We now have 18 published reports behind us: 14 product categories, 34 products, 6,891 written Amazon reviews, the newest batch pulled this month.

Last month we ranked all of it the other way around, highest negative review rate first, and the finding was that the danger sits almost entirely on the power side of a product. This time we went bottom-up, hoping to answer a more practical question: if you just want a category where things don't go wrong, what does that list look like?

What we found is that the list doesn't answer the question.

A negative review rate tells you how much danger lives in a category. It says nothing about whether anyone in that category has figured it out. The second thing is what actually determines whether you can win there. Two categories can both sit around 30%, but if the products in one of them are 2x apart, somebody in there is doing something right and you can learn from it. If a category sits at 11% and the products are only 1.1x apart, then nobody is doing it right and nobody is doing it wrong. You'd land in the same place.

So this piece has two halves. First the list, since that's what most people came for. Then the second dimension, which is the part we think is actually useful.

1. How the list was built

The methodology is identical to the previous ranking, so here are the essentials only; the link at the end goes to the full write-up.

Negative review rate is 1–2 star reviews divided by that product's written reviews. Worth repeating: this is not the star rating shoppers see on the listing. The headline rating is a weighted average of every rating including the many buyers who never wrote a word. Written reviews are a much smaller pool, and the two numbers not matching is normal, not an error.

When a category has more than one report, the reports are merged and weighted by review count. Neck massagers, for instance, combine to 32.3% across 711 reviews.

The window is August 2024 to October 2026. The sample floor is 100 written reviews per product; below that we don't draw conclusions. One exception appears in this piece and it's flagged where it shows up.

Since the last ranking, the library grew from 15 reports to 18, adding three categories: self-cleaning litter boxes, USB cables, and USB-C cables. Total review volume went from 6,054 to 6,891.

2. All 14 categories, ranked

Starting with the complete table, because showing only the bottom half is how people end up assuming the rest doesn't matter.

Rank Category Negative rate Sample Reports
1 Self-cleaning litter box 47.5% 259 1
2 Power banks 35.5% 400 1
3 Smart pet feeders 33.7% 403 1
4 Neck massagers 32.3% 711 2
5 Wired earbuds 31.1% 389 1
6 Pet water fountains 29.5% 478 1
7 Chargers 28.8% 640 2
8 Electric shavers 28.0% 403 1
9 Night lights & sleep aids 24.7% 737 2
10 Electric toothbrushes 19.9% 857 2
11 USB-C cables 18.1% 216 1
12 USB cables 14.1% 362 1
13 Translation earbuds 14.0% 522 1
14 Wireless earbuds 10.9% 514 1

Top to bottom that's a 4.36x spread, wider than the 3.26x we reported last time, and the new entry is the reason. Self-cleaning litter boxes land 12 points above the previous worst.

Only five categories come in under 20%: wireless earbuds, translation earbuds, USB cables, USB-C cables, and electric toothbrushes. We'll take them one at a time, because the useful question is whether each low number comes from doing the job well or from not having much of a job to do.

3. The five lowest-rate categories, and why they're low

Electric toothbrushes β€” 19.9%, 857 reviews

Starting here because it's the clearest example of why the average is a bad summary.

857 reviews, two reports, four products, 19.9% on average. The individual products: 8.6%, 19.5%, 24.7%, 29.4%. Best to worst is 3.42x, the widest internal spread of any category we've analyzed.

Where does the 8.6% come from? It's the rate for one product measured against the product it was compared with in the same report, which came in at 29.4%. The same buyers, the same window, same methodology, 3.4x difference in complaint volume.

The complaints point in different directions too. For the 8.6% product, the top groups are comfort and vibration at 20.6% (a single item, "motor too weak," is 14.7%), cleaning performance at 14.7%, and durability at 14.7%. The 29.4% product leads with battery and charging at 23.8%. In the other report, the comparison unit leads with charging and battery at 37.2%, where "charging base fails to charge" alone accounts for 14.3%.

Across all four toothbrushes, the battery-and-charging group runs 11.8%, 21.4%, 23.8%, 37.2%. A 3.15x gap inside one category, on one dimension. The 19.9% headline describes none of the four products.

USB-C cables β€” 18.1%, 216 reviews

216 reviews across two products: one with 155 reviews at 14.2%, one with 61 reviews at 27.9%.

By our own floor, 61 reviews isn't enough to draw a conclusion, so this section reports the numbers without treating them as evidence. The category total of 216 is fine. Anyone using "18.1%" to describe either product individually is reading it wrong.

Both products concentrate their complaints on charging, at 37.5% and 15%. At item level: "stopped charging" 28.1%, "loose connection" 15.6%, "doesn't support 100W" 9.4%, "USB 2.0 slow data transfer" 9.4%, "not durable" 9.4%.

One detail worth noting: buyers don't only complain about failure, they complain about specs that don't hold up. The two items about missing 100W support and slow USB 2.0 transfer add up to roughly 19%. Overselling a spec gets invoiced in this category.

USB cables β€” 14.1%, 362 reviews

362 reviews, Anker against UGREEN, at 13.5% and 14.9%. A 1.10x gap.

Complaints pile into a single group, charging, at 61.9% for one product and 37.5% for the other. Items: "won't fast charge" 33.3%, "stopped charging" 28.6%, "loses connection sporadically" 14.3%, everything else in single digits.

The reason for the low number is straightforward: a cable doesn't promise much. It has two jobs, conducting power and moving data. No battery management system, no app, no motor, no moving parts to wear out. There's no version of this product where the server goes down and the cat misses dinner. When it fails, the cable failed, and the reason is usually one of two things.

So the low rate here is bought with a short list of promises, not with careful engineering. The same sentence works in reverse: there's not much room to differentiate either, which is why two competing products land at 13.5% and 14.9%.

Translation earbuds β€” 14.0%, 522 reviews

522 reviews, two products at 10.0% and 14.6%.

The source of this low number is entirely different, and honestly counterintuitive. Translation earbuds promise a lot β€” multiple languages, real time, sometimes active noise cancelling on top β€” and they still rank second-lowest.

The complaint structure explains it. For one of the two products, the "other / long-tail" group takes 71.4% of all complaint mentions. In other words, the complaints are spread across dozens of small items and none of them can hold the room. The items you can make out: poor translation accuracy 28.6%, earbuds fail to pair with phone 28.6%, poor audio quality 14.3%, battery drains quickly 14.3%.

The low number isn't evidence of good execution. It's evidence that buyers' disappointment has no shared direction. Expectations for this product type are vague to begin with; nobody knows what level of translation counts as acceptable before they buy, so the disappointment arrives in fragments and never consolidates into a dominant complaint cluster. The metric looks healthy, but it isn't describing a stable product. It's describing a category with no agreed passing grade.

The distinction matters because the two situations call for opposite responses. The cable category's low rate is something you can copy. The translation earbud low rate tells you nothing about what to build.

Wireless earbuds β€” 10.9%, 514 reviews

514 reviews, two products at 12.8% and 9.0%, a 1.42x gap, and the lowest rate of all 14 categories.

Complaints cluster in three places. One product leads with connectivity at 36.5%, the other with battery at 37.0%. At item level: "one earbud stops working" 23.1%, "Bluetooth connection drops" 13.5%, "poor sound quality" 11.5%, "earbuds fail within weeks" 11.5%, "battery drains quickly" 9.6%.

Here's the interesting part: the vocabulary of complaint is almost identical to what we see in wired earbuds, which sit at 31.1%. On the wired side, one product leads with durability and build at 22.6%, audio quality at 17.7%, component failure at 17.7%; the other leads with fit and comfort at 35.5%. On the wireless side it's connectivity at 36.5%, then battery at 23.1%, audio quality at 11.5%, durability at 11.5%.

A 2.9x difference in negative review rate, and almost the same list of words. Buyers want the same three things from both, connect, hold a charge, don't break. The difference is how often this price band of wireless earbuds manages to fail at them.

4. One dimension isn't enough

Those five categories share a low rate. Only one of them is worth entering.

The reason is that a negative review rate is an absolute level. It says nothing about how far apart the products inside a category are. That needs a second number: internal spread, the highest negative review rate in a category divided by the lowest, counting only products that clear the sample floor.

Here are the 14 categories with both dimensions side by side:

Category Negative rate Best / worst inside Spread
Electric toothbrushes 19.9% 8.6% / 29.4% 3.42x
Chargers 28.8% 18.5% / 39.7% 2.15x
Neck massagers 32.3% 21.0% / 37.5% 1.79x
Pet water fountains 29.5% 23.3% / 35.5% 1.52x
Smart pet feeders 33.7% 26.6% / 40.0% 1.50x
Translation earbuds 14.0% 10.0% / 14.6% 1.46x
Wireless earbuds 10.9% 9.0% / 12.8% 1.42x
Self-cleaning litter box 47.5% 42.1% / 54.4% 1.29x
Night lights & sleep aids 24.7% 22.6% / 27.8% 1.23x
Power banks 35.5% 33.2% / 37.9% 1.14x
USB cables 14.1% 13.5% / 14.9% 1.10x
Electric shavers 28.0% 27.1% / 29.2% 1.08x
Wired earbuds 31.1% 30.5% / 31.6% 1.04x
USB-C cables 18.1% β€” not enough data

(For USB-C cables, the second product has only 61 reviews, below the floor, so no spread can be calculated.)

Cross the two dimensions and you get four quadrants. The dividing lines: a rate of 20% (lower than 20% counts as low) and a spread of 1.5x (1.5x and above counts as wide). Of the 14 categories, USB-C cables can't be placed because the data is short, and the other 13 divide evenly.

5. What each quadrant means

Low rate, small spread: safe, and no room for you

USB cables (14.1% / 1.10x), wireless earbuds (10.9% / 1.42x) and translation earbuds (14.0% / 1.46x). The three lowest-rate categories all land here.

This combination means everyone in the category is performing about the same and there's no obvious winner. You get certainty that you won't blow up, and in exchange you give up the ability to differentiate through product quality, because there's no weak rival to attack and no better approach to copy.

USB cables are the extreme case. 13.5% against 14.9% means that in buyers' eyes these two products are the same product. Whatever this category gets decided by, it isn't the product. It's price, listing quality, and ad placement, none of which is what this dataset helps you with.

If what you want is stability, this quadrant delivers. If what you want is to win on the product itself, this quadrant can't give you that.

Low rate, wide spread: a rare window

Exactly one of the 14 categories lands here: electric toothbrushes (19.9% / 3.42x).

The category is broadly healthy and one product inside it is at 8.6% while another is at 29.4%. Two things are true at once: the basics aren't hard, and there's both a benchmark worth copying and a weakness worth attacking.

The value of this shape is that the cost of validation is low. You don't have to bet on whether the category works, because someone already made it work. All that's left is figuring out where the 3.4x comes from. In this data the answer isn't far away: the battery-and-charging group runs from 11.8% to 37.2% across the four products, so the gap sits on that one dimension. Getting the charging base right is an immediately bankable point of difference here.

The fact that only one category out of 14 has this shape is itself the information. Most categories don't have a structure this comfortable.

High rate, wide spread: someone figured it out, go read it

Chargers (28.8% / 2.15x), neck massagers (32.3% / 1.79x), pet water fountains (29.5% / 1.52x), and smart pet feeders (33.7% / 1.50x).

The danger is real, but it isn't evenly distributed. Charger products run from 18.5% to 39.7%; neck massagers from 21.0% to 37.5%. Inside a single category, one product's negative rate is double another's. That tells you the failures here belong to specific products, and product-level failures can be studied and solved.

If you're set on entering one of these categories, the first thing to do isn't to survey the market. It's to take apart the best performer and work out what it avoided.

High rate, small spread: a category problem nobody has solved

This is the quadrant to avoid, and it happens to contain most of the worst offenders: self-cleaning litter boxes (47.5% / 1.29x), power banks (35.5% / 1.14x), wired earbuds (31.1% / 1.04x), electric shavers (28.0% / 1.08x), with night lights and sleep aids (24.7% / 1.23x) sitting here as well.

A high rate isn't the remarkable part. Everyone being equally bad is. Wired earbuds: 30.5% and 31.6%. Electric shavers: 27.1% and 29.2%. Power banks: 33.2% and 37.9%.

That pattern means the failure is structural. It isn't something you can spend your way around or fix with one more design revision.

Self-cleaning litter boxes are the cleanest example. A 47.5% negative rate, an average market rating of 2.8 (the only category in our set below 3.0), and yet the two products are only 1.29x apart. No winner. Their top complaint groups don't even match: one leads with cleaning performance at 44.9% (a single item, "waste gets stuck in rake prongs," at 12.1%), the other with mechanical failure at 54.5% ("broke in less than 5 months" at 6.1%). One product's tray drew complaints for being made of cardboard, at 6.6%.

Why does nobody escape it? Because a self-cleaning box has to solve mechanical scooping, sensor detection, odor control, and cleaning at the same time, and "waste gets stuck in the rake" has almost no tolerance for error. Get it right nine times out of ten and buyers will remember the tenth. In a category like this, the product's real competitor is physics.

This quadrant is the part we most wanted people to see. A high negative review rate doesn't mean a category is off limits. What matters first is whether someone got it right or whether nobody can. The first is researchable. The second is something to route around.

6. How to use this

Reverse the order: read the internal spread first, the rate second.

Swap the order and the action flips. A category at 47.5% with a 2.5x spread is an opportunity, because somebody inside it got down to 19%. A category at 47.5% with a 1.29x spread is one to route around, because it's equally hard for everyone. The two look identical in a ranking table, and the difference is entirely in whether you looked at the second dimension.

Then split the three reasons a rate can come out low, because each one calls for a different move.

Low because the product promises little (cables, wireless earbuds) is safe to enter, at the cost of commodity dynamics: hard to win on the product itself. A scattered complaint pattern produces a different kind of low (translation earbuds, 71.4% long-tail), and there's nothing to copy from that number, since all it says is that the category has no shared standard. The third shape is the one worth hunting for, where the low average comes from a leader doing genuinely excellent work, like the toothbrush at 8.6%. That's the only kind of low that gives you both a safety margin and room to move.

Read the spread as a proxy for how copyable a category is.

A wide spread means there's a known answer to reference and a known weakness to attack. A spread near 1 means there's no answer to copy, because there isn't an answer. This reading only works alongside the one above it.

One last note, not really a method: sample size is a signal too.

Samples in this piece range from 216 to 857 reviews. Categories with more reviews have more solid demand and cheaper research. Reading 857 reviews tells you what buyers care about; 216 may leave you collecting more data yourself before you're ready to decide.

7. What this data can't prove

The boundaries need to be stated, or the quadrants above will get used as conclusions.

It only covers the categories we've analyzed. 14 categories, 18 reports. Amazon has tens of thousands of categories and none of the others are included. The accurate phrasing is "of the categories we've analyzed, these five have the lowest rates," not "these are the five lowest on Amazon."

Internal spread depends on the pairs we chose to compare. This one deserves to be spelled out: the two products in a report were selected for comparison, not randomly sampled. So "spread" measures the gap between the competitors we compared, not the dispersion of the whole category. Swap in two other products and the number changes. Its value is in showing whether a given pair has a clear strong and weak side. It can't be used to characterize a category's entire competitive structure.

A negative review rate is not a return rate. A negative review means someone bothered to write. A return means someone voted with their feet. The two groups overlap without being identical, and plenty of people who return a product never write a word. Return losses have to be calculated separately.

Written reviews carry their own bias. The people who type out a review are already the extremes, the very satisfied and the very disappointed. The silent middle isn't in the sample. So these percentages describe the distribution of extreme experiences, not of all buyers.

A low rate doesn't mean an easy category, and this article argues the opposite. Of the five lowest-rate categories above, exactly one is worth entering. That sentence on its own should tell you a rate can only serve as a first filter.

The same product gets analyzed more than once. The 6,891 figure counts reviews by report, and two products were analyzed twice against different comparison units. Deduplicated, it's roughly 6,500. This doesn't move any ranking, but the number should be stated accurately.

Stating what a dataset can and can't prove matters as much as computing it correctly. These boundaries are part of the data, and they come with it.

Related reading