How to Spot AI-Generated Fake Reviews Before They Cost You Money
AI now writes fake reviews that are grammatically clean and hard to tell from real ones. Here is how to read a product page like a moderator: check the distribution, read the 3-star reviews, and click the reviewer.

The average star rating is the least useful number on the page
Fake reviews used to be easy to spot. They were badly written, obviously templated, or posted in suspicious bursts. That no longer works. Generative AI now produces reviews that are grammatically clean, contextually plausible, and stylistically varied, and a Yale study found that most people cannot reliably tell AI-written reviews from human ones in blind tests.
The scale is large. A 2024 study by the Transparency Company analyzed 73 million reviews and estimated that nearly 14% showed signs of being AI-generated. Amazon, Google, and Meta now remove hundreds of millions of suspicious reviews per year. In Singapore, the Competition and Consumer Commission announced its largest fake-review probe on September 30, 2026, finding that a marketing firm called Reputifly had supplied AI-written fake reviews to about 100 businesses.
The good news is that spotting fakes is a skill, not a superpower. Once you know the signals, you can read any product page critically in under a minute. Here is how to spot fake AI reviews before they cost you money.
Read the 3-star reviews first
Three-star reviews are the least faked category on the internet. Brands buying reviews want 5 stars, competitors planting reviews want 1 star, and reviewers paid with free products overwhelmingly skew positive. Nobody pays to fabricate a lukewarm review.
That makes the 3-star band the closest thing to ground truth on a product page. Read five to ten of them. They typically describe real friction: a zipper that broke after three weeks, a color that looks darker than the photos, a battery that drains fast but a camera that is genuinely good. If the 3-star reviews sound like real people with real complaints, the product is probably fine. If there are almost no 3-star reviews at all, that is a warning sign on its own.
Check the rating distribution, not the average
Never trust the average star rating alone. Click into the distribution histogram, which Amazon, Google, Trustpilot, and the App Store all show. Genuine products tend to follow a J-curve: mostly 5 stars, a meaningful tail of 4 stars, and a smaller tail of 1 to 2 stars from genuinely disappointed customers.
A product with 98% five-star ratings and almost nothing else is suspicious, not excellent. Real products disappoint somebody. Also check the dates: open the review feed sorted by most recent and look for a sudden burst of five-star ratings within days, especially after a period of little activity. In the Singapore investigation, three five-star reviews appeared within ten minutes from different accounts on one business.
Click the reviewer, not just the review
A single review tells you very little. The reviewer's history tells you almost everything. On every major platform, click the reviewer name to see their other reviews. Warning signs include little or no history, many reviews posted within a short period, or endorsements for unrelated businesses in distant locations.
A real reviewer usually has a messy, human history: a restaurant here, a hardware store there, a complaint about a delivery two years ago. A fake account usually has either nothing or a suspiciously uniform burst of praise. If the platform lets you, check whether the purchase is verified, which is a small but real filter.
Look for fingerprints in the text
AI-generated and farm-written reviews share a recognizable register. They tend to repeat the full product name and key features, as if the writer was fed the product title as a prompt. They use vague, hyperbolic language with no specific use case: "thoroughly impressed with the craftsmanship" or "exceeded my expectations in every regard." They skip practical detail entirely: no sizing, no comparison to similar products, no mention of delivery, packaging, or first-use friction.
Read several reviews back to back. AI tools often generate from the same underlying template, so even when the wording differs, the rhythm stays the same: initial hesitation, quick reassurance, flawless experience, strong recommendation. Three reviews that all follow that arc are not three happy customers. They are one template.
One caution: no single clue proves a review is fake. A new account can belong to a real first-time reviewer, and an enthusiastic comment can describe a genuinely good product. Look for clusters of signals across the listing rather than accusing an individual reviewer based on writing style alone.
Cross-check somewhere else
The strongest defense is leaving the page. Compare the product on another platform, check whether the business exists and holds any license it needs, and look for evidence that comes from somewhere other than the business's own marketing. A product with hundreds of five-star reviews on its own site and silence everywhere else is telling you something.
Regulators are catching up. The US Federal Trade Commission's Consumer Review Rule took effect in October 2024, explicitly banning AI-generated fake reviews with civil penalties of up to $53,088 per violation, and in December 2025 the FTC issued warning letters to 10 more companies suspected of violating it. That enforcement helps, but it will never keep up with the volume. Your own judgment is still the first line of defense.
A 60-second routine that works
Here is the routine to use before any purchase that matters. Check the distribution histogram first. Read five 3-star reviews. Click two or three reviewer profiles. Scan ten reviews together for repeated storylines and brand-heavy wording. Then check one other platform.
That takes about a minute, and it will save you from most fake-review traps. The same habit works for spotting AI voice clone scam calls and for checking whether your phone has spyware. The internet rewards people who verify before they trust, and reviews are where that habit pays off fastest.


