‹ Marketplaces

28 Labs Study · AI Discovery Intelligence

AI Visibility in Travel Marketplaces: who the engines recommend when a traveler asks where to book.

When a traveler asks an AI engine where to book a flight, a hotel, or a stay, the engine recommends a marketplace - and most marketplaces have no idea whether it is them or a competitor. We ran a study to make that visible: the share of AI recommendations each leading travel marketplace holds across the four major engines, on the high-intent queries that actually drive bookings.

5 destinations · 4 engines · intent-scored travel query set · share of AI recommendations · every figure confidence-tagged

Why travel

Travel is one of the first categories AI engines learned to answer directly.

Booking a trip is a research-heavy, comparison-heavy decision - exactly the shape of question travelers now hand to an answer engine instead of a search box. "Where should I book," "which site is cheaper," "is this one trustworthy" are questions the engines answer confidently, naming a marketplace as the destination. That makes travel a clean place to measure something every marketplace faces: when the engine becomes the front door, whose listings does it send the traveler to?

This study is not a ranking we publish to embarrass anyone. It is a method, run on a public category, to show what AI Discovery Intelligence measures - and what a marketplace can do with it.

How we ran it

Share of recommendations, not a count of citations.

We measure the share of AI recommendations a marketplace holds - the percentage of high-intent travel queries where the engine names it as an answer, against the competitors named alongside or ahead of it. A raw citation count tells you a brand appeared; share tells you whether it is winning the decision.

The query set

An intent-scored set across the traveler journey, run across five destinations so share is comparable city to city - weighted to the high-intent queries near booking, pricing, and trust, not the long tail of inspiration research.

The engines

Every query put to ChatGPT, Claude, Gemini, and Perplexity, with each response captured verbatim and scored for which marketplaces it recommended.

The read

Share of recommendations per marketplace, broken down by engine and by intent stage, with each finding tagged HIGH, MED, or LOW by how directly we observed it.

The query universe

The traveler journey, scored by what each stage is worth.

Travelers do not ask one question. They ask across a journey, and the stages near the booking carry the revenue. We weight what we measure to where the booking is decided.

01
Research
"best places to travel in spring"
→
02
Comparison
"which site is cheaper for flights"
→
03
Pricing
"cheapest way to book a hotel in Paris"
→
04
Availability
"last-minute hotels in Dubai"
→
05
Booking
"best site to book a stay in Bali"
→
06
Trust
"is this site safe to book on"

High-intent queries near booking, pricing, and trust carry most of the revenue. Open inspiration queries carry almost none. The share number that matters is the one weighted to where the booking is decided - not a vanity average across the long tail.

Destinations studied

New YorkLondonParisDubaiBali

The marketplaces studied

Booking.comExpediaAirbnbKayakTripadvisorAgodaHotels.comSkyscannerVrboTrivagoPriceline

What we found

The findings.

Findings pending live run

This study publishes once the run is complete. The structure below is locked; the numbers come straight from the measured data - we do not estimate a finding we have not observed.

  • Share of AI recommendations by destination - who leads New York, London, Paris, Dubai, and Bali, and where it flips
  • Headline share by marketplace, across all four engines and all five cities
  • Where the leader's grip is strongest and weakest, by intent stage
  • How the four engines diverge on who they recommend
  • Where the engines answer the traveler directly instead of sending them to any marketplace - the disintermediation signal
  • How the travel pattern echoes what we see in other marketplace categories

How we keep it honest

No one can prove a single AI recommendation caused a single booking, and we do not claim it. We label the confidence of every read, so the finding carries exactly the weight the evidence supports.

HIGH

Directly observed. The recommendation is present in the captured engine responses. We saw it happen.

MED

Inferred. Multiple independent signals converge on the same read, but no single source confirms it outright.

LOW

Industry intelligence. The read rests on category benchmarks and external patterns, not the run itself. Treated as directional.

What this study does not claim

The honest boundary.

We name our limits before anyone asks.

  • We measure recommendation behavior, not training data. This study reflects what the engines retrieve and recommend when answering travel queries. We do not measure how often a marketplace appears in an engine's training corpus - that signal is not reliably available from the outside, and anyone claiming to measure it is overpromising.
  • A snapshot is a snapshot. Engine behavior shifts as models update and the web is re-crawled. A single run is a point in time; the value is in measuring the trend, which is why we run it as a live read for clients rather than a one-off.
  • Public category, public method. This study uses public travel marketplaces to demonstrate the method. We never expose a client's data to make a point, and the cross-vertical comparisons here are anonymized.

If you run a marketplace

The same read, run on your category.

We can measure your share of AI recommendations on the high-intent queries that drive your revenue, tie it to the traffic and cost your own data already shows, and hand your teams the moves to win it back. A 60-minute working session walks a sample audit and scopes a live run.

Talk to Us AI visibility for marketplaces