The hotel discovery layer: AI search and the disappearing review

Reviews used to inform the choice. Now they inform the AI that informs the choice

A traveler asks ChatGPT to compare three hotels in Lisbon. The AI answers with a single paragraph — atmosphere, room quality, breakfast, neighborhood, common complaints, who the property is best for. She reads the paragraph and chooses one. She never opens TripAdvisor. She never scrolls Booking.com reviews. She never reads a single guest comment.

The reviews still exist. They were read — by an AI.

This is the function AI search has taken from the booking channel. Reviews haven't disappeared. Their primary reader has changed. The traveler used to read reviews and form a judgment. Now the AI reads the reviews, forms the judgment, and hands the traveler a summary.

What reviews used to do on the booking channel

For two decades, reviews on the booking channel were the friction that resolved the shortlist into a choice. A traveler with three properties in mind read a handful of reviews on each, looked for patterns, weighed photos against complaints, and picked one.

Review platforms were built around this. TripAdvisor's rank, Booking.com's review score, Google's star rating, the count of recent reviews — these were published signals hotels could see, measure, and compete on. Reputation was a visible distribution input. The work of interpreting reviews happened in the traveler's head, on the channel, against signals the hotel could read at the same time the traveler did.

That work has moved.

What changed

Three structural shifts have moved review reading off the booking channel and into AI summaries.

The first is that AI summaries are replacing review scrolling. A recent hotelrank.ai study of 4,000 Google AI Mode hotel queries across eight cities found that OTAs receive only 3.6% of clicks despite providing 46% of the sources the AI cites. The reviews are being read; they just feed a system that routes the traveler somewhere else.

The second is that the AI is now the primary reader of reviews. Hotels have spent two decades writing thoughtful responses, soliciting reviews from satisfied guests, and managing reputation across platforms — work that assumed a human audience. The audience has changed. The reviews still inform the choice, but they inform the AI first, and what the traveler reads is the summary.

The third is that hotel reputation has moved from a published rank to an interpreted signal. The AI's answer isn't a number. It is a paragraph, generated for a specific query, weighted in ways no hotel can see. The same property may be summarized differently for "romantic weekend in Lisbon" than for "business trip Lisbon," and differently again for "where to stay near Alfama." Reputation is no longer a position. It is a description, written on demand.

The asymmetry

Hotels can see their TripAdvisor rank. They can see their Google star rating. They cannot see how an AI summarizes them. There is no equivalent dashboard, no rank to track, no auction to bid in, no published signal to optimize for.

The intermediation creates a deeper problem. When a traveler read reviews on Booking.com, the relationship was direct — review to reader. The hotel could read the same reviews the traveler read, identify what was driving sentiment, respond to specific complaints, watch the score move in response. With AI summaries, the model sits between the review and the reader. The hotel writes a response to a negative review and has no way to know whether the model surfaces it, weights it, or ignores it. Reputation management has become indirect.

The asymmetry compounds because the answer is not stable. The same reviews may produce different summaries for different queries, different users, even different versions of the same AI. A property that ranks well in one summary may not rank in another. The signals are not published. The weights are not disclosed. The appeal mechanism does not exist.

What this means

Hotel reputation has been a measurable distribution input for two decades. Star ratings, review scores, TripAdvisor rank — these were observable, comparable, and contestable. AI search has changed all three. Reputation is no longer a visible position in a ranked list. It is a paragraph generated on demand, in a window the hotel cannot see into.

For commercial teams, a familiar discipline — review management, reputation strategy, response cadence — is now being practiced with a different audience in mind. The traveler still reads. But the traveler increasingly reads what the AI says about the reviews, not the reviews themselves. Hotels are writing for two audiences now, and only one of them ever responds.

Next in this series: Metasearch, eaten by what it invented

by Markus Busch, Editor/Publisher Hospitality.today

Enjoying this analysis? Hospitality.today delivers daily insights on hotel distribution, AI trends, and travel commerce — straight to your inbox. Subscribe for free at Hospitality.today →