Why sentiment scores get reviews wrong, and what to ask instead
Sentiment scores get reviews wrong because they squeeze a whole review onto one line from negative to positive. Sarcasm reads as praise, a review that loves the design and hates the strap averages out to neutral, and product words like “this vacuum really sucks” can flip the meaning. The fix is to ask separate questions: what the review is about, whether it is mixed, and how serious the problem is.
What this comes down to
- One score per review hides mixed reviews, which are often the most useful ones because they name exactly what fell short.
- Sarcasm, negation and product vocabulary can all push a score the wrong way.
- How negative a review sounds is not how serious its problem is; judge severity separately from tone.
- Whatever tool you use, read a sample of its labels against the reviews before trusting the totals.
The problem: one line for a whole review
A sentiment score puts each review somewhere between negative and positive. That works for Love it
and for Arrived broken
. It fails on the reviews that matter most, the ones that say several things at once, because a single position on a line cannot hold them.
Five patterns fool a score again and again. The examples are written for this page.
“Great, it broke on day two” is not a positive review.
Five ways reviews fool a score
| Pattern | Example | What goes wrong |
|---|---|---|
| Mixed | Beautiful design. The strap snapped in a week. | Praise and a defect average out to neutral |
| Sarcasm | Great, it broke on day two. | Positive words, negative meaning |
| Product vocabulary | This vacuum really sucks, in the best way. | A negative-sounding word is the product working |
| Comparison | My old one was awful. This one is fine. | Negative words describe a different product |
| Not about the product | Lovely mug, courier threw it at my door. | The complaint belongs to delivery, not the product |
Mixed reviews are the ones worth reading
A mixed review comes from someone who liked enough to be fair and still found a flaw. That flaw is usually specific: the strap, the lid, the sizing. A scoring system that turns those reviews into neutral files your most precise feedback under nothing to see
.
That is why mixed should be its own answer, not a midpoint. The Review Analyzer labels each review positive, neutral, negative or mixed, and counts negative and mixed reviews together when it ranks topics.
Tone is not severity
ABSOLUTELY THE WORST PURCHASE OF MY LIFE, colour is slightly off
is very negative and not very serious. The charger got hot and smelled of burning
is calm and very serious. A score ranks the first as the bigger problem.
So ask about severity separately, and describe the levels by what happened rather than how it was worded. The severity guide gives a five-level scale, with the instruction that capital letters are not severity.
Ask separate questions instead
One score answers one question badly. Four narrow questions answer four well:
- What is it about? One or more topics, from a list you define.
- What is the overall tone? Positive, neutral, negative or mixed.
- How serious is the problem? A severity level based on what happened.
- Does it need a reply? Yes or no, on its own.
No method reads every review correctly, including ours. Read twenty labelled reviews against their text before you trust the totals, and adjust the topic definitions if the same mistake keeps appearing. The glossary entry on sentiment analysis has more on the method.
Questions
Why is sentiment analysis inaccurate for product reviews?
A single score cannot hold a review that praises one thing and criticises another. Sarcasm, product words such as “sucks” for a vacuum, comparisons and complaints about delivery also push scores the wrong way.
What is a mixed review?
A review that is positive about some things and negative about others, such as loving the design and reporting a broken strap. Treat it as its own category rather than averaging it to neutral.
Is a very negative review always more urgent?
No. Tone and severity are different. A calm review about a charger overheating is more serious than an angry review about a slightly different colour, so judge severity by what happened.
Read next
What is customer review analysis? What it finds that star ratings hide
Review analysis turns reviews into counts: topics, sentiment, severity and which need a reply. What it produces, and what a star average cannot tell you.
How to prioritise negative reviews by severity
Not every one-star review is urgent. Rank complaints by severity and how often they recur, so the fix list starts with what actually hurts.
Which customer reviews matter most? It depends on the decision
Safety reviews first, common complaints over loud ones, and the overlooked three-star and specific five-star reviews. Which reviews to read for each decision.