On August 27, Google switched on hotel booking inside AI Mode for users in the United States. A traveler describes the trip, compares rooms in the conversation, reads the cancellation terms and pays with Google Pay without leaving the chat. The launch partners were Booking.com, Expedia, Hotels.com, Priceline, Trip.com, Choice, Hilton, IHG, Marriott and Wyndham.
In October 2025, Booking.com and Expedia had already opened apps inside ChatGPT. Since then hotel groups have followed with their own assistants and apps, and technology vendors have started selling direct-booking connections into the same chats on a flat fee, with no commission.
I have spent 22 years running distribution for leisure hotels in Northeast Brazil: OTAs, bedbanks, tour operators, air charters, metasearch and the direct channel. Every one of those channels arrived the same way. First as a novelty, then as volume, and only later as a contract someone else had already written. AI answer engines are at the novelty stage. That is exactly when a hotel still has room to set the terms.
What is actually new
It is not the interface. Search moved to apps, apps moved to metasearch, and hotels adapted. Two things are different this time.
The answer replaces the list. On a results page the traveler sees ten options and chooses. In a conversation the assistant recommends two or three, and the traveler often stops there. Being on the shelf is no longer enough. The hotel has to be the answer.
Nobody has settled the economics. The question is still open: does the assistant charge the hotel? Today three models coexist. The OTA sells inside the chat and charges its usual commission. An aggregator connects the hotel to the assistant and charges a new fee that did not exist before. Or the hotel connects its own inventory and keeps the margin, paying for the technology instead. Google and OpenAI, for now, prefer that the partner charge the card and own the guest service. The value is moving to whoever owns the conversation with the traveler, not to whoever issues the invoice.
In September we showed that agentic booking has already happened in Brazil (article in Portuguese). The next step is deciding on what terms the hotel enters this channel.
Six rules I would set now
1. Decide who sells on each assistant. For every surface (Google AI Mode, ChatGPT, Gemini, the next one), write down whether the hotel sells directly, through an OTA or through an aggregator, and at what cost. Decide this on paper before the first booking arrives through that channel, not after.
2. Extend rate parity to the conversation. An assistant quotes the price it can read. If an OTA shows a lower rate than the hotel website, the AI will repeat it in a confident sentence, with no list around it to soften the gap. Parity audits need to include what the assistants are saying, not only what the OTA extranets show.
3. Treat content as inventory. Assistants recommend what they can understand: room types, real policies, distances, what is included in the rate. A hotel with one clean, machine-readable source of truth for these facts gets cited. A hotel with three versions of its cancellation policy gets skipped, or worse, gets quoted wrong.
4. Put a ceiling on the mix. An AI-originated booking that arrives through an OTA app pays the full OTA commission. That can be fine as acquisition, but only inside a target. Set the share you accept per channel, the same way you would for any OTA or bedbank allotment.
5. Measure it as a channel. Most hotels today cannot tell how many bookings started in an assistant. Separate the traffic and the bookings that come from AI referrals in your reporting, and review them in the weekly commercial meeting next to the other channels. What is not measured gets negotiated by someone else.
6. Build the direct path before you need it. The open protocols that let an assistant read live availability and rates already exist, and connecting to them does not require changing the PMS or the booking engine. The cost of being ready is small. The cost of arriving late is another decade of commission.
Why this matters for Latin America
The launch in AI Mode is United States only, in English. Latin America is likely next, and it will meet a market where OTAs already concentrate a large share of leisure demand and where many independent hotels still update rates by hand. Northeast Brazil and the Caribbean share the same pattern: sun-and-beach demand, all-inclusive product, strong seasonality and dependence on air capacity. In markets like these, the hotels that arrive at the assistant with clean content, a parity rule and a measured mix will keep their direct share. The others will learn the commission from the invoice.
Every distribution channel I have worked with eventually set the rules for the hotels that did not set their own. This one is still open. It will not stay open for long.
Sources
- Skift, "Google's Agentic Hotel Booking Tool Comes to AI Mode", August 27, 2026. https://skift.com/2026/08/27/googles-agentic-hotel-booking-tool-comes-to-ai-mode/
- OpenAI, "Introducing apps in ChatGPT and the new Apps SDK", October 6, 2025. https://openai.com/index/introducing-apps-in-chatgpt/
- Hospitality Net, "The Agentic Booking Question No One Has Answered: Who Gets the Commission?", August 2026. https://www.hospitalitynet.org/opinion/4133767/the-agentic-booking-question-no-one-has-answered-who-gets-the-commission
- Accenture Newsroom, Radisson Hotel Group and Accenture on ChatGPT, July 28, 2026. https://newsroom.accenture.com/news/2026/radisson-hotel-group-and-accenture-redefine-travel-discovery-on-chatgpt
- Hotel Management, Lighthouse launches a direct booking app in ChatGPT, March 2026. https://www.hotelmanagement.net/tech/lighthouse-launches-hotel-direct-booking-app-chatgpt
Sources accessed on October 12, 2026. Unsplash photographs used under the Unsplash License, cropped. None of them shows a hotel served by Sereia.