One score, and every input behind it.
A rating you can't interrogate is decoration. So here is the whole method — what counts, what it's worth, what we throw away, and the cases where we'll tell you to book somewhere else.
The sources that count
Every rating is normalised onto a 0–100 scale first, because a 4.5, an 8.9 and a 92% are not measuring the same thing. Each source then contributes according to how much we trust it and how much data stands behind it.
Weights are relative, not absolute. When a platform has nothing for a property, it drops out and the rest are renormalised — the score never quietly assumes data it doesn't have.
Why volume matters, but not linearly
A hotel with 4,000 Google reviews is a better-evidenced 8.6 than one with 40. But 4,000 reviews is not 100× more informative than 40 — the extra reviews mostly repeat each other. So each source's influence is scaled by the logarithm of its review count rather than the count itself, bounded between 0.45× and 1.4×. A strong TripAdvisor sample of 400 still gets heard next to a huge Google one.
Why social engagement is not in the score
We read Instagram, TikTok and YouTube, and we show you everything we find. None of it touches the score.
Engagement measures how photogenic a lobby is. That is frequently uncorrelated with whether the air conditioning works, the walls are thin, or the front desk answers at 2am. Folding reach into a quality number would be exactly the conflation this whole product exists to undo — so social sits next to the score, labelled as reach, and nothing more.
How themes are counted
Themes are named by a model and counted by us. The model only ever sees a sample, so asking it how many guests mentioned breakfast returns a guess. We ask it for keyword sets instead, then scan every stored review to count real matches.
The sample it names from is stratified across platform and rating band — 1–2 star, 3 star, 4–5 star — so it sees the complaints as well as the raves rather than whichever source dumped the most text.
When counting, a review from the last 18 months or one left by a verified stay counts roughly twice as much. A theme has to clear both two separate guests and 4% of the corpus before we call it recurring.
What the bands mean
When we'll tell you not to book
Every other booking site earns money the moment you click confirm, which is precisely why none of them will ever talk you out of one. We would rather have you back than have this booking, so a profile can carry a verdict that costs us the sale:
- The recurring complaints outweigh what the star average suggests.
- The rate today is well above what this property normally charges.
- There isn't enough independent data to stand behind a number at all.