The Hedonic Loyalty Gap · 259,647 reviews · 117 hotels · 8 countries

Same rating. Who comes back?

Two hotels. Both 9.1 stars. Same booking page, same five yellow stars — both look like a safe bet. One keeps more than double the guests who write they'll return.

Hotel A · Upscale · Portugal
9.1★
Tap to see who returns
+49
net loyalty · 53% write they'll return
Hotel B · Luxury · Switzerland
9.1★
Tap to see who returns
+22
net loyalty · 24% write they'll return

Which one do guests actually return to? Tap your pick.
Most people pick the Luxury one. Most people are wrong.

Loyalty here is read from what guests write, not from their bookings — it signals a return, it doesn't confirm one. Every number on this page re-runs from the same public review base.

Those two weren't cherry-picked

It's not two hotels. It's thirty-one.

Narrow all 117 hotels to one near-identical band — 9.0 to 9.25 stars — and 31 survive. They span 47 points of loyalty. Luxury sits at both ends. So does Upscale. Neither the star nor the tier tells you which guests signal they'll return.

+17
fewest return · Upper-Upscale · CH
+64
most return · Luxury · CH
47 points

Each dot a hotel, all at 9.0–9.25★ · darker = Luxury (scattered low and high) · A & B are your two cards, located

Across the full cohort, the guest rating explains barely half of the loyalty read — R²=0.53. The other half is invisible on the page you book from.

Once you stop trusting the star, the same reviews start handing you decisions

Four things the rating can't show you.

1,863

The perfect ten who's already gone.

That many guests, in their own words, name a rival as the better choice. Nearly half — 47% — still rated the stay 7 or higher; 230 of them gave a perfect 10. On a dashboard they're green. Handled. They'd already gone.

"Beautiful hotel, breakfast great — but the location was away from everything. I'd choose the historic district next time."a 10/10 review · read from the text, not a tracked switch
82%

When they name a rival, the leak is in the lines you control.

When a guest names a competitor as better, the rival wins on food 82% and value 79% — but location is a coin-flip, 52%. The thing you can't move rarely loses them; the things the operator runs do.

Food
82%
Value
79%
Facilities
74%
Location
52%
2024+, of guests who named a rival: food 108/132 · value 155/196 · location 16/31 (a coin-flip)
1 in 9

The 9-plus that hides a building.

Among guests who scored a property 9–10, roughly one in nine narrate unresolved maintenance in the same glowing review — and the per-property share runs from under 1% to north of 20% at near-identical scores. The score certifies the welcome, not the walls.

"Great service, lovely stay. The room may need some renovation — it's showing its age."scored 9/10 · severity-gated · promoter reviews
+32pp

The save that decides the year.

Same complaint, two operators. Within the same rating band, an operator who reaches out before the guest complains buys +13 points of advocacy on top stays — and +32 on the middling ones, where a save actually decides whether they ever come back. The words show you which operator has the reflex.

7,088 recovery events · within-band lift (the raw +85/−2 headline is mostly rating mix)
Where the read stops

The fastest way to trust a number is to be told what it can't do.

So here's the perimeter, volunteered.

Inference
Return is inferred from the guests' own words, not from booking records. We report it as a signal of intent — never a confirmed re-stay.
Source
It's public-review intelligence — no PMS, no operator data, no guest PII. Public reviews are the whole input — so the read needs no cooperation, no login, and no operator relationship with the property. Anything resting on a handful of reviews is flagged directional, with its sample size and a 95% confidence interval.
Skew
The cohort skews European and Portugal-heavy. Read it as one well-read slice of the market, not a global law.
Not "truer"
A booking score and a loyalty read measure different things. The Gap is the distance between them — and that distance is where the decision lives.

Everything above re-runs from the same public base — we'll show you the working before you ever pay for a number.

The named version, on your asset

The score is public. The read is not.

Everything above is a hotel we don't operate — read from public reviews alone. What the rating page never gives you is the finished read of that property, one-off, in the hands of the seat deciding on the asset rather than running it. The next one is named, and it's yours — a target before the bid, a rival across the street, the operator up for renewal.

You just watched a rating fail to grade two hotels. Your pipeline is full of the same ratings.

The first read is €3,000 for one hotel you name, 14 days, delivered first — and you pay only if it surfaces a finding worth acting on. Take the call first; the point of it is to see whether there's a finding worth paying for, not to sell you one.

Know someone underwriting a hotel on its rating? Forward this — the number that breaks them out of it is the first one up the page.

Book a 20-min read intro