OpenAI just saved Booking.com

The AI checkout that could have broken the OTA's grip on hotel distribution just got shelved — and someone still needs to close the deal

In January, we wrote that Booking.com had built one of the internet's great businesses by controlling what travelers see, and that agentic AI was transferring that control to a dozen platforms it couldn't own or replicate. The thesis was directionally sound. The timeline was wrong.

This week, OpenAI quietly scaled back its plans to enable direct purchases inside ChatGPT. The company had been building toward a world where travelers could discover, compare, and book entirely within the chat interface — no redirect, no OTA, no Booking.com in the chain. That vision has been shelved. For now, ChatGPT will recommend and send. Someone else closes the deal.

That someone is still Booking.com.

The market reached the same conclusion in real time. Booking Holdings shares jumped over 8.5% on Thursday — the day the news broke. The market's read was immediate: OpenAI's retreat is good for the incumbents.

What actually changed

The original erosion thesis rested on a specific sequence. AI platforms would capture discovery first, then extend naturally into transaction. Once they owned the checkout, Booking.com's role collapsed to inventory supplier — a commodity position with no pricing power and no margin defence.

OpenAI's retreat disrupts that sequence. Building full transaction infrastructure inside a conversational AI interface turns out to be genuinely hard. Payments, inventory synchronisation, merchant agreements, dispute resolution, cancellation logic — these aren't product features. They are operational businesses. Booking.com has spent two decades building exactly this infrastructure, and it still consumes billions in annual investment to maintain.

The complexity isn't a bug in Booking.com's model. It's what keeps everyone else out.

Every AI platform that decides commerce is too difficult to own is implicitly deciding that someone else should own it. Right now, that someone has 31 million listed properties across 220 markets, a merchant model that generated roughly 60% of 2024 revenue, and payment rails optimised for markets that most technology companies would rather not deal with. OpenAI's retreat validates everything Booking.com built.

The referral engine problem

There is a version of this story where the retreat doesn't help Booking.com at all. If ChatGPT — which now has 900 million weekly active users — becomes a recommendation layer sending travelers to book externally, the critical question is: where does it send them?

The optimistic read for Booking.com is that structured data, broad inventory, and brand recognition make it the default destination. Travelers get a recommendation, click through, and complete the booking on a platform they already know.

But referral economics are not the same as discovery economics. Booking.com built its margins by charging hotels for placement inside an interface it controlled. As a referral destination, it becomes one of several options an AI platform might send traffic toward — alongside hotel direct websites, Expedia, regional OTAs, and any channel manager with an MCP connector. The visibility hierarchy moves upstream.

The question isn't whether Booking.com survives as a transaction destination. It will. The question is whether it can charge for placement when it no longer controls the interface where placement decisions are made.

The metasearch trap, revisited

This has happened before.

When metasearch emerged — Trivago, Kayak, Google Hotel Ads — the premise was that comparison-layer platforms would commoditise OTAs by owning the discovery moment. Hotels would bypass OTAs and bid directly for traffic. OTA margins would compress as travelers shopped around.

What actually happened: Booking.com and Expedia out-spent everyone on metasearch placement, absorbed the channel, and used it to reinforce their dominance rather than lose it. The comparison layer became another distribution cost — eventually paid by hotels, routed through the OTAs. Booking Holdings spent $7.3 billion on marketing in 2024 alone, roughly 31% of revenue. That muscle was built for exactly this kind of arms race.

If ChatGPT and its peers become referral engines rather than transaction platforms, Booking.com will optimise for AI visibility the same way it optimised for Google and metasearch. It has the budget. It has the data. It has teams who have run this playbook before. Hotels, once again, will fund the placement through commission structures they don't control.

OpenAI retreating from transactions doesn't eliminate this dynamic. It may accelerate it.

What the partnerships look like now

In January, we characterised Booking.com's partnerships with OpenAI, Copilot, and Gemini as capitulation — inventory supplier relationships dressed up as strategic positioning. That characterisation needs revisiting.

If AI platforms stay in discovery mode, those partnerships are something more useful: preferred referral relationships with platforms that collectively reach billions of active devices. Booking.com isn't surrendering the transaction layer; the AI platforms are declining to compete for it.

The partnership logic flips. Booking.com isn't feeding inventory to platforms that will ultimately replace it. It's establishing itself as the destination those platforms send travelers to when they're ready to book. That's a materially different position than the one we described in January.

The caveat that matters

OpenAI has scaled back its shopping plans. It has not abandoned them. The technical complexity that makes full in-chat commerce difficult today will not make it impossible permanently. Every month that passes is a month in which payment infrastructure matures, inventory API standards consolidate, and the marginal cost of adding a checkout layer to a recommendation falls.

The other platforms haven't retreated at all. Google is still building agentic booking for Gemini. Amazon has Alexa+ in market. Apple is integrating Gemini into a rebuilt Siri on two billion devices. The competitive pressure to own the transaction — and the margin that comes with it — hasn't gone away. It has been temporarily deferred by one player.

Booking.com's lifeline is real. It is also contingent.

What this means for hotels

For hoteliers, the temptation is to read OpenAI's retreat as breathing room. It isn't. It is a delay — and the delay favours the incumbent with the largest budget, the deepest inventory, and the longest history of absorbing new distribution channels at the expense of the properties listed on them.

The window in which a truly neutral AI transaction layer might have given hotels direct access to travelers — without Booking.com as the obligatory intermediary — was always narrow. It just got narrower. When the next version of that window opens, the question will be the same one it has always been: whether hotels have built the direct booking infrastructure to compete, or whether they are once again waiting for someone else to solve the problem.

Every previous version of this story ended the same way. The hotel brands and tech companies that write a different ending are the ones worth watching.

by Markus Busch, Editor/Publisher Hospitality.today

Read also The erosion of Booking.com

Enjoying this analysis? Hospitality.today delivers daily insights on hotel distribution, AI trends, and travel commerce — straight to your inbox. Subscribe for free at Hospitality.today →