53 points
The gap between two questions in one market
In the United States the hotel question held 96% of its sources and the flight question 43%. Same market, same day, same method.
Travel and hospitality
See which travel brands AI recommends before the booking starts.
Travellers now ask an assistant where to book before they open a booking site. EchoWi runs those questions across the AI engines your customers use, repeatedly, with country and language fixed, and shows you which brands are named, which sources built the answer, and where you are missing.
AI answer
What held in every run
- Metasearch and comparison sites all of them
- One brand, a local site per market both markets
- Sites built to close a booking none
What we measured
Four findings from one measurement
Same question, Google AI Overview, country and language set explicitly on every run. Here is what came back.
3 points
39% and 42%. The two questions describe each other there, and in the US they do not, which is why a category average tells you nothing.
24 of 25
Sources that held on the strongest question
And 24 of 26 in a second session, the same twenty four domains by name. A high number checked twice is a map you can work from.
5 of 35
Sources that held on the weakest
The Spanish flight question. Thirty of its thirty five sources appeared exactly once and were gone in the next run.
Two questions, hotels and flights, in four markets on three AI surfaces, all on 10 August 2026. Every figure links to the study. Read the four-market measurement.
The pattern is the same every time we look
The brands that survive in these answers are not the ones spending most. They are the ones publishing content that compares.
- Every stable booking domain in the Spanish run was a metasearch engine. Comparing is what they publish, and comparing is what the question asks for.
- The transactional sites, the ones built to close a reservation, did not reach the stable set in either market. A calendar and a price grid give a retrieval layer nothing to lift when the question is which one to use.
- The one travel brand cited reliably in both markets runs a separate local property per market rather than one translated site.
- We found the same split in two unrelated sectors, so this is not a quirk of travel.
A market where nothing holds is a market you can enter
The most useful thing a measurement can tell you is that AI cited your own page and still did not name you. That happened to a Spanish hotel chain in our run: its official site was the first source in the answer, ahead of every directory, and the chain is absent from the text. When the model has already read your page, the excuses are gone. Your site is crawlable, your content is readable, your page was judged relevant, and it still did not put you in the summary. What is left is a content problem, and content problems get fixed.
One booking question, four markets, four different answers
We asked which website is best for booking a hotel, and then for booking a flight, in four markets with country and language fixed on every run. The two columns to compare are how many domains were cited and how many of those held across every run.
- Domains cited
- Held in every run
United States, hotels 24/25
Spain, hotels 10/21
France, hotels 10/20
Germany, hotels 7/18
United States, flights 9/21
Spain, flights 5/35
Germany, flights 11/26
| Market and question | Domains cited | Held every run |
|---|---|---|
| United States, hotels | 25 | 24 |
| Spain, hotels | 21 | 10 |
| France, hotels | 20 | 10 |
| Germany, hotels | 18 | 7 |
| United States, flights | 21 | 9 |
| Spain, flights | 35 | 5 |
| Germany, flights | 26 | 11 |
Google AI Overview, AI Mode and Gemini, measured 10 August 2026, two runs each, country and language set explicitly. France on flights answered once rather than twice and is left out, because with a single run every domain cited is trivially stable and the column would read 11 of 11. The spread is the finding: in American hotel answers almost everything cited came back every time, and in Spanish flight answers five of thirty-five did.
What a revenue and distribution team gets
Numbers a commercial director can act on, per market, not one score for the group.
Your buying questions, not ours
The questions a traveller types before they book, without your brand in them. A query that already names you measures nothing.
Every market run separately
Country and language fixed on each run. The same question answers differently in every market, and an average across four can hide being invisible in three.
The sources, not just a score
Which domains built the answer, so you know who to be published alongside rather than only that you were missing.
Repeated, so the number means something
Answers vary between identical requests. One check cannot tell a stable citation from a coincidence, so we do not report one.
We confirm the answer layer exists in each source market before you pay to watch it
AI coverage is not universal. Measuring five categories in Spain in August 2026, two returned no AI Overview at all across thirteen runs, with no errors. A brand buying monitoring for one of those categories today would be paying to watch a surface that does not render. Flight booking and hotels do get answers in Spain, which is why this page exists, but we run the check before quoting anyone.
Who we do this with
We work with travel and distribution brands, and Phocuswright and PhocusWire are in our logo wall. And because we publish what we measure rather than what suits us: a brand we track appeared in none of the runs behind the figures above, and we printed that in the study too.
What travel teams ask us first
The questions that come up on the first call, answered here so you do not have to book one.
Our brand is named in the answer but our site is never a source. Why?
Because a booking question is comparative and a property or airline site describes one thing. The answer is assembled from sites that rank, compare and aggregate: metasearch, review platforms, guides and directories. That is why being named and being cited move independently, and why a measurement that only reports whether you were mentioned hides the half you can act on.
Why does the same question look so different by market?
Because the source set is different, not merely smaller. Asked which site is best for booking a hotel, the American answer cited twenty-five domains and twenty-four of them came back in every run; the German answer cited eighteen and seven held. A stable answer set means the positions are taken and hard to move. An unstable one means the answer is still being decided, which is the cheaper market to enter.
Does this measure OTAs and metasearch as competitors?
It measures whoever the answer is built from, which in this sector is usually them rather than another hotel or airline. That is the useful part: the list tells you which intermediaries the answer depends on in each market, so you know where being present actually changes what a traveller reads.
Can you measure by destination or by property rather than by brand?
Yes, and it is often the sharper question. A group can hold a strong brand answer and disappear on the destination question that decides a booking, because those are different queries with different source sets. Each is run with its own country and language rather than averaged into a group figure.
Is there an answer layer for every travel question?
No, and that check comes first. In our August 2026 sweep of results pages, questions asking which local business is best returned no AI Overview at all across every run, while country-level and destination-level travel questions did. If one of your questions has no answer layer, optimising for it is optimising for a box that is not on the page, and we say so before quoting.
What does the first measurement deliver?
Your booking and destination questions run repeatedly on the surfaces that answer them in each source market, with which brands were named, which domains built the answer, and which of those held across every run rather than appearing once. The stability column is what separates a source worth pursuing from noise, and it varies more between markets than anything else we measure.
See what AI says about your brand
Send us your site and the market you sell into. We will run five buying questions and show you the brands that appeared and the sources the answer was built from.
Measure my booking questions (opens in a new tab)No card. If a source market of yours has no answer layer yet, we say so and we do not bill for watching it.