The algorithm doesn’t read reviews like a human, but it does look for specific marker phrases that statistically predict a real problem — or a real plus — better than the score alone. Here are six we watch closest. 🔍
“Not like the photos” — the strongest negative marker: a direct gap between the promo image and reality, exactly what the whole index is built to catch.
“Thin walls” / “could hear the neighbours” — soundproofing you can never see in a photo, but feel clearly at 3am.
“The host met us and helped” — a personal-contact marker, typical of honest small places and almost never seen in assembly-line reviews.
“It smelled” / “smell” — critical for hygiene and almost never present in templated fakes: scripted reviews rarely invent sensory detail.
“Even better than the photos” — rare but a strong positive signal: the hotel is underselling itself in promo, not overselling.
“They swapped the room without any fuss” — a service-recovery signal, a good sign even if there was a complaint to begin with.
All six work because faked reviews tend to be templated and vague — “loved everything, recommend” — while inventing convincing sensory detail at scale is much harder. Formally this is part of the “tone” signal, which feeds the index’s confirmation alongside live photos.