Hotels in the cart

Google just filed lodging under the same commerce rail as sneakers and groceries

In November, agentic hotel booking was a travel story. By May, it was a shopping one.

The shift is easy to miss, because both versions end in the same place: a guest completing a reservation inside a conversation with Google, never touching a hotel website or an online travel agency. But the route Google now describes to get there has changed. And where the booking now lives is the part worth reading closely.

At the Phocuswright Conference last November, Julie Farago, Google's vice president of engineering for travel and local search, framed it in travel's own language. The goal was to let users book flights and hotels directly in AI Mode, built out with a roster of travel partners — Booking.com, Expedia, Marriott, IHG, Choice, and Wyndham. Google's accompanying post described the finish line the way the industry would recognize it: compare options, refine them, and then complete the booking with the travel partner of the user's choosing. Partners. Travel partners. The booking lived in the travel surface, routed to the companies that have always carried the inventory.

On May 19, at Google I/O, hotels reappeared somewhere else entirely. Not in a travel announcement. In a shopping one — "Introducing the Universal Cart," written by the VP and GM of Ads and Commerce. Hotels weren't the subject. They were a clause: Google's Universal Commerce Protocol checkout, the post said, is expanding to more verticals, starting soon with hotel booking and local food delivery.

Read the two side by side and the migration is plain. A hotel reservation went from a travel feature built with travel partners to a vertical of a retail commerce protocol, named in the same breath as food delivery, inside a launch built around Nike, Sephora, Target, Walmart, and Shopify merchants. Same destination. Different home.

So the question here is not whether Google can put a hotel in the cart. It's what the cart was built to hold.

What UCP actually is

Google's own documentation is unusually direct on this point. The Universal Commerce Protocol is presented as an open standard for the future of commerce, built to turn AI interactions into instant sales across AI Mode in Search and Gemini, running on merchants' existing Merchant Center shopping feeds. And in the line that matters most for lodging, the documentation states that UCP's "primitives map 1:1 to standard retail operations such as checkout."

One to one. With retail.

The architecture follows from that. Merchants pick a native checkout wired straight into Google's surfaces, or an embedded, iframe-based path reserved for those with bespoke branding or more complex flows. The published roadmap names multi-item carts, loyalty-program account linking, and post-purchase support oriented to tracking and returns. Payments run on a separate protocol, AP2, which lets a user set guardrails for an agent — a spending ceiling, specific brands, specific products — and writes the result into what Google calls a tamper-proof digital mandate. Throughout, the merchant of record stays the brand. Google's claim is that whichever way you buy, "the brand stays the merchant of record."

UCP is interoperable with MCP rather than opposed to it; the protocol supports an MCP binding as one transport among several. Which fits the way we've framed the two layers before — MCP as the context layer that tells an agent what exists and whether it's available, UCP as the commerce layer that closes the transaction and moves the money. A complete agentic booking needs both. What changed on May 19 is which layer Google named when it named hotels. It named the commerce one. Its own.

What a room isn't

Here is where the one-to-one mapping starts to strain.

A retail checkout settles a final, payable total against a SKU (stock keeping unit) that sits in a warehouse, ships, and can be sent back. The post-purchase lifecycle is tracking and returns. That is the lifecycle the roadmap names, and it is the wrong shape for a room.

A room doesn't ship. It doesn't return. It cancels — sometimes free until a deadline, sometimes never — depending on a rate-level attribute the guest chose at booking that binds different obligations on each side. It gets modified: a date moved, a night added, an early departure. It carries a price that, outside the markets now mandating all-in display, often isn't final when it enters a cart — and even where total-price rules apply, occupancy and city taxes can still resolve at the property, in local currency, under local rules. The settled total a retail cart assumes, a hotel rate frequently doesn't have yet.

Some of the roadmap does bend toward lodging. Loyalty-program account linking is genuinely useful to a chain, and it's listed. But a spending guardrail is not a grasp of the terms. An agent cleared to spend up to four hundred dollars on a hotel is not an agent that distinguished a non-refundable rate from a flexible one, bound the cancellation deadline, and understood what the guest owes when plans change. AP2's mandate authorizes a payment. It does not, on its face, authorize comprehension of the contract that payment enters into.

Then there is the merchant of record — the assertion doing the most quiet work in the announcement.

For Nike, merchant of record is settled. Nike sells the shoe. For a hotel rate, it depends entirely on the channel the rate traveled. The November roster put two merchant models in one list: hotel groups that sell their own inventory, and online travel agencies that resell it under arrangements that have spent two decades being litigated, renegotiated, and fought over — precisely because the answer to "who is the merchant" was never fixed. A Marriott rate reaching the cart directly is one merchant of record. The same room reaching the cart through an OTA is potentially another. A rate that passed through a wholesaler, then a reseller, before it surfaced is a third conversation entirely. "The brand stays the merchant of record" answers a question retail had already closed. In lodging, it was never settled — it was the whole fight.

The rail underneath

Step back from the mechanics and the strategic shape is starker than the November framing let on.

Google is assembling every layer of the agentic stack at once. The demand surface — Search, Gemini, AI Mode — where the guest's question gets asked. The commerce rail, UCP, that closes the sale. The payments protocol, AP2, that moves the money. No other AI platform holds all three. Most hold one.

UCP is the connector layer in that picture: the rail that OTAs and chains alike plug into to be bookable inside Google's surfaces. And a connector standard co-developed and stewarded by the same company that owns the demand and the payments is a different proposition than a neutral one. Adopting it means accepting a transaction surface whose terms, ranking, and roadmap are set elsewhere. Naming hotels as a UCP vertical is the moment that proposition went public for lodging.

Two pictures that don't yet line up

A flag on interpretation here, because the public record genuinely doesn't resolve this.

November describes one mechanism: book in AI Mode, with the travel partner of your choice. May describes another: hotels as a vertical of the Universal Commerce Protocol, the retail rail. These are not obviously the same system. One is partner-routed and travel-native; the other is feed-driven and retail-shaped. Whether they converge into a single booking path, run in parallel, or one quietly supersedes the other is stated nowhere Google has published. The reading offered here — that the May framing signals the direction of travel — is inference, not confirmation. Google has named a vertical. It has not shipped a hotel booking flow, a lodging-specific specification, or a partner list for this phase.

What is on the record is the migration itself. Across two announcements, five months apart, the same capability changed conceptual homes — from the language of travel distribution to the language of retail commerce.

What's open

Which leaves one question, and it's structural, not rhetorical.

A retail rail can be extended to lodging in one of two directions. The rail can bend — taking on the cancellation logic, the fork between refundable and flexible, the unsettled total, the contested merchant of record, all the things that make a booking not a purchase. Or lodging can be flattened to fit the rail — booked, increasingly, as though a room were a SKU, with the parts that don't map treated quietly as edge cases.

The November-to-May migration doesn't settle where this lands. It shows which way the gravity runs. Not a verdict. A direction.

The cart is being built either way. What's still open is whether hotels get to stay hotels inside it.

by Markus Busch, Editor/Publisher Hospitality.today

Read also: Google was never going to save you

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 →