A negative review rate is the share of 1β2 star reviews in a product's written reviews. That is not the star rating you see on the listing. Across our 22 reports, five products sit at exactly 3.7 stars, and their negative review rates run from 10.9% to 26.8% β a 2.5x spread under an identical headline number.
Everyone in Amazon selling uses the phrase "negative review rate." Very few people actually calculate one. What usually happens is a glance at the star rating: four stars feels fine, three stars feels dangerous, and the decision gets made right there.
Those two numbers are not close.
One comparison from our library makes the point faster than any definition. A charger pair at 3.6 stars carries a 31.2% negative review rate. A translation earbud pair at 3.9 stars carries 14.0%. A third of a star apart, more than double the risk. On a listing page these two look almost identical.
So this piece is about the metric itself: how it is calculated, where it diverges from the star rating, what counts as normal, and what it cannot tell you.
1. How to Calculate a Negative Review Rate
One line: 1β2 star reviews divided by total written reviews.
Three parts of that line get botched routinely.
The numerator is 1 and 2 stars only. 3 stars does not count. Definitions vary here. Some tools fold 3-star reviews into the negative bucket. We keep them out, because on Amazon a 3-star review usually still carries "it works, but" β the buyer is hesitating, not rejecting. Folding 3 stars in inflates every rate by a wide margin and destroys comparability between categories.
The denominator is written reviews, not all ratings. This is where the negative review rate and the star rating part ways permanently. That 4.3 on the listing has every person who ever tapped a star in its denominator, and a large share of them never typed a word. You cannot see what they said, and you cannot see what they were unhappy about. The negative review rate only counts the people who wrote something β the ones you can read, categorize, and hand to an engineer.
Those are not the same population. People who bother to type are the ones with the strongest experience, good or bad. The middle of the distribution walks away silently. That gives the rate a built-in amplification: it describes the extremes, not the average buyer.
The third trap is the word "total." One number, three denominators, three different percentages that are all correct. A single pain point mentioned 36 times is 52.2% of all pain mentions for that product, 67.9% of its negative reviews, and 19.3% of its total reviews. Three answers to three questions: how dominant is this complaint among complaints, how many unhappy buyers raised it, and how many buyers overall ran into it.
The negative review rate only travels in the last direction, with a denominator slightly narrower than "all reviews" β written reviews only.
A related confusion worth naming: a pain-point group share is not a negative review rate. The group share slices inside the complaints. The rate slices across the buyer population. Write 52.2% into a listing analysis as if it were a negative review rate and every downstream conclusion flips.
Here is the arithmetic on a real product. Anker's power bank in our report has 187 written reviews, 53 of them at 1β2 stars. 53 Γ· 187 = 28.3%, which is the number printed on the report page. UGREEN in the same report is 50 Γ· 201 = 24.8%. Put those two side by side and the question "which of this pair is steadier" becomes answerable. A lone 28.3% answers nothing.
2. Why the star rating barely moves
There is a simple reason, and it is arithmetic rather than opinion.
A star rating is a weighted average. Around 4.0, a 1-star and a 5-star differ by 4 points of contribution, so holding the average steady means one 1-star needs three 5-stars to cancel it. The equation is 5x + y = 4(x + y), which solves to x = 3y. At 4.5 it gets worse: one 1-star needs seven 5-stars.
The closer a product sits to the top of the scale, the more efficiently bad news gets buried.
Two distributions make it concrete. Both are realistic Amazon shapes:
| Distribution | 5β | 4β | 3β | 2β | 1β | Average | 1β2β share |
|---|---|---|---|---|---|---|---|
| Shape A (smooth) | 52% | 28% | 14% | 3% | 3% | 4.23 | 6% |
| Shape B (polarized) | 78% | 0% | 0% | 10% | 12% | 4.22 | 22% |
An average gap of 0.01. A negative review rate gap of 3.7x.
Shape B is not exotic. "Lots of five stars, a real tail of one stars, nothing in the middle" is common in budget electronics and pet gear, where buyers either feel they got a bargain or feel they got cheated. That product shows a handsome 4.2 on the listing while more than one buyer in ten is writing something angry.
The average is what remains after the extremes cancel out. The rate measures only the losing side. One tells you how the books look. The other tells you the size of the actual loss. Product decisions belong to the second number, because product changes act on the loss side directly β complaints move when you fix things, praise does not necessarily follow.
3. Five reports at 3.7 stars, 2.5x apart
Sort the library by average rating and the ties do the talking.
| Average | Reports | Rate range | High / low |
|---|---|---|---|
| 3.8 | 2 | 18.3% β 26.3% | 1.44x |
| 3.7 | 5 | 10.9% β 26.8% | 2.46x |
| 3.6 | 2 | 17.8% β 31.2% | 1.75x |
| 3.5 | 3 | 22.1% β 31.1% | 1.41x |
| 3.3 | 3 | 28.0% β 35.5% | 1.27x |
| 3.2 | 2 | 29.5% β 34.9% | 1.18x |
The 3.7 row is the clearest: wireless earbuds at 10.9% on 514 reviews, a cat tree at 18.4% on 380, two night lights at 24.5% and 25.0% on 408 and 329, and a power bank at 26.8% on 388.
All five average 3.7 in our data. Filter a category by star rating and these five land in the same bucket. What differs is the actual level of dissatisfaction by a factor of 2.5. In the earbud report, 56 of 514 reviews are 1β2 stars. In the power bank report, 104 of 388 are.
That gap is not sampling noise. Every sample clears 300 reviews and our 100-review floor, all five were pulled inside the same window, and the method is identical.
A metric that cannot separate 10.9% from 26.8% at a fixed value is not doing much work in the 3.5 to 4.0 band. Which is exactly where a large number of Amazon products live, especially older listings whose average has been anchored by years of accumulated positive reviews and cannot move.
4. Cross-rating comparisons are worse
If tied averages can be dismissed as buckets that are too coarse, comparisons across different averages close that escape route.
A 3.6-star charger pair (UGREEN vs INIU, 325 reviews) runs 31.2%. A 3.9-star translation earbud pair (Csasan vs Soundcore, 522 reviews) runs 14.0%. Three-tenths of a star apart, 2.2x the risk.
Another: wireless earbuds at 3.7 stars come in at 10.9%, lower than a pet grooming tool at 4.3 stars and 12.9%. Six-tenths of a star lower, less risk.
And a third: USB-C cables at 4.1 stars sit at 18.1%, a cat brush at 3.8 stars sits at 18.3%. Three-tenths of a star apart, two-tenths of a point apart on the rate.
Put those together and the conclusion is unavoidable. A higher average does not mean lower risk, and a lower average does not mean higher risk. Across the 3.5 to 4.3 range these two numbers behave almost like independent variables.
The reason is category behavior. Buyers score pet products generously and charging products harshly; the same defect draws different stars on either side. The rate is also affected by that, but it measures the share of people complaining rather than how severely they complain, which travels a little better across categories.
Which is why comparing averages across categories tells you almost nothing, while comparing rates at least tells you something.
5. What counts as normal
Combining multiple reports in one category is done by weighting on review count, not by simple averaging. Electric toothbrushes: 437 reviews at 17.8% and 420 reviews at 22.1%. Simple average 19.95%, weighted (437Γ17.8 + 420Γ22.1) Γ· 857 = 19.9%. The samples are similar in size, so the difference is negligible. Power banks: four products average 31.05% simply, 31.2% weighted, again close. But put a 60-review product next to a 500-review product and the simple average lets the small sample drag the whole category. The convention is weighted from the start.
Across 16 categories and 22 reports, negative review rates run from 10.9% to 47.5%, with a median of 26.4%. Four bands:
| Band | Categories | Reading |
|---|---|---|
| Under 15% | Wireless earbuds 10.9%, translation earbuds 14.0%, USB cables 14.1% | Either a simple product or forgiving buyers |
| 15% β 20% | Pet grooming 16.1%, USB-C cables 18.1%, cat trees 18.4%, electric toothbrushes 19.9% | Low-ish; worth knowing why |
| 20% β 30% | Night lights 24.7%, shavers 28.0%, chargers 28.8%, pet fountains 29.5% | The normal range |
| Above 30% | Wired earbuds 31.1%, power banks 31.2%, neck massagers 32.3%, smart pet feeders 33.7%, self-cleaning litter boxes 47.5% | High risk, for different reasons |
Three rules of thumb:
Under 15% is healthy, but find out why. It could be a genuinely simple product (cables), vague buyer expectations (translation earbuds), or a category where nobody does well enough for anyone to complain. The right response differs for each.
20% to 30% is the Amazon baseline. There is room to improve, but a low rate will not win the category on its own.
Above 30% β do not decide yet. Check the internal spread first. Self-cleaning litter boxes at 47.5% with a 1.29x spread means nobody has solved it; chargers at 28.8% with a 2.15x spread means somebody has. Same high risk, opposite conclusions.
None of these bands are absolute. Sixteen categories is what we have analyzed; Amazon has tens of thousands. If your category is not in the list, treat the bands as a reference frame, not a verdict.
6. Three ways people get the number wrong
Using the listing's rating distribution as the denominator. Amazon shows five bars for every product, and it is tempting to read "1-star plus 2-star share" straight off them. That denominator includes everyone who rated without writing, people whose reasoning is invisible to you. The two figures will not match, and which one comes out higher depends on how talkative that product's buyers happen to be. Substituting the distribution for the rate quietly counts a crowd of silent buyers into your analysis.
Treating a pain-point share as a negative review rate. Section 1 covered the 52.2% versus 19.3% pair. The high number means the complaint is concentrated, the low one means the overall product is holding up. Both are useful together and misleading apart.
Letting one product stand in for the category. Power banks appear in two reports across four products: 24.8%, 28.3%, 33.2%, 37.9% β a 1.53x best-to-worst gap, with two of those being models from the same brand that still differ by 1.34x. Grab one at random and your number describes that product, not the category.
A variant of the same mistake: computing a rate off a few dozen reviews. Seventeen negatives out of 61 is 27.9%, which sounds alarming and has a confidence interval wide enough to be useless for a decision. Our floor is 100 written reviews per product; below that you can read direction, not draw conclusions.
7. What the number is good for
As a floor for risk. The rate tells you how bad the category gets if you do nothing special and perform at the average. That is your cost ceiling β returns, complaints, support tickets, listing health all derive from it.
As a first filter, not a final answer. Above 30%, be careful. Under 15%, be comfortable. In between, you need the second dimension.
Paired with internal spread. The rate is an absolute level; the spread between products in a category is relative. Cross them and you learn whether anyone has solved the category. At 47.5%, a 2.5x spread means an opportunity because somebody got to 19%, while a 1.29x spread means a ceiling. On a ranking table those two are indistinguishable.
If you keep one sentence: the rate answers how hard the category is, the internal spread answers whether anyone has solved it, and you need both.
8. It is not a constant
A negative review rate is computed inside a time window. Change the window and the number moves.
Cat trees are the clean demonstration. Across the full period the two products read 18.9% and 17.9% β essentially tied at 1.06x. Over the last three months that becomes 17.9% and 7.3%, and in the most recent month 10% against 4.3%. Same products. Looking back two years, the story is "these two are the same." Looking forward three months, it is "one of them fixed it." Opposite conclusions from one dataset.
So check the window before you check the number. Our reports default to the full period and give the last three months and the most recent month in the trend chapter; reading both is steadier than reading one. Cross-category comparisons are especially exposed β a two-year report against a six-month report is not a comparison at all.
This also explains why a category ranking can shift after a few months. The products may not have changed. The recent buyers may simply be a different group of people.
9. What this metric cannot prove
It only covers categories we have analyzed. 16 categories, 22 reports, 42 products, 8,211 written reviews. Amazon has tens of thousands of categories and none of the rest are included. The accurate phrasing is "within the categories we analyzed, rates fall in this range," not "Amazon's rates fall in this range."
It is not a return rate. A negative review is someone willing to write. A return is someone voting with their feet. The groups overlap without being identical; plenty of returns never generate a word, and some one-star reviewers keep the product.
Written reviews carry their own bias. The people who type are the extremes. The silent middle is missing from the sample, which means a high rate can reflect a category of expressive buyers as much as a category of bad products.
The average and the rate come from different pools. The average weights every rating; the rate counts only written reviews. They are supposed to disagree. Every comparison in this piece uses each metric on its own terms, with no conversion between them.
Some products get analyzed twice. The 8,211 figure counts reviews by report, and two products were analyzed against different competitors. Deduplicated it is 7,873. No ratio is affected, but the total should be stated precisely.
Writing the boundaries down matters as much as getting the numbers right. The value of a metric is not in what it proves. It is in knowing what it cannot prove, and still trusting it enough to decide.
Related reading
- The Complete Guide to Amazon Review Analysis: A Four-Layer Method From Nearly 3,000 Reviews β the negative review rate is the first metric in that method; this piece isolates it
- Which Amazon Categories Have the Lowest Negative Review Rates? β this article defines the metric, that one ranks 16 categories with it and adds internal spread
- Wireless Earbud Review Analysis: Aptkdoe vs Btootos β the full 14-chapter report behind the 3.7-star / 10.9% example, with all 514 reviews available to check