How to Spot Fake Reviews (When the Fake Ones Are Written Better Than the Real Ones)
The short version: you cannot tell from how a review reads. Bought and earned praise use much the same words, which is why the standard checklist flags real customers and lets paid ones through. What separates them is how the praise was acquired — and acquisition leaves traces the writing does not: when the reviews arrived, what arrived before the opinion, who is vouching for whom, and whether the praise remembers the experience or the brochure.
Real praise remembers the scratch. Bought praise remembers the spec sheet.
The usual advice has stopped working
Guide after guide teaches the same list: watch for vague or exaggerated language, too many exclamation marks, suspiciously perfect grammar, reviewer accounts with one review, a rating curve that is all five stars. Some of that is still useful. Much of it now does damage, for two reasons.
It was calibrated on human fakers working at volume. The tells it teaches — clumsy phrasing, thin detail, copy-paste rhythm — are artefacts of someone writing forty reviews an hour for very little money. A language model clears every one of those bars on the first attempt. The advice to distrust polished writing is now advice to distrust the competent.
And it flags real people. This is the part that goes unsaid. Genuine enthusiasm looks like the exclamation-mark tell. A non-native speaker looks like the template tell. A meticulous professional looks like the AI tell. A one-review account is what almost everyone who has ever been moved enough to write a review actually has. Run the standard checklist honestly and it convicts a great many innocent customers — which is worse than useless, because it teaches you that everything is fake, and someone who believes everything is fake has no way left to recognise the real thing.
The question is not “is this fake?” It is “was this bought or earned?”
Trust arrives in one of two ways. It is earned — someone had an experience and chose to say so — or it is rented: bought, staged, incentivised, or routed into place by an arrangement. That is a fact about the transaction behind the praise, not about the prose in front of you, and it is why the tells that survive are all about provenance.
The federal rule already works this way, and it is the part the standard checklist has no room for. Under the FTC’s Endorsement Guides (16 CFR Part 255), paying someone to write a review that is required to be positive is deceptive even if the payment is disclosed. But a paid review the person was free to make negative is legitimate, provided the connection is disclosed clearly and conspicuously — which the FTC defines as difficult to miss and easily understood, not buried where nobody scrolls.
Read that twice, because it inverts the usual instinct. The presence of money is not the tell. The absence of a free judgment is. A creator who takes the cheque, says so at the top, and then tells you what the product does badly has given you something real. An “organic” rave from an account with no disclosure and no downside has given you an advertisement you were not told about.
Every artifact on this page is fictional and pattern-true — invented brands, invented people, real patterns. They are drawn from the item bank of Real or Rented, a daily game built on these eight tells. Nothing here describes a real company or a real person, and no figure in them is a statistic about anything.
The eight tells
Each one reads the acquisition rather than the adjectives. None of them depends on who, or what, wrote the sentences.
1Timing clusters
Conferral trickles in. Purchases arrive in batches.
Rented — Fourteen months of silence, one week of applause, silence again. Praise that only exists during campaigns is campaign material — real users don't observe fiscal quarters.
2Template language
Different people do not share a skeleton. Vendors supply skeletons.
Rented — Three companies, three metrics, one silhouette: "3x'd our ___ in ~9 weeks." When the numbers all wear the same outfit, marketing dressed them.
3Incentive residue
Follow the thing that arrived before the opinion did.
Rented — "I'd say this anyway" is what people say when they wouldn't. Watch the residue: the free months arrived before the opinion did.
4Reciprocity rings
A endorses B endorses C endorses A, inside one week. Rings are round for a reason.
Rented — Three authors, one publisher, a closed triangle of praise timed to one catalog. The blurbs may even be sincere — but the routing was arranged, and arranged praise is rented even when the words are true.
5Specificity inversion
Real praise remembers the experience. Bought praise remembers the brochure.
Rented — Her skin apparently reads ingredient panels. The pH is precise to the decimal; the "glow" is a shrug. When the chemistry is more vivid than the experience, the experience didn't happen.
6The wrong flaw
A real complaint costs someone something. A manufactured one is a compliment in a trench coat.
Rented — A four-star costume on a five-star ad. A real fourth star has a reason — a rattle, a scratched fender — not a compliment in a trench coat.
7Engagement mismatch
The audience can be purchased. The attention cannot.
Rented — 412,000 followers, 214 likes. The audience was purchased; the attention can't be — and the gap has a name: the Counterfeit Discount.
8Manufactured urgency
Reviews describe the past. Salesmen describe the deadline.
Rented — Reviews describe the past; salesmen describe the deadline. A review with a countdown in it is an ad wearing stars.
Five reviews that look bought, and are not
This is the half the checklists leave out, and it matters more than the tells do. A detector that only ever finds fakes has not been tested; it has only been pointed at things it already suspected. Each of these breaks a rule you have been taught, and each one is genuine.
Breaks the rule: “Be suspicious of sponsored content.”
Real — REAL. He leads with the ad, names what the product doesn't do, and sends insomniacs to a doctor. Disclosure isn't rented — the conferral is the honest judgment, not the placement.
Breaks the rule: “Typos and sloppy writing mean a fake.”
Real — REAL. The misspellings, the accidental order, Dale colonizing a second business — nobody invoices for typos and lore. The genuine has details no one would bother to invent.
Breaks the rule: “Polished, formal, statistic-laden writing means AI.”
Real — REAL. It reads like a claims report because he is a claims adjuster — the wrist device, the 41 minutes, the professional contempt for feelings. Stiffness is a voice, not a tell. Templates repeat across people; this could only be one guy.
Breaks the rule: “Gushing five-star praise is a red flag.”
Real — REAL. Half the review is about Miguel and a grandfather's truck. No brand would pay for praise this off-topic — which is exactly why you can trust it.
Breaks the rule: “A sudden follower spike means the audience was bought.”
Real — REAL. The step has weather (the viral clip), proportional engagement, and — the detail nobody fakes — unfollows inside the spike. Real crowds shed; purchased ones only stack.
The one-line version
Stop asking whether the praise sounds honest. Ask what it is made of — when it arrived, what arrived before it, who arranged it, and whether it remembers anything that actually happened.
Everything above is one distinction applied eight ways: attention and trust are either contested — taken, bought, competed for — or conferred, given by someone who decided you had earned it. Counterfeit conferral is the whole business model of a fake review, and counterfeits are detectable because they have to skip the part that costs something.
Now drill it
Reading about the tells is not the same as catching one at speed. Real or Rented gives you five artifacts a day — a review, a testimonial, a caption, an endorsement, a chart — and one question each: earned, or bought? Every reveal names the tell. Some of the real ones will catch you out, which is the point.
Play today’s five →Common questions
How can you tell if a review is fake?
Not by how it reads. Bought and earned reviews use much the same vocabulary, which is why text-based checklists flag real customers and miss paid ones. What separates them is how the praise was acquired, and that leaves different traces: timing (conferral trickles in, purchases arrive in batches), incentive residue (something arrived before the opinion did), routing (A endorses B endorses C endorses A inside one week), and specificity inversion — real praise remembers the experience, bought praise remembers the brochure.
Are sponsored posts and #ad reviews fake?
No, and treating them as fake is a common mistake and an expensive one. Under the FTC's Endorsement Guides, paying someone to write a review that is required to be positive is deceptive even when the payment is disclosed; a paid review the person was free to make negative is legitimate provided the connection is disclosed clearly and conspicuously. So the question is not whether money changed hands. It is whether the judgment was free, and whether you were told. A creator who says what a product does not do, in a disclosure you did not have to hunt for, is giving you a real one.
Do fake review detectors work?
They help, and they are not a verdict. Any detector has to draw its line using the text, and the text is the layer where bought and earned look alike — which is what the five examples above are for. Each of those is a genuine review that a strict rule would throw away. Treat a score as one input rather than a ruling, and use the tells here on the review actually in front of you.
Has AI made fake reviews harder to spot?
It has broken the old advice specifically. Guidance to distrust poor grammar, generic phrasing or suspiciously polished writing was calibrated on human fakers working at volume, and a language model clears every one of those bars. What AI does not have is the thing genuine reviews are full of: the scratched floor, the delivery driver's name, the flaw that cost the writer something. Provenance tells — timing, incentive, routing, disclosure — are unaffected by who or what wrote the sentences.
Sources. The disclosure rules described above are the Federal Trade Commission’s Guides Concerning the Use of Endorsements and Testimonials in Advertising, 16 CFR Part 255, and specifically § 255.5, Disclosure of material connections. This page is not legal advice.
The weekly case file
The game runs daily. This is the weekly version: every Friday for twelve weeks, one email covering that week’s five case files — the artifact built to catch you, the tell that settles it, and the principle underneath. Then it stops. Twelve is the whole promise.