What is sentiment analysis?
Sentiment analysis reads a piece of written feedback and labels it negative, neutral or positive. Applied to every comment in a survey, it shows the mood of all the answers at once and how that mood changes from one period to the next.
Reading the tone of every answer
A survey with an open question can collect hundreds of comments a week, more than anyone can read closely. Sentiment analysis gives each comment a label for its tone, so you can count how many are negative, see whether that share is rising, and read the negative ones first.
Modern tools use language models that read whole sentences in context and in many languages. Older methods counted positive and negative words from a list, which missed sarcasm and negation, such as “not bad at all”.
Sentiment is most useful beside a theme. Knowing that a third of comments are negative helps less than knowing that most negative comments are about delivery, which is where thematic analysis comes in.
- Labels
- Negative, neutral and positive, sometimes on a finer scale
- Reads
- Written answers, reviews, support tickets and chat messages
- Used for
- Spotting changes in mood and reading unhappy comments first
- Pairs with
- Themes, which say what each comment is about
- Also called
- Opinion mining, tone analysis
How sentiment analysis works
Each comment goes to a model that reads it in context and returns a label, often with a level of confidence. A comment can carry more than one feeling, such as “Love the app, hate the new pricing”, so some tools read the tone of each part or each topic on its own.
The labels are then counted. A common measure is the share of negative comments, and another is net sentiment: the share of positive comments minus the share of negative ones.
Net sentiment = % positive − % negative
It runs from −100, when every comment is negative, to 100, when every comment is positive. Neutral comments count toward the total but add to neither side.
Sentiment analysis examples
A few comments from one delivery survey, and how each one reads. The last two show why reading in context matters: word by word, they would get the opposite label.
| Comment | Sentiment | Why |
|---|---|---|
| “Delivery came a day early, great service” | Positive | Clear praise |
| “The parcel arrived on Tuesday” | Neutral | A fact without a feeling |
| “Second time my order was late” | Negative | A repeated problem |
| “Not bad at all for the price” | Positive | A negation that reads as praise |
| “Great, another password reset email” | Negative | Sarcasm |
A net sentiment example
Say 400 customers write a comment in a month. 120 read as positive, 188 as neutral and 92 as negative, so positive comments make up 30% and negative ones 23%.
The number matters most as a trend. If net sentiment falls from 7 to −5 the next month, open the negative comments and look for the theme that grew.
- Positive
- 120 comments, 30%
- Neutral
- 188 comments, 47%
- Negative
- 92 comments, 23%
- Net sentiment
- 30 − 23 = 7
Limits of sentiment analysis
Sentiment is a reading of tone, and tone is easy to misread. Short answers, mixed feelings and words that mean something else in a given trade all make it harder.
Check a sample of labels by hand now and then, and correct the ones that are wrong, so the counts you report stay trustworthy.
- Sarcasm and irony, which read as praise word by word
- Mixed comments that praise one thing and criticize another
- Very short answers such as “ok” or “fine”, which carry little tone
- Slang, such as a “killer feature” or a “sick” new design
- Polite complaints, where a serious problem is described calmly
How Usertick reads sentiment
On every plan, Usertick's AI reads each response with something written as it arrives, in any language and without drawing on an allowance. Each written answer gets a sentiment, negative, neutral or positive, along with a category, a theme and a type, and each response gets an urgency.
On a text question's card in Results, every answer shows its sentiment and theme, and the answers filter by both. The Signals strip counts sentiment for the date range against the period before, and the count opens those responses in the inbox, where sentiment is also a filter.
On every plan, a workflow can start after the AI reading and check the sentiment, so negative answers post to Slack or Microsoft Teams or reach the right person. Anyone with permission to manage responses can correct a sentiment, and the correction wins everywhere, from the inbox to workflows.
Mistakes to avoid with sentiment analysis
Small choices that keep the results useful.
- 01
Reporting sentiment without themes
A drop in sentiment gives no direction on its own. Break the negative answers down by theme to find the cause.
- 02
Trusting every label
Sarcasm and mixed comments get misread. Read a sample each month and correct what is wrong.
- 03
Treating neutral as satisfied
A neutral comment often states a fact or a mild problem. Read the neutral answers that come with low scores.
- 04
Comparing periods with few comments
With 20 comments, a few angry ones can swing the share by many points. Check how many answers sit behind a change.
Frequently asked questions
It depends on the model and the text. Clear praise and clear complaints read well, while sarcasm, mixed comments and very short answers are harder, so check a sample by hand and correct mistakes.
No. CSAT comes from a rating the customer gives, and sentiment is read from the words they write. The two often agree, and the answers where they don't are worth reading.
Usertick labels each written answer negative, neutral or positive and counts the labels for any date range against the period before. It doesn't combine them into one score, so compute net sentiment from the counts if you need it.
Modern models read most languages directly. Usertick reads written answers in any language and labels them without translating the responses.
On every plan, an alert can fire when urgent responses, or responses of a type such as complaints, rise above a line you set. A workflow can also post each negative answer to Slack or Microsoft Teams as it arrives.
Listen better, act faster, and build what truly matters.
Start free, see what your customers mean, and get each answer to the person who can act on it.
- 150 responses a month on Free
- Every channel and AI analysis
- No card needed