Read the shape, not the average
How to spot fake Amazon reviews starts with putting the star rating aside. It is a summary of a sample you have not inspected, and two products showing 4.5 can have completely different evidence behind them.
What separates a genuine 4.5 from a manufactured one is shape. Two hundred reviews collected steadily over two years and two hundred collected in a fortnight describe different products wearing the same badge, and the difference is visible on the page without any special tools. This sits inside the wider question of how to read an Amazon listing at all, covered in the Amazon buying guide.
Look at when the reviews arrived
Reviews are a by-product of sales, so they should arrive roughly in proportion to them. That produces an uneven but explicable line: a slow accumulation with bumps at a seasonal peak, a discount event, or a mention somewhere with an audience.
The pattern worth stopping on is a dense cluster on a young listing, almost all five stars, with nothing in the product's sales position to explain it. A burst of reviews without corresponding sales activity means something other than customers wrote them.
A second shape worth knowing is the rating shift: a product that held a 3.8 for a year and now shows 4.6 across recent reviews. Sometimes that is a genuine revision of a flawed product, which is good news and usually mentioned in the reviews themselves. Sometimes it is a listing that changed what it sells. Reading a few recent entries separates the two in about a minute.
Velocity is a signal rather than a verdict. A product can go viral legitimately and a launch can be well marketed. What a spike warrants is a closer look at the text, not a conclusion, and the detail of how to read those patterns is in review velocity red flags.
Read the three-star reviews first
The most informative entries on any listing are the middling ones, and almost nobody reads them.
A genuine three-star review names a concrete disappointment. The strap frayed after four months. It works but the app is unusable. It is louder than expected in a small room. That specificity is hard to manufacture, because it requires having used the thing.
A manufactured review praises in general terms. Great quality, fast delivery, exactly as described, would recommend. None of that requires possession of the product, and a page whose criticism is entirely absent or entirely generic is telling you the sample is not real feedback.
So the fastest quality check is not counting stars. It is reading five recent three-star reviews and asking whether the people writing them appear to own the item.
Check the reviews are about your item
This is the most common way a rating misleads and it has nothing to do with fraud.
One product page can carry dozens of variations: sizes, colours, capacities, pack counts. Reviews on the parent listing are frequently pooled across all of them, so the 4.6 above a specific colour may have been earned mostly by a different colour, or by the two-pack rather than the single, or by last year's revision of the same model.
The sharper version is variation abuse, where a seller adds an unrelated product as a variation of an established listing and it inherits the review history on day one. A phone case listing acquires a hair dryer. It is against the platform's rules and it happens, and it is obvious within a few reviews.
Before trusting any rating, click into the exact variation you intend to buy and read what is shown for that one. If the reviews describe a different size, a different colour or a different product, the number above them is not about your item.
Incentivised is not quite the same as fake
This distinction matters, because treating them identically makes the whole exercise less accurate.
A card in the box offering a gift card for a five-star rating is against the platform's policy and remains common. It does not usually produce fabricated reviews. It produces filtered ones, because the people willing to take the offer are disproportionately the people who were already happy. The individual reviews are real and the sample is not.
The effect is a rating pulled upward by a percentage nobody can measure from outside, on a product that may be perfectly fine. What it costs you is resolution: the listing has lost its ability to tell you about the unhappy minority, which is the part you needed.
A related case is the review swap or the refund-for-review arrangement, where a buyer is reimbursed after posting. Those look genuine by every surface measure, including verified-purchase status, which is why verified purchase is a weaker signal than most people assume.
The signals that are weaker than they look
Three checks are widely recommended and worth less than their reputation suggests.
Verified purchase. It confirms the reviewer bought the item through the platform, which a refund-for-review arrangement satisfies trivially. The buyer purchases, posts, and is reimbursed afterwards. Every surface signal reads as genuine.
Reviewer history. A profile with a long record looks reassuring, and established accounts are bought and sold precisely because they look that way. A short history is weak evidence; a long one is barely stronger.
Photographs. Images in a review feel like proof of possession and are trivially sourced from the listing itself or from another site. Useful when they show the item in a real setting that matches the complaint. Meaningless as a count.
None of these are useless. They are just not the checks to lead with, and a page that tells you to count verified purchases is giving you a number that the manufactured reviews will also score well on.
What to do with what you find
Four practical conclusions, in order of usefulness.
Weight recent reviews more heavily than old ones. Products change, manufacturing moves, and a listing's reviews from three years ago may describe a different item than the one shipping today.
Treat a thin sample as unknown rather than as bad. A product with eleven reviews has not been judged, and pretending otherwise in either direction is guessing. That is a reason to lean on the return policy, not a reason to avoid it.
Distrust a perfect rating more than a good one. Genuine products at scale accumulate some dissatisfied buyers. A 4.9 across several hundred reviews is less reassuring than a 4.4, because real distributions have tails.
Report the ones that look manufactured. A return closes your own case; a report is what makes the pattern visible to the platform, and it is the only part of this that affects anyone but you.
One last habit worth building: sort by most recent rather than most helpful. The helpful sort surfaces long, early, enthusiastic reviews, which is the opposite of what you want. Recent reviews describe the item as it ships today, from the supply chain currently filling orders, and a product whose quality has slipped usually says so there months before the average moves.
TickClip reads rating and review-count history rather than the average alone, and when the evidence is too thin to judge it says so instead of producing a score anyway. To run those checks on one listing rather than doing it by hand, use the Amazon review checker.
None of this requires accusing anyone of anything. The reviews on a listing are evidence of varying quality, and the only question worth answering is how much weight this particular set can carry.
