Sentiment vs customer satisfaction: what is the difference

Short answer

Customer sentiment and customer satisfaction are not the same thing. Sentiment is the tone of the language used in a conversation: the customer's sentences sound positive, neutral or negative. Satisfaction is the customer's own rating of the service, usually measured with a survey (for example CSAT, the customer satisfaction score). A third, separate thing is whether the problem was resolved, and a fourth is how the customer actually feels inside — which no tool measures directly.

Mixing these four leads to wrong decisions: a customer who starts angry and gets the problem solved can leave with negative sentiment but high satisfaction; a customer who writes politely but whose problem is not solved can leave with positive sentiment and never come back.

Four different measures

1. 1SentimentThe tone of the text or speech. Source: the conversation itself. Shows: how the customer expressed themselves.
2. 2Satisfaction (CSAT)The customer's own rating. Source: a survey. Shows: how the customer rates the service — but only for those who answer.
3. 3ResolutionWhether the problem was actually solved. Source: CRM, later contacts, order status.
4. 4Inner emotionWhat the customer really feels. Source: none — text only shows expression.

How sentiment is calculated

The exact sentiment method varies from tool to tool. As an example, AWS Contact Lens documentation describes it like this: each speaker turn is rated positive, neutral or negative; then two factors — frequency and streaks of the same tone — are used to give each portion of the call a score from −5 to +5, and the overall score is the average of those portions. The documentation presents sentiment as a way to choose which contacts to investigate — not as a measure of satisfaction.

Two conclusions follow: sentiment is based on language and is calculated differently by different tools, so two tools' scores cannot be compared directly.

When sentiment misleads

• Describing a problem: "my card was charged twice" is negative in tone, but the customer is just stating a fact.
• Politeness: many customers write politely even when unhappy — the tone looks positive.
• Irony: "great, late again" can be read as positive from its words.
• Language and spelling: mixed languages, misspellings and short messages weaken tone estimates.
• Topic: customers writing to a complaints service will always sound more negative — that does not mean the service is bad.

When CSAT misleads

Surveys have their own bias. The response rate is usually low, and responders tend to be either very happy or very unhappy. If the survey is sent right at the end of the conversation, the customer may not yet know whether the problem was solved. And the rating is often given for the employee's politeness, not for the resolution.

A satisfaction-adjacent signal from conversations

Without a survey, a conversation does not measure satisfaction, but it gives a few explicit signals close to it: the customer thanks you at the end and confirms the problem is solved, does not come back with the same problem within a week, or, conversely, writes "still nothing has changed". Recording these as a separate field is more precise than overall tone, because it rests on the customer's own words. Still, do not call it "satisfaction" — "confirmed resolution" is a more honest name.

How to show it in reports

1. 1Separate columnsSentiment, CSAT and resolution are shown separately, not merged into one "customer happiness" index.
2. 2CoverageHow many conversations each covers: the CSAT response rate, conversations where sentiment was computed.
3. 3Cross-tabThe most useful view: sentiment × resolution. Negative tone + resolved = good service; positive tone + unresolved = hidden risk.
4. 4TrendChange, not absolute level: does the tone shift while topic and channel stay the same?

Four quadrants

• Negative start, resolved: support's most valuable work; useful as training examples.
• Negative start, unresolved: needs urgent review — churn and negative-review risk.
• Positive tone, resolved: the normal flow.
• Positive tone, unresolved: the most overlooked group — a polite customer leaves with a problem.

The difference between agent quality and sentiment is a separate topic: customer sentiment vs agent performance.

What sentiment can be used for

• Choosing conversations to read: sharply negative tone, or tone worsening during the conversation.
• Trends by topic: is the tone on one topic getting worse over time?
• Channel comparison, carefully: channel audiences differ.

What sentiment should not be used for

• As the sole basis for evaluating, rewarding or penalising an employee.
• For decisions about a customer (service, price, credit).
• To claim "X% of customers are satisfied" — that needs CSAT or another direct measure.

Illustrative example

This is an illustrative example. In an internet provider's monthly report, the share of negative-toned conversations rises and leadership assumes support has got worse. A sentiment × resolution cross-tab shows the rise comes from a line outage in one region: customers write in a negative tone, but most problems are solved within a day or two, and those customers' CSAT does not fall either.

The real risk shows up elsewhere: in the "positive tone, unresolved" group, modem replacement requests — customers wait politely but get no answer for weeks.

Typical mistakes

• Equating negative sentiment with dissatisfaction.
• Merging sentiment, CSAT and resolution into one index.
• Comparing sentiment scores from different tools.
• Drawing conclusions about a customer's inner feelings from text.

Limits

• Text-based sentiment does not see tone of voice or intonation.
• Tone estimates are weak on short and mixed-language messages.
• The accuracy of a sentiment score must be checked separately in your language and domain.

This topic in Vexvon

Vexvon analytics is built around fields: the company sets up fields such as "problem resolved", "complaint type" or "topic of dissatisfaction" itself, and AI picks them from the customer's words as signals — which is more useful for decisions than overall tone. If there is no data for a message-level sentiment measure, the panel's AI assistant says so plainly rather than returning a misleading "0 negative" from an empty table. More: Vexvon analytics.

Next step

If your current report puts sentiment and satisfaction in one number, separate them and add a "resolution" column. Then read 20 conversations from the "positive tone, unresolved" group. For churn signals, see churn signals in conversations; we can build the report together in a demo.

Further reading on this topic: negative customer feedback signals, conversation analytics dashboard metrics.