How to Decide If an Amazon Category Is Worth Entering

Over the past two weeks we put every published report we have side by side and ranked the whole set twice: once highest negative review rate first, once lowest first. That's 15 categories, 36 products, 7,271 written Amazon reviews, spanning August 2024 to October 2026.

After both lists went up, the question we got most was the same one: so is my product worth doing? A ranking can't answer that. A ranking is about a category, and what you want to know is what happens once you go in. There's a person standing between those two things.

So this piece is built differently. It walks through the decision in four steps, three checks and one calculation, and the links at the bottom let you verify every number.

1. Most people check the wrong thing first

Four approaches come up again and again, and they all share one flaw.

Look at BSR and category growth charts, and whatever ranks high and climbs fast must be worth selling. Look at monthly sales estimates and search volume curves, and bigger numbers mean better business. Look at what other people did, and a product with a few thousand reviews proves the position is solid. Look at the star rating, and anything above 4.5 stars is low risk.

All four answer the same question: is anyone buying?

None of them answer a different one: what happened to the people who bought.

The gap between those two is bigger than it sounds. Across our 15 categories the lowest negative review rate is wireless earbuds at 10.9% and the highest is self-cleaning litter boxes at 47.5%, a 4.36x spread that is completely invisible on a BSR chart. Both look healthy on a rankings page.

The litter box number is worth dwelling on. Its market average rating is 2.8 out of 5, the only category in our set below 3.0. By the "low rating means nobody buys it" logic, this category should be dead. The reason we could analyse 259 written reviews is that people are still buying it, and writing that much about it.

The star rating is even less reliable. The headline score is a weighted average of every rating, including a large share of buyers who never wrote a word. A one-tap star and a three-paragraph complaint count the same. The only reviews you can turn into product changes come from the people who wrote something.

So the order should be reversed: risk first, opportunity second. Risk is how much complaining a category generates no matter who you are, and it has little to do with how good your product is. Opportunity is whether you can do better than the field, and that part you can influence. Risk sets your floor, opportunity your ceiling, and the order itself changes conclusions — same data, different sequence, opposite answer.

2. Check one: is the demand real, or is it fed to you

Start by confirming that the demand comes from buyers actively looking, not from a feed or an ad budget. Both look identical from the outside. They collapse at very different speeds.

The test is crude but reliable: can you write down three search terms a buyer would actually type?

Category terms are easy. Cat tree, neck massager, dash cam. Anyone can list three.

Qualified category terms matter far more. Cat tree for large cats. 65W USB-C charger. Wireless earbuds for small ears. Qualifiers tell you buyers aren't picking something up at random, they're choosing with specific requirements in mind. The more qualifiers a category has, and the finer they get, the more entry points exist.

Brand terms are the most valuable layer. That's a buyer typing Feandrea cat tree, Anker 65W, PETLIBRO feeder. If you can write brand terms, the category has real brand recognition, buyers are comparing brands against each other, and that traffic can be borrowed or intercepted.

If you can't write one of the three layers, stop and think about why.

Translation earbuds illustrates this well. The terms are new, plenty of people search them, and the search curve looks great. Then you look at the complaint structure and something is off. In our report the "Other / Long-Tail" group accounts for 76.7% of complaints, spread thin across dozens of small items with no shared letdown holding the field. The reason is that buyers themselves don't know what level of translation counts as acceptable, so there's no pass mark to complain against. When demand is newly manufactured, the terms are manufactured and so is the standard, leaving you no clear standard to win on.

The opposite case is chargers and cables. Brand-term density there is the highest of any category we cover. Buyers search anker 65w charger, ugreen nexode. Active search is so obvious it doesn't need proving. The cost comes with it: the more certain the demand, the more people enter, and chargers run a 28.8% negative review rate with the two products we compared sitting at 18.5% and 39.7%, a 2.15x gap. Brutal competition, and somebody in there did win on product.

One more signal gets ignored constantly: sample size itself. Electric toothbrushes have 857 reviews, night lights and sleep aids 737, neck massagers 711. Buyers in those categories are unusually willing to write, which tells you the demand is real. At the other end, USB-C cables total 216 reviews, and the smaller product in that report has 61, below our own floor. A thin sample isn't just less data, it also means buyers there don't care enough to type.

The one-line version: if you can't write three terms a buyer would actively search, don't talk about whether the category is workable yet.

3. Check two: has anyone in this category figured it out

Demand confirmed, next look at the competition. There's exactly one metric to use here, and we call it internal spread: within a category, among products with enough sample (at least 100 written reviews), the highest negative review rate divided by the lowest.

The category's absolute negative rate tells you how much risk sits in the market. The spread tells you something else entirely: whether anyone in this market is doing it differently from everyone else. Neither number means much alone.

Here's all 15:

Read the table against a single line: 1.5x.

Above 1.5x, somebody in this category has solved something the others haven't. You can take them apart and see what they avoided, or attack the weak end of the range. Below 1.5x, the products in that market are effectively the same thing in a buyer's eyes, and you won't land anywhere different from where they are.

The three high-spread categories get there for different reasons.

Electric toothbrushes at 3.42x deserve the most attention. In the same report, the two products compared sit at 8.6% and 29.4%. Same buyers, same standard of extraction, a 3.4x difference in complaint volume, and it has very little to do with how well the brush cleans. Across four products the "Battery & Charging" group accounts for 11.8%, 21.4%, 23.8% and 37.2%, a 3.15x gap between best and worst, and the difference lives in one part: the charging base.

Chargers at 2.15x are a different shape. The risk is spread across three dimensions — durability at 29.6%, connection stability at 24.1%, charging speed at 21.3% — all three of them engineering problems, which is how 18.5% and 39.7% get that far apart.

Neck massagers at 1.79x are highly concentrated. The top complaint, durability, takes 42.3%, so nearly half the complaints point at the same component. The 21.0% to 37.5% gap is basically whether that one part was made properly.

The low-spread categories are more interesting, because they look safe while carrying less information.

Wired earbuds sit at 1.04x, with products at 30.5% and 31.6%. Cat trees at 1.06x, 18.9% and 17.9%. Electric shavers at 1.08x, 27.1% and 29.2%. All of them hugging 1.

The negative rate levels across those three are wildly different, from 18% to 31%, and the read is the same: you won't do any better than they did.

This breaks open a vague phrase everyone uses — "intense competition." There are two completely different situations hiding in it. One has a visible winner. The other has nobody winning at all. The first is researchable; you know what to chase. The second leaves you fighting on price, listing quality and ad placement, which happens to be exactly what review data cannot help with.

One caveat: the "best" and "worst" in that table are two products we picked specifically to compare, not a random sample. The spread measures the gap between those particular competitors. Swap the competitors and the number moves. Use it as a read on whether this pair has a clear strong side, not as a claim about the category's whole competitive structure.

4. Check three: is the risk concentrated, or spread out

Third step, look at what the risk looks like. A negative review rate tells you how many people complain, not where the complaints land. The same 30% rate behaves differently when three quarters of the complaints point at one part versus when they scatter across dozens of small issues.

The metric here is top complaint share: across a batch of reviews, the most-mentioned group of problems as a share of all problem mentions. Across the 15 categories:

That share means "among the people who complained, how many complained about this," not "how many buyers hit this problem." The two are far apart. One more note: group names aren't standardised across reports, so the same charging issue appears as "Battery & Charging" in some and "Charging & Battery" in others. I merged by name, which means one component can show up as two rows.

High share and low share correspond to two different kinds of risk.

Concentrated risk can be located, scoped and verified. You know which component to change, and you know how you'd confirm the change worked, because the share of complaints pointing at that dimension is your acceptance test. USB cables are the extreme case, with charging at 54.3% of mentions.

Spread-out risk gives you nothing to aim at. Change any single thing and you gain no position, because you don't know which thing the buyer cares about. The 76.7% long tail in translation earbuds is the type specimen: dozens of small entries that never add up to a shared complaint.

Here's the part that needs care, though, because high share doesn't mean easy and low share doesn't mean hard. They measure different things.

Self-cleaning litter boxes run 36.7% mechanical failure and 32.7% cleaning performance, so concentration isn't low. But both are physical-condition problems: the product has to solve scooping, sensor detection, odour control and cleaning at once, and "waste gets stuck in the rake" has almost no tolerance. Get it right nine times out of ten and the buyer remembers the one time. In that category your real competitor is physics.

Flip it around: electric toothbrushes show a top complaint share of only 17.1%, which reads as spread out. But in those 857 reviews, Charging & Battery at 17.1% plus Motor & Performance at 15.1% is just two things — charging and the motor. Both are component problems, both fixable. The share looks low only because the two products each failed on a different one, splitting the complaints across two groups.

So when you see a low share, don't call it "spread out" yet. Go one level down into the individual items and check whether several groups are pointing at the same underlying problem.

A brand new report that came out today makes a good example. Cat trees, Feandrea versus Yaheetech, 380 reviews, 18.4% negative rate, 3.7 average rating. On the surface, a healthy category.

Its top complaint is durability at 44.5%. The item-level breakdown reads "not sturdy" 19.3%, "carpet rips easily" 12.0%, "platform broke off" 12.0%, followed by "hard to assemble" 9.6%, "missing a piece" 8.4%, "too small for my cat" 7.2% and "customer service won't replace part" 7.2%.

The structure is clear: concentrated, and static. Materials, hardware, structure — problems with established fixes in the industry, not physical limits. And yet the internal spread is only 1.06x. Neither of the two products we compared solved it.

So this step produces two separate sentences: can it be located, and has anyone solved it. They're independent questions, and only together do they form a judgment.

5. The calculation: concentration times internal spread

The first three checks give you three readings: whether demand is real, whether the field has layers, where the risk concentrates. The fourth step crosses the last two. Split both concentration and internal spread at the midpoint and you get four cells.

One: concentrated, and somebody solved it. Electric toothbrushes (17.1% top share but concentrated in charging and motor, 3.42x spread), chargers (durability 29.6%, 2.15x), neck massagers (durability 42.3%, 1.79x).

This is the comfortable cell. You know what to change, and somebody has already proven it can be changed. The work here is turning that into an acceptance test rather than a slogan: fix the charging base, push durability past what the 18.5% product does, and the share of complaints on that dimension afterwards is your result.

Two: concentrated, and nobody solved it. Cat trees (durability 44.5%, 1.06x), wired earbuds (fit and comfort 34.7%, 1.04x), electric shavers (shaving performance 39.1%, 1.08x).

This is the cell people misread most often. The problem is concentrated and the direction looks obvious, yet nobody in the field has solved it. Usually that means the cost of fixing it exceeds the return, or the concentrated problem is simply the default configuration at that price point.

Cat tree durability is the example. Thicker board and better hardware would genuinely fix it, but the cost goes up, and the same buyers complaining about the platform breaking are also complaining the tree is too small for their cat. They want cheap, big and sturdy, and at that retail ceiling you can't have all three.

You can enter this cell, but do the arithmetic first: what does fixing this cost, can you carry the cost into price, and if you can't, you've spent money for nothing. Our data can't answer that. It can only tell you where the problem is.

Three: spread out, and somebody solved it. Pet water fountains (durability 22.8%, cleanliness 17.8%, 1.52x), smart pet feeders (reliability 25.8%, 1.50x).

Winners in this cell usually win on one specific link in the chain, and not necessarily the loudest one. Feeders are the example. The top complaint is reliability at 25.8%, but what actually separates products might be something like whether the device still works when it drops off the network — few complaints, but the buyers who hit it leave one star immediately. What you study in this cell is what the winner did right, not what buyers complained about.

Four: spread out, and nobody solved it. Translation earbuds (76.7% long tail, 1.46x) and the two cable categories (charging at 54.3% and 41.7%, spreads of 1.10x and sample-too-small) all sit near here.

Translation earbuds, as covered above: no agreed pass mark, nothing to copy. Cables are a different animal. They simply don't promise much. A cable does two things, carries power and carries data, with no battery management, no app, no motor and no wearing parts in between, so there are only two or three places it can break. When it fails, the cable failed.

Those two kinds of spread have nothing to do with each other and still produce the same answer: in this cell, what you can do lives mostly outside the product.

6. The four steps compress into one scorecard

Three dimensions, 3 points each, 9 total:

7 or above: worth a full research pass. Widen the competitor set from two products to five or eight, score every dimension. 5 to 6: don't decide yet, go get more competitor data first, because the two you're looking at may not be the two best in that market. 4 or below: this method won't help you. Your edge has to come from price, traffic or supply chain, and product isn't the main battlefield. That doesn't mean the category is unworkable, it means you need a different playbook.

We score our own content topics the same way, on term coverage, available material and competitive difficulty, 3 points each. It's a filter whose only job is to remove what's clearly not worth the effort.

7. What this method can't prove

Time to lay out the boundaries, or the cells above will get quoted as conclusions.

The internal spread depends on which competitors we picked. This matters most. The two products in each report were chosen for comparison, not sampled randomly. Swap them and the number moves. Treat it as a read on whether that particular pair has a clear strong side, never as a statement about the category's competitive structure.

Top complaint share is "among complainers, how many complained about this," not "how many buyers hit this problem." One denominator is complaints, the other is buyers, and they're far apart. Easy to mix up when writing copy.

Locatable doesn't mean worth fixing. That's the cat tree cell. The problem concentrates in durability and the technical path is clear, but the cost arithmetic is yours to do. Our data has no cost data, and no pricing or margin either.

A negative review rate is not a return rate. Reviews are people willing to write; returns are people voting with their feet. The two groups overlap without being equal, so return losses need separate work. Written reviews also carry their own bias — the people who bother to type are the ones with the most extreme experiences, and the silent middle isn't in the sample at all.

The coverage is what it is. 19 reports, 15 subcategories, 36 products, 7,271 written reviews, two products analysed twice because they were matched against different competitors, which brings the de-duplicated total to roughly 6,880. Amazon has tens of thousands of categories, and anything we haven't covered isn't counted. If the category you're evaluating isn't one of these 15, you can still use the method, but none of the comparison numbers above belong to it.

Discount categories with thin samples. The 61-review product in the USB-C cable report is below our own floor and flagged as such in the table. 40% of 50 reviews and 40% of 5,000 reviews are not the same claim.

Writing the boundaries down matters as much as getting the numbers right. What a dataset can prove, and what it can't, is part of its instructions.

Related reading

• How to Analyze Amazon Reviews: A Four-Layer Method Built on Roughly 3,000 Negative Reviews — this piece judges the result; that one turns a pile of reviews into data you can judge
• 10 Amazon Categories With the Highest Negative Review Rates: We Read 6,054 Reviews, and the Danger Sits in the Same Place — the risk-heavy end of the same dataset, and how top complaint share works as a second filter
• Cat Tree Negative Review Report: Feandrea vs Yaheetech — the full 14-chapter report behind the 18.4%, 44.5% and 1.06x example, where you can check every number