New Tool Catches Fake Review Rings by the Shape of Ratings

Fake reviews rarely come from one person. They come from a group — dozens or hundreds of accounts that all push a product up the charts at once. Most defenses try to catch them by looking at who the accounts are, who they follow, or what they write. This team tried something different: ignore the identities entirely and stare at the overall shape of the ratings.

The engine only sees a histogram (a bar chart of how many 1-star, 2-star, and so on), a total count, and a picture of what "normal" looks like for that product. It deliberately hides the account names until the evidence is locked in, so scammers can't dodge detection by simply switching who does the voting. It then subtracts the randomness you'd expect anyway, leaving a clean signal of deliberate distortion.

They tested it on historical Amazon review streams, pairing real browsing behavior with fake accounts whose membership was known in advance. Accuracy — measured on a 0-to-1 scale where 0.5 is a coin flip — rose from 0.500 to 0.797 when the same fake votes were spread across more and more re-used accounts. In tests focused purely on the shape of the ratings, accuracy hit 0.909 and 0.967. Bolting this onto existing "these accounts keep showing up together" checks lifted a combined score from 0.750 to 0.874.

The honest caveat: this is a lab result, not a live deployment. It was tested on historical data with controlled, synthetic attacks — real-world manipulation is messier and adaptive. There's also a flip side worth watching: a real surge of genuine excitement (a small brand going viral, an author's fans turning out) could look statistically similar, so the tool has to be paired with human judgment before anyone gets punished.