How to win the AI booking recommendation
The guest asks an assistant where to stay and gets one short answer with a handful of names. Whoever is in it does not need to be first in a list of ten blue links, because there is no list.
AI sessions measured across the sites we track, last 30 days.
Quick answer
Winning the AI booking recommendation means being the property an assistant names and links when someone asks where to stay or what to book. Two measurements say the window is open now. In a sweep of the public ad library on 3 September 2026, Booking.com was already running ChatGPT ads titled with individual hotel names, while Melia, RIU, Barcelo, Iberostar and Palladium had none. And in the AI channel we measure, ChatGPT sends more than twenty times the next assistant, with the live 30 day counts in the panel at the top of this page.
The intermediaries are already bidding on your name
The ad library finding is worth sitting with. An ad titled with a hotel name, bought by a platform that takes a commission on that hotel's rooms, placed inside the surface where the guest is deciding. Five large Spanish groups had nothing there on the same date.
That is not an argument for buying ads. It is evidence that the channel is being valued by the party with the best data on what a booking is worth, and it says something about where the recommendation is being decided.
The organic side of that surface is decided by what the model read. In the corrected measurement window, the answers about one cenote park in Punta Cana were assembled from 784 citations across 271 domains, of which the measured site supplied 73 and ten booking and ticketing brands supplied 144. Both were in the room.
Named, linked, and the click you can count
A recommendation reaches a guest in two ways and only one of them is measurable. Your name inside the answer text travels even when nobody clicks. Your URL in the source list brings a session, and because ChatGPT tags its outbound links with utm_source=chatgpt.com, that session is countable.
The gap between the two is usually not where hotels expect. Of the 74 appearances in that window, 51 were a link with the site name absent from the answer, and exactly 1 was a mention with no link. The traffic was arriving while the name stayed invisible. That is one pass on 7 domains of our own travel network, so a single reading proves nothing about your property, but the shape of the gap is worth checking on your own questions.
If your analytics has no AI channel, both halves are dark. The visit from an assistant arrives with a referrer your reports do not recognise, so it is filed under direct traffic and the whole channel reads as zero. A channel that reads as zero is a measurement problem far more often than it is a business conclusion.
What to do about it this quarter
Start with the baseline, because everything above is somebody else's number until it is yours. A pass in reporte.ai is a fixed list of the questions your guests ask, run across ChatGPT, Claude, Gemini, Google AI Mode and Perplexity, stored with the full answer and its sources so you can see who beat you and on which page.
Then close the two gaps separately. The pages you own are covered in content that AI cites, and the pages you do not own in off-site mentions. Rerun the pass after each batch, so the movement is measured rather than assumed.
FAQ
Yes. In a sweep of the public ad library on 3 September 2026, Booking.com was running ChatGPT ads titled with the names of individual hotels. Melia, RIU, Barcelo, Iberostar and Palladium had none. That is one intermediary bidding on the names of the properties themselves.
The click, yes. ChatGPT tags its outbound links with utm_source=chatgpt.com, so the session arrives labelled. Without a tool that reads that label, the same visit lands in your analytics as direct traffic and the channel looks like it does not exist.
Across the sites we measure, ChatGPT sends around nine of every ten AI sessions and more than twenty times the next assistant. The panel at the top of this page carries the live 30 day counts, and those sites are our own travel websites rather than hotel clients, so read the concentration and not the volume.
You have to be in the same answer, which is a lower bar. In the corrected window the measured site was cited 73 times while ten booking and ticketing brands took 144 between them, so both were present and the guest chose.
What the sources it read say about it. In that window the answers drew on 784 citations across 271 domains, so the recommendation is assembled from other pages rather than issued from a ranking you can buy into.
Not on its own. Being named travels without a click, being linked brings the session. Of the 74 appearances we measured, 51 were a link with no mention and only 1 was a mention with no link, so most places have the opposite gap to the one they expect.
With a baseline, because every claim above is only useful against your own numbers. Ours come from one pass on 7 of our own travel domains, and a single reading proves nothing. A pass gives you your questions, your surfaces and the sources that beat you, and the pixel tells you what that turns into on your site.