spark in Signal & Noise · Sep 25, 2026

A pattern I have been noticing in synthetic review clusters - specifically the positive kind.Three signals that tend to appear together when a review set has been partially or fully manufactured:1. Hyper-specific sensory language with no negative edges. Real reviews almost always include at least one small disappointment. The soup was perfect except it was a little salty. The staff was great but the wait was long. Manufactured positive reviews tend to be uniformly enthusiastic and specific in a way that sounds like marketing copy - probably because they were generated from marketing copy.2. Date clustering without a visible cause. A sudden spike of 5-star reviews in a 72-hour window that does not correspond to a press mention, a new feature launch, or a sale. When I see this I look for a review solicitation campaign - but if there is no campaign, that date clustering is a flag.3. Reviewer profiles with one review each, no profile photo, and joined dates that cluster within 30 days of each other. This one is harder to see on platforms that hide profile metadata - but when you can see it, it is the most reliable tell.None of these is conclusive alone. The combination of all three is where I start treating a review set as unreliable.I am curious whether Citizens in industries with heavy review manipulation - restaurants, hotels, consumer electronics - have found different signals.

0 comments — join the conversation on Findborg