Faked reviews are an old industry problem: review farms, “review for a discount” swaps, sometimes plain bots. You can never tell a fake from a real one with full certainty — but three signals give a confident enough shift.
First — time spikes: 40 reviews in a single day after months of silence is statistically unlikely for an organic guest flow, and almost always lines up with a paid campaign.
Second — templated text: high textual similarity across reviews from different “guests” (same phrases, same structure, even the same typos) is a clear sign of one source, not many people.
Third — unnaturally even category scores. Real guests disagree with themselves: they’ll honestly complain about Wi-Fi while praising the location. Faked five-stars are usually flat across every category at once — that’s the “evenness” signal, and it lowers that review block’s contribution to the index.
We don’t fully ban suspicious reviews — the risk of wrongly cutting a real guest is too high. Instead we lower their weight, by the same logic that ages out old reviews over time rather than deleting them. We’d rather show an honest, more modest number than a pretty, unverifiable one. 🕵️