Every layer of product research above the checkout has the same weakness: it can exist without a single sale. Ads prove that someone is spending money to sell a thing. Search volume proves that someone is curious about it. TikTok views prove that someone found it entertaining. None of it proves that anyone is buying.
Amazon is where that gap closes. A Best Sellers Rank only moves when orders are completed, and a verified review can only follow a purchase. If you want to know how to check product demand on Amazon, the raw evidence is published in plain sight on nearly every listing — the trouble is that both numbers are routinely misread. Rank gets treated as an absolute score when it is relative. Review counts get treated as demand when they are mostly history.
What follows is a field method for reading both correctly: what a rank actually encodes, how to estimate current sales flow from review dates, how to read the shape of a first page, and where this evidence stops being useful.
Why retail proof is the honest signal
Ranks and reviews are byproducts of completed purchases. That single property separates them from everything upstream. A brand can run ads for months on a product that never earns back its spend. A trend can collect millions of views without producing a checkout. A subreddit can complain at length about a problem nobody will pay to fix.
A category's review flow cannot be faked the same way. One seller can buy a batch of incentivized reviews for one listing — it happens, and it is worth watching for. But a whole first page of listings quietly accumulating dated, verified reviews month after month is the residue of real orders. Nobody counterfeits demand across a hundred competitors at once.
That is why we treat Amazon as the retail-proof layer: the place in the research stack where you stop reading promises and start reading receipts.
Reading Best Sellers Rank without a decoder ring
The most common question — what is a good BSR? — has no answer as asked, because BSR is category-relative. A rank of 10,000 in Electronics, one of the largest categories on the platform, still describes a product selling steadily every day. A rank of 10,000 in a small sub-category might describe a product selling once a week. The number only means something next to the size of the shelf it sits on.
The second thing to understand is that BSR is recency-weighted and volatile. It reacts to the last day or so of orders, so a single lightning deal or a viral mention can catapult a listing to #50 for an afternoon. One day's rank is weather. What you want is climate: direction and stability over weeks.
The manual method for Amazon BSR product research is unglamorous but honest. Pick five to ten representative listings in the market — a couple of leaders, a few mid-pack, one or two new entrants. Note their sub-category ranks today. Note them again in a week, and again a week after that. A set of ranks that holds steady or improves describes a market with a pulse. A set that decays in lockstep describes one that is cooling, whatever any single day's number said.
Field judgment, not research: in a mid-sized sub-category, a listing that holds inside the top 1,000 for a month is selling every day. That stable rank tells you more than any one-day spike to #50 ever will.
Review velocity beats review count
Lifetime review count is a history book. A listing wearing 4,000 reviews may have earned most of them three years ago and be coasting on the accumulated weight. If you want to know how to tell if a product sells on Amazon now, the number that matters is new reviews per month — review velocity — because reviews trail orders at a roughly steady ratio within a category.
You can estimate it by hand from the date stamps. Sort a listing's reviews by most recent and count how many carry dates inside the last 30 days. Or, faster: find the tenth most recent review, note its date, and divide. Ten reviews over five days is a very different business from ten reviews over five months.
Only a small fraction of buyers ever leave a review — the exact ratio varies by category and price point, but orders outnumber reviews many times over. So a listing gathering 30 or 40 new reviews a month is, with near certainty, shipping a healthy multiple of that in units. You are not computing revenue; you are establishing an order of magnitude, and an order of magnitude is usually all a sourcing decision needs.
Velocity also answers a question nothing else can: did demand outlive the moment that created it? A product that spiked on TikTok six months ago and is still accumulating reviews at pace has crossed from trend to staple. That distinction — spike or signal — deserves an article of its own, but the review date stamps are where the verdict shows up first.
The shape of the first page
Search the market's main buyer keyword and read the top ten organic results as a distribution rather than a list. Amazon competition analysis mostly comes down to the shape of the review counts across that first page, and there are three shapes worth recognizing.
- One giant, weak also-rans. A leader with tens of thousands of reviews above a field of listings with a few hundred each. Demand is proven and concentrated — and the runner-up slots are winnable. A wedge exists for a differentiated entrant.
- Ten deep incumbents. Every slot on the page holds thousands of reviews. Demand is beyond doubt, but so is the moat: buyers here have abundant trusted options, and displacing any of them is a long, expensive campaign. Raw competitor counts mislead about this — saturation is about the shape, not the number.
- A page of sub-50-review listings. Tempting, because it looks wide open. But thin review counts across the whole page mean demand itself is unproven. You might be early. You might be looking at a shelf nobody visits. Either way, the market has not yet shown it can absorb inventory.
While you are there, note the price band the page clusters around and how many of the top slots are sponsored. A page carried mostly by ads, with weak organic ranks underneath, is demand being rented rather than owned.
Complaint mining: the free product roadmap
The one-to-three-star reviews of a category's top sellers are the cheapest product development brief you will ever read. Open the leaders' critical reviews and tally recurring themes. If buyers keep writing too dim, broke in a week, sizing runs small, or impossible to clean, they are dictating the spec sheet of the improved version — and telling you the incumbents have left a door open.
Two habits make this reliable. First, count rather than collect: a complaint that appears once is an anecdote, while one that appears in a fifth of critical reviews is a market. Second, cross-reference outside Amazon. If the same failure shows up in community threads on Reddit — unprompted, in buyers' own words — the gap is real and unsolved, not the artifact of one bad batch.
The limits of trailing evidence
Everything above shares one weakness: retail proof looks backward. Ranks and reviews confirm that a market exists — that real money has changed hands, recently and repeatedly. They cannot tell you where the market is heading. A category can post its best review-velocity month at the exact moment its underlying trend peaks and turns.
So treat Amazon as the confirmation layer, not the compass. Pair it with the present tense — which competitors are spending money right now, visible in the Meta ad library — and with direction, which Google Trends sketches better than any marketplace can. A market showing retail proof, live ad spend, and rising search interest at the same time is a different bet from one showing retail proof alone. Reading all four signals together is the whole discipline.
The compressed version
Done by hand, the method in this post — rank checks spaced across weeks, review-date counting, first-page distribution, complaint tallies — costs a few honest hours per market. Kestrel runs this retail-proof check as one of four evidence layers, alongside live Meta ads, search direction, and community chatter, and folds them into a 0-100 market score with a Hot, Promising, or Weak verdict. The specimen report shows exactly what the rank-and-review evidence looks like in that format, and the first 20 market scans are free, no card required. The manual method still matters: it is how you audit any score, ours included.
Receipts, not promises
Amazon is the only stop in product research where the evidence is made of completed transactions. That makes it the strongest single signal you can read for free — and also the easiest to over-trust, because a receipt tells you what happened, never what happens next.
So use it in its place. Let rank stability over weeks, not one day's number, establish whether the market has a pulse. Let review velocity, not lifetime counts, estimate the current flow. Let the shape of the first page tell you whether there is room to stand, and let the critical reviews tell you what to build. Then look up from the shelf and check which way the wind is blowing before you order inventory.
