scoring shows what customers are happy or unhappy about.
• A good score depends on the channel, so benchmark against your own history.
• Sarcasm, mixed reviews and non-English feedback can all skew the number.

The number on your dashboard is only half the story

Your dashboard says customer sentiment is +0.4. Is that good? Leadership will ask, and the honest answer is that it depends on the tool.

Google’s Natural Language API scores text from -1.0 to 1.0. Qualtrics scores survey comments from -10 to +10. Amazon Connect scores each portion of a call from -5 to +5. The same +0.4 is fairly positive on Google’s scale and close to neutral on the other two.

Scale is only the start. It also matters whether the tool scored the whole review or each topic inside it, and how many comments sit behind the number. A healthy average can hide a delivery problem that grows every week. Here’s how a sentiment score is produced, how to read it, and where it falls short.

Why Listen to Us?

Scoring sentiment by theme is what Chattermill does every day for enterprise CX, VoC and product teams. Our customer experience intelligence platform analyzes surveys, reviews, support tickets, social posts and call transcripts together, so teams see what’s driving a score, not just the score itself. Here’s what that has delivered for our customers:

• Uber has worked with Chattermill for over eight years, with 400+ users on the platform.
• E.ON Next grew its billing NPS from +9 to +22, a 144% increase.
• HelloFresh lifted average order value by 30%.
• Footasylum cut contacts per transaction by 42.8%.

Read more in our customer stories.

What is a sentiment score?

A sentiment score is a number that measures how positive, negative or neutral a piece of text is. Sentiment analysis software reads a review, survey comment, support ticket or social post and places its tone on a fixed scale, often -1 to +1, where 0 is neutral. Higher scores mean more positive feeling, and lower scores mean more negative feeling.

You’ll also see it called a sentiment analysis score, though the exact sentiment score meaning varies by tool.

How is a sentiment score calculated?

Three choices shape the number a tool gives you.

The scale and the net sentiment formula

Google’s API and VADER, a free open-source scoring tool, both use -1 to +1. VADER treats anything at +0.05 or above as positive and anything at -0.05 or below as negative.

To roll thousands of comments into one number, most teams use net sentiment, which works much like NPS:

Net sentiment = (positive mentions − negative mentions) ÷ total mentions × 100

Here’s how to calculate a sentiment score for a whole data set. Say you have 2,000 mentions: 1,100 positive, 500 negative and 400 neutral. That’s (1,100 − 500) ÷ 2,000 × 100, which gives you +30.

Some tools leave neutral mentions out of the total. Talkwalker divides by positive plus negative mentions only, so the same data would score +38 there. Check which version a tool uses before comparing platforms.

Polarity and magnitude

Polarity is the direction of a feeling, positive or negative. Magnitude is how strong it is. Google’s API reports both, and its magnitude tends to grow with longer text.

You need both for mixed feedback, because praise and complaints cancel out. In Google’s own examples, a neutral document scores 0.1 with a magnitude of 0.0, and a mixed one scores 0.0 with a magnitude of 4.0. Nearly the same score, very different feedback.

Document-level vs aspect-level scoring

Document-level scoring gives a whole comment one number. Aspect-level scoring, or aspect-based sentiment analysis, identifies each topic a customer mentions, such as delivery, price or support, and scores the sentiment toward each one. Document-level tells you whether customers are happy. Aspect-level tells you what to fix.

Chattermill assigns sentiment to each theme in a comment, then calculates net sentiment from those theme scores.

One review, scored two ways

Take this review from an online shopper:

“The jacket fits perfectly and the quality is great, but delivery took nine days and nobody answered my emails.”

Document level: VADER scores the whole review +0.63. So your dashboard logs a happy customer. Neither complaint registers, because “delivery took nine days” and “nobody answered my emails” contain no negative words. VADER scores both at 0.0.

Aspect level: The same review splits into four themes.

Two positives and two negatives give the review a net score of 0, plus two problems your team can act on. Across 10,000 reviews, your average can stay positive while delivery complaints pile up underneath it.

How to read a sentiment score alongside volume and theme

A score alone can’t tell you how much to care. A theme scoring -80 from 12 mentions looks alarming, but one scoring -30 from 4,000 mentions affects far more customers. So read every score next to two things:

• Volume: how many mentions sit behind the score
• Theme: what those mentions are about

When sentiment on a theme falls while its volume rises, that’s usually the first thing to fix. Our guide to customer sentiment analysis covers the wider workflow.

Reading by theme is how Mindful Chef spotted that families rated portion sizes lower than smaller households. After increasing family portions, sentiment for that segment rose from 38 to 57.

With Lyra Agent, now in early access, you can ask why sentiment moved on any theme and get answers backed by verifiable numbers and real customer quotes.

Sentiment score benchmarks by channel

A good sentiment score depends on the channel, because some lean far more positive than others. Public benchmarks are scarce, but here’s how each one skews.

The safest benchmark is each channel’s own history, so be careful blending channels into one company-wide score. A wave of happy reviews can hide a support inbox that’s getting worse.

The limits of a sentiment score on its own

Three kinds of feedback catch most models out.

Sarcasm

“Great, another late delivery” uses a positive word to say something negative, and VADER scores it +0.62. In research on sarcastic tweets, correctly detecting sarcasm improved sentiment detection by nearly 50 percentage points.

Mixed reviews

One number can’t hold two opinions. In a study of misclassified game reviews, reviews weighing pros against cons with words like “but” and “although” were the biggest single cause of errors, at 30%.

Language

Accuracy often drops outside English. When researchers tested ChatGPT on one sentiment task, it scored 84% in English but 56% in Buginese, a regional language of Indonesia.

So when a score surprises you, read the comments behind it before you act.

Start scoring sentiment by theme

A sentiment score is a fast read on how customers feel, as long as you know the scale, the formula and what got scored. Track net sentiment for each theme, watch volume alongside it and dig in wherever sentiment falls while mentions climb.

To see sentiment by theme across every channel in one place, book a demo.

FAQs

What is a good sentiment score?

It depends on the scale. On -1 to +1, anything at +0.05 or above usually counts as positive. For net sentiment, anything above 0 means positive mentions outnumber negative ones.

Can ChatGPT do sentiment analysis?

Yes, and it handles simple positive-or-negative calls well. In a 2024 study, GPT-4 averaged 93% accuracy on positive versus negative with no extra training, and accuracy fell once a neutral option was added. To track thousands of comments by theme over time, you’ll want a dedicated sentiment analysis tool.

How do you measure customer sentiment?

Collect feedback from every channel, score each comment by theme, then track net sentiment and volume over time. Our guide to analyzing customer sentiment with AI walks through the full process.

What is sentiment analysis in simple terms?

Sentiment analysis is software that reads text and decides whether the writer feels positive, negative or neutral. The sentiment score is its output.

What does sentiment mean in simple terms?

Sentiment is the feeling or opinion behind what someone says. In customer feedback, it’s whether a customer sounds pleased, frustrated or indifferent about your brand, product or service.