Reading reviews: telling real feedback from manufactured praise

Skip the five star reviews entirely. They are the cheapest kind to manufacture, the easiest to solicit from people who received something free, and the least informative even when they are genuine. Go to the three star reviews instead: those are written by people who used the thing, liked parts of it, and had a specific complaint they felt was worth typing. Then search the review text for the words that describe failure, and check whether the reviewer history and the listing itself hold together. That takes a few minutes and tells you more than an hour of scrolling praise.

Where fake reviews come from

Fake reviews are not mostly written by bored individuals. They are produced at scale by several distinct methods, and each leaves different traces, which is why knowing the production line makes the detection obvious.

Incentivized reviews are the commonest. A card in the box offers a refund, a gift card or a free replacement in exchange for a five star review, or a private group hands out products to members who post afterwards. The review is written by a real buyer with a genuine verified purchase badge, which is why the badge proves so little.

Review farms rent or buy established accounts and post to order. The accounts have history, sometimes years of it, because a fresh account is easy to filter and an aged one is not.

Brushing is a seller shipping a cheap item to a real address harvested from a leak, creating a verified purchase on an account they control. It is also why people occasionally receive parcels they never ordered.

Listing manipulation moves reviews rather than writing them. A seller merges a new product into an old listing, or converts a well reviewed product page into something else entirely, and thousands of reviews for a phone case end up attached to a hair dryer.

Review gating is the legal grey version. The seller emails every customer, asks how things went, and steers only the happy ones to the public review page while routing complaints into private support. Nothing posted is false, and the average is still a fiction.

Listing tricks: merged products and hijacked pages

This is the pattern people miss because it is invisible if you only read text. Review counts are a valuable asset, so they get moved around.

Watch for a mismatch between the product in front of you and the product being reviewed. If the page sells a standing desk and a cluster of reviews discusses a phone charger, the listing has been repurposed. Variation listings do the same thing more subtly: reviews for the popular small size are shown on the expensive large one, even though they are different products with different faults.

Look at the date distribution as well. A listing with thousands of reviews where every recent one is negative and every old one is glowing has usually changed hands, changed factory, or changed contents. The reverse also happens: a brand new listing with a few hundred reviews all within a fortnight did not sell a few hundred units in a fortnight.

On marketplaces where anyone can sell, seller reputation is a separate question from product reputation, and a seller account can be bought outright along with its feedback. If you are buying from an individual rather than a shop, the patterns to weigh are different again and are set out in buying from a stranger safely.

The patterns that give a fake away

No single signal is proof. Two or three together usually are.

Generated text has made the language signals weaker. Reviews written by a model read fluently, avoid the spelling errors that used to give farms away, and can be produced in any quantity. What they still lack is specific incident: the day it arrived damaged, the exact noise it makes, the thing that did not fit. The habits that help elsewhere apply here too, and they are collected in how to tell whether a piece of text was written by AI. Treat fluency as neutral and specificity as the thing that carries weight.

Read the three star reviews first

A three star review is written by somebody with no incentive to exaggerate in either direction. Nobody is paid to write a lukewarm review, and nobody writes one in a rage. That is where the useful sentences live: what it does well, what it does badly, and whether the bad part would bother you.

Four star reviews are the second most useful, for the same reason. One star reviews are worth reading only for the fault they name, because a large share of them are about delivery, packaging or the buyer ordering the wrong thing, none of which tells you anything about the product.

Weight reviews by elapsed time above everything else. A review that says it still works after eight months is worth fifty that say it arrived quickly. Most products fail in the second half of their first year, which is precisely when nobody is writing about them, so sort by most recent and look for the updated reviews where someone came back to add a line.

Searching the reviews instead of reading them

Almost every review section has a search box, and using it is faster and less biased than reading in order. You are looking for the words people use when something goes wrong.

• Time words: after a month, after a year, second one, still going.
• Failure words: broke, cracked, stopped, died, leaked, loose, faded, returned, refund.
• Fit and sizing words: small, tight, heavy, loud, smell.
• Support words: warranty, replacement, customer service, no response.
• The name of the specific part you expect to fail first: hinge, zip, battery, cable, seal.

Two things matter more than the number of hits: whether the same failure appears repeatedly, and whether it is the failure you would care about. A hundred complaints about packaging on a product with a common hinge fault is a better outcome than five complaints that are all the hinge.

Then look at whether anyone from the seller answers. A brand that replies to problems with something concrete, a replacement part or a specific fix, is a brand that expects to be around next year. Canned apologies that ask you to email an address are worth nothing. Check the questions section too, because it is less curated than the reviews and people are blunter there.

What a rating number cannot tell you

An average star rating compresses everything that matters into one figure, and three different things produce the same number.

Sample size first. An average built on a handful of reviews is close to noise, and a slightly lower average across a large number of reviews is usually the safer purchase. Then the shape: genuine ratings tend to cluster at the top with a real tail of complaints, so a distribution with no low ratings at all is more suspicious than one with a visible tail.

Averages also hide age. A product that was excellent for three years and then moved production carries its old score for a long time, which is why the recent reviews deserve more weight than the headline number. And ratings across platforms answer different questions: a retailer rating covers the purchase, an independent testing organization covers performance, and an enthusiast forum covers what breaks in year three. Judging which of those sources deserves your trust is the same skill as telling whether a source online is worth believing.

The same applies to a viral recommendation in a video or a post. Sponsorship disclosure is inconsistent, affiliate links are normal, and the person may never have used the product for longer than the filming, so treat a strong claim the way you would treat any other, by checking it before you pass it on.

A ten minute routine before you buy

1. Read five three star reviews and two four star reviews. Write down the complaint that appears in more than one.
2. Search the reviews for three failure words and the name of the part most likely to break.
3. Sort by most recent and read the last ten, checking that they describe the product actually on the page.
4. Open two glowing reviewer profiles and look at what else they have reviewed and when.
5. Check the rating distribution for a missing middle or an absent tail of low ratings.
6. Search the product name outside the shop, on a forum or a repair community, adding the word problem.
7. Read the return policy and the warranty terms, because they are the only part of this that is enforceable.

That last step is the one that makes the rest low stakes. A clear return window and a credit card behind the purchase means a wrong decision costs you a trip to a parcel drop rather than the price of the item, and the escalation route if the seller stops replying is set out in how chargebacks and payment disputes actually work. If the shop itself is unfamiliar, check the shop before you check the product, using the checks that expose a shop you have never heard of.

One closing caution. Fake reviews and fake shops increasingly travel together, complete with generated photographs and invented brand histories, which is a change of scale rather than a change of method and is covered in what AI actually changed about scams and what it did not. The routine above still works, because it tests the one thing that is expensive to fake: a consistent account of a real object being used over real time.