Short answer: a 3.5-star average almost never means “an average experience”. On most review sets it means roughly a 55% chance of a good experience and a 30% chance of a bad one, because ratings cluster at the ends rather than the middle. The average is a summary of a distribution, and for review data that distribution is usually J-shaped — a tall bar at 5★, a second bar at 1★, and very little between. Reading the average alone throws away the only information that would actually help you decide.
Why review scores are bimodal, not bell-shaped
Exam marks form a bell curve because everyone sits the exam. Reviews do not, because reviewing is voluntary and motivated. Two states of mind reliably produce a review: delight and anger. An experience that was merely fine produces nothing at all, because nobody feels compelled to write about a parcel that arrived when it said it would.
This is called selection bias, and it means a review page is not a survey of customers. It is a survey of customers who felt strongly enough to type. The silent middle is missing, which is why the 3★ bar is nearly always the shortest one on the chart.
What each average actually implies
| Average | What people assume | What it usually means in practice |
|---|---|---|
| 4.8–5.0 | Outstanding business | Either very few reviews, or a filtered/incentivised set. Genuine businesses at volume rarely exceed about 4.7. |
| 4.2–4.7 | Very good | Genuinely strong. Most customers satisfied, a real minority unhappy. This is the realistic top of the range. |
| 3.8–4.2 | Good | Solid but inconsistent. Usually one recurring failure mode dragging an otherwise fine business down. |
| 3.2–3.8 | Average / mediocre | Polarising, not mediocre. Roughly half delighted, roughly a third genuinely unhappy. Outcome depends on which group you land in. |
| 2.5–3.2 | Poor | A real and repeated operational problem, usually visible as the same complaint again and again. |
| Below 2.5 | Terrible | Either a genuinely failing service, or a review-bombing incident. Check the dates before concluding. |
The three numbers that beat the average
- The negative share. What percentage of reviews are 1★ or 2★? This is the closest thing to your risk of a bad outcome, and it is far more actionable than the mean. A 30% negative share means roughly one customer in three had a bad time.
- The count. An average from 12 reviews is noise; one new rating can move it by a tenth of a star. Under about 30 reviews, treat the number as decorative. The often-quoted “0.1 star difference” between two brands is almost always smaller than the sampling error.
- The recency. A three-year-old score describes a business that may no longer exist in the same form. Read the most recent reviews separately, because management, ownership and logistics partners all change.
Why a 0.1-star gap between two brands means nothing
If two businesses score 3.52 and 3.44, it is tempting to call the first one better. Usually you cannot. With a few dozen to a few hundred reviews, the uncertainty around each average is comfortably wider than that gap, so the ordering could flip with a handful of new ratings. Ranked lists that present such gaps as a verdict are reporting noise as though it were signal.
A difference starts to be worth anything when it is large (roughly half a star or more), when both sides have a few hundred reviews, and when the distributions differ in shape and not just in mean.
How to read a review page properly, in order
- Look at the distribution bars first, before the average. The shape tells you whether the business is consistent or a coin flip.
- Read the 1★ reviews and look for repetition. One recurring specific complaint is a real defect. Twenty unrelated grumbles are ordinary noise.
- Read the 3★ reviews. They are the rarest and often the most honest, because the writer had no strong incentive and is weighing things up.
- Filter to the last few months. You are buying from today’s business, not the one that earned the score.
- Check whether the complaints even apply to you. Most negative reviews describe a failure mode that is specific — a particular city, plan, or product line. If it is not your use case, it is not your risk.
Frequently asked questions
Is 3.5 stars a good rating?
It is a mixed rating, but it does not mean a mediocre experience. Because review distributions are bimodal, a 3.5 typically means around 55% of reviewers were positive and around 30% were clearly negative, with a thin middle. In other words it describes a business that is inconsistent rather than uniformly average, and your outcome depends on which group you fall into. Read the one-star reviews to find out what causes the bad outcomes and whether that failure mode applies to you.
Why do review scores cluster at 1 and 5 stars?
Because writing a review is voluntary and takes effort, so only strong emotion produces one. Delight and anger both motivate people to type; an experience that went exactly as expected motivates nobody. The result is selection bias: the review page captures the two tails of customer experience and misses the satisfied, unremarkable middle. This is why the three-star bar is usually the shortest on any review chart.
How many reviews make a rating trustworthy?
Below roughly thirty reviews an average is unstable, because a single additional rating can shift it noticeably. Between thirty and a hundred the shape of the distribution becomes meaningful even if the exact average is still soft. Above a few hundred the average is statistically stable, though you should still check recency, since a large old sample can conceal a recent decline in service.
Is a 4.9-star business better than a 4.3-star one?
Not reliably, and the higher score can be the bigger warning. Businesses serving large numbers of real customers very rarely average above about 4.7, because genuine volume always includes people who had an ordinary or bad day. A 4.9 usually indicates either a small number of reviews or a set that has been filtered, incentivised or solicited selectively. A 4.3 built on several thousand reviews is normally the more credible signal.
Should I trust the average or read the reviews?
Use the average only to decide whether the listing is worth your attention, then read. The average compresses everything useful into one number and discards the distribution, the recency and the reasons. Two businesses can share an identical 3.6 where one is consistently unremarkable and the other alternates between excellent and awful — and those are completely different purchases.
What is a J-shaped distribution in reviews?
It is the characteristic shape of voluntary review data: a large spike at five stars, a smaller but substantial spike at one star, and a shallow dip through two, three and four. Plotted from left to right it resembles the letter J. It appears because reviewing is self-selected rather than sampled, and recognising it is what stops you from misreading a polarised business as an average one.
Comments (0)