Insights

When AI books the room, who gets the guest?

Fewer than one in every hundred room nights: that is the share Booking Holdings currently attributes to visitors referred by ChatGPT, Gemini and other AI chatbots, its chief financial officer said when the company reported second-quarter results in early August. On that number alone, you could shelve the subject for now.

The big players are preparing for it all the same. Expedia and Booking.com have had their own apps inside ChatGPT since October 2025, and Accor followed in January 2026. Since 27 August, travellers in the US have been able to book hotels directly in Google’s AI Mode, with Hilton, Marriott, IHG, Choice, Wyndham and the big OTAs on board from the start. And Apaleo, a provider of property management software, reports that use of its interface for AI agents rose by 70% between July and September.

My view: AI is barely taking bookings yet, but it is already doing part of the shortlisting. Whether a hotel makes that shortlist depends partly on whether a machine can read its data.

What the numbers actually say

In March 2026, YouGov surveyed more than 5,700 adults in the US, the UK and India on behalf of Expedia. Of those, 53% are comfortable letting AI suggest travel options. Only 8% feel comfortable booking through an AI platform, and 68% would rather book with a travel brand they trust, even when AI booking is available.

On the hotel websites analysed by the technology provider Lighthouse, AI assistants account for just under 1% of visits. In an analysis from March 2026, BCG cites Skift’s Travel Outlook 2025 as the source of the figure that 37% of travellers have used a large language model to help plan a trip. BCG does not say when, or among whom, this was measured.

My reading: people ask the AI early, book later, and so far rarely inside the chat. What matters most for a hotel, then, is whether it appears in the answer that names three hotels to a guest three weeks before the trip.

What happened in 2026

  • January: Accor launches its ALL Accor app in ChatGPT. Guests search in the chat and book on Accor’s own site. According to Skift, Accor pays no fees for this traffic for now, but expects that to change.
  • March: OpenAI steps back from built-in purchasing in ChatGPT (“Instant Checkout”). Merchants can use their own checkouts instead, while OpenAI concentrates on product discovery. Its reason: the first version was not flexible enough.
  • 27 August: Google’s AI Mode starts booking hotels, for now only in the US. Payment goes through Google Pay, while the hotel or OTA remains the merchant of record, sends the confirmation and handles any changes.
  • 8 September: In the EU, the hotel unit for direct suppliers in Google’s search results stops showing prices. At the end of September, Google also retires the feature it used to source third-party rates for Hotel Ads itself. Hotels that want to appear there with a price now supply their rates through Hotel Center or a connectivity partner.
  • 28 October: Apaleo reports the 70% increase. The interface is used mainly for internal work such as revenue management, forecasting, reservations and reporting. Agents that talk to guests are still the exception.

Europe has bought time

Booking in AI Mode is not yet available in the European Economic Area. I think it would be risky to conclude from this that the shift will pass Europe by.

It has already arrived here, just in a different form. Prices have gone from the hotel unit for direct suppliers, and the OTAs now sit above it. In November 2024, Google itself wrote that affected businesses, including hotels, airlines and small retailers, had reported up to 30% fewer free direct-booking clicks since earlier changes. That figure has not been independently verified. What I take from it: when a platform sits between guest and hotel, how visible the hotel is depends heavily on what that platform shows and where it gets its data from.

Who supplies the data that AI reads?

BCG sketches three possible paths: language models partner with the OTAs, with the global distribution systems (GDS) through which travel agents and OTAs book rooms, or directly with the hotel brands. In the first two, BCG says, brands could lose influence, because the AI draws on third-party reviews and descriptions.

For independent hotels, that need not be bad news. According to Lighthouse, independents gained a disproportionate share when ChatGPT began including far more links in its answers in 2025. The analysis puts this down to their content being more distinctive than that of chain hotels.

For hotel groups, the bottleneck lies elsewhere. The h2c Hotel Chain Tech Report 2026, supported by several technology providers including Apaleo, finds that 91% of the hotel chains surveyed use AI. According to Apaleo’s account of the report, 15% have introduced a standardised AI data access layer, such as MCP. And the State of Distribution 2026 report by NYU, RateGain and HEDNA finds that 55% of the hotels surveyed have changed their distribution strategy little or not at all, even though, according to the report, AI search is already producing measurable bookings.

What I would advise head office to do now

With shares this small, I would not start a dedicated AI booking project. But I think four things make sense under any of the three paths.

1. A single source for the facts. Room types, amenities, cancellation terms, dog policy, parking, accessibility: maintained in one place and distributed from there to the website, the OTAs and Google. When an AI finds three different versions, it is more likely to quote the wrong one.

2. Rates and availability through an interface. Anyone choosing or renewing a property management or reservation system in the next two years should ask whether AI assistants can access it, starting with read-only access. One standard for this is MCP, an open protocol that lets AI assistants talk to other systems.

3. Measure before you invest. Track visits from AI assistants as a separate channel in your web analytics, and check regularly what ChatGPT, Gemini and Perplexity say about your properties. Some of the errors in those answers can be traced back to your own data and fixed there.

4. Decide who owns it. If the property maintains the description, head office the brand and revenue management the rates, someone should also keep an eye on the overall picture a machine pieces together from all of that. That job belongs with sales leadership, not with IT alone.

Seen from the guest’s side

If my reading is right, many guests ask an AI early and book later with whoever they trust. Where that ends up depends in part on what the AI told them about your property beforehand, and whether it was true.

At 5213 LABS, we come at the same question from the other side. Much of the property data an AI search needs is also needed by any AI meant to work inside the hotel, whether on the phone, in the inbox or at shift handover. Put that data in order once and it serves both.

Sources

Next step

Which workflow costs the most hours every week?

20 minutes. One workflow. A clear next step.

Book a first conversation