AI Can Find Your Hotel. But Can It Pay for It?

Table of Contents

  1. Introduction: From Search to Execution
  2. Agentic Bookings: Moving from Idea to Reality
  3. Booking a Hotel Is Not the Same as Buying a New Pair of Sneakers
  4. AP2: Solving the Authorization Problem
  5. The Most Interesting Question: Who Actually Gets Paid?
  6. What Could This Mean for Hotels?
  7. What Could This Mean for OTAs?
  8. The Next Challenge: Trust and Responsibility
  9. Conclusion
  10. Glossary
  11. Sources

Introduction: From Search to Execution

The real question in AI travel is no longer whether it can find a hotel. It is whether it should be trusted to book and pay for one.

For years, innovation in travel has been about discovery — better search, better recommendations, better personalization. AI is now pushing that boundary further, from suggesting options to actually completing bookings on behalf of users.

But the moment an AI agent moves from suggestion to execution, a completely different layer of complexity appears: payments, authorization, liability and control.

Because booking a hotel is not just a search problem. It is a financial transaction that happens across multiple systems and comes with contracts, rules, and real money at stake.

That is why the conversation is shifting — from what AI can find to what it should actually be allowed to do.

Agentic Bookings: Moving from Idea to Reality

Over the last few months, several announcements have signaled a significant shift in how travel could be reshaped in the future:

  • September 2025: The Agent Payments Protocol (AP2) was introduced. 
  • November 2025: Google expanded agentic booking capabilities in AI Mode for selected reservation categories and announced plans to enable flight and hotel bookings directly in AI Mode in the future.
  • May 2026: Google I/O announced Universal Cart and confirmed plans to expand the Universal Commerce Protocol (UCP) into hotel bookings.
  • June 2026:  Mastercard, ING and Worldline announced a successful live end-to-end agentic payment transaction in Europe.

Taken together, these developments suggest that AI-driven booking is moving from concept to reality.

Although hotel booking through UCP is not yet broadly available, Google has already signaled that hotels are one of the next verticals planned for expansion. The direction of travel is becoming increasingly clear.

Booking a Hotel Is Not the Same as Buying a New Pair of Sneakers

Much of the conversation around AI commerce has been shaped by retail. If an AI agent can buy a pair of sneakers, why shouldn’t it be able to book a hotel room?

The answer lies in the complexity of travel transactions.

Buying sneakers is relatively straightforward. There is usually one merchant, one payment, one fulfillment process and limited post-purchase interaction.

Travel works differently.

A hotel room may be booked months before arrival. Guests may modify dates, cancel reservations, request refunds or fail to show up entirely. Payment may be collected immediately, shortly before arrival or at the property itself. Tax rules, cancellation policies and regulatory requirements vary across destinations.

The technology stack is equally complex. Inventory and availability data often flow through property management systems, central reservation systems, channel managers, payment gateways and card networks before a reservation is confirmed.

This is why booking a hotel is not simply a search problem. It is also a payments problem.

AP2: Solving the Authorization Problem

One of the most interesting developments in this space is AP2, the Agent Payments Protocol.

Traditionally, a customer completes a booking by clicking a button and entering payment details. An AI agent cannot do that unless the traveler has explicitly delegated purchasing authority.

AP2 addresses this challenge by separating user intent from transaction execution through two distinct mandates.

The first is the Intent Mandate. This is where the traveler defines the boundaries within which the AI agent can operate. In a travel context, these instructions could include a maximum budget, preferred hotel category, location requirements, cancellation policies, loyalty preferences, or other booking criteria. Rather than approving a specific hotel, the traveler approves a set of rules under which the AI operates.

The second is the Cart Mandate. Once the AI agent identifies an option that satisfies the Intent Mandate, a transaction-specific authorization is generated for that particular booking. This mandate is tied to a single shopping cart and can only be used for that specific purchase.

Together, these two mandates create a framework in which the traveler remains in control while allowing the AI agent to act independently within clearly defined limits.

The Most Interesting Question: Who Actually Gets Paid?

When people first hear about agentic commerce, they often assume that Google will become the new travel intermediary.

The reality appears far more nuanced.

In Google’s current Universal Cart model, the brand remains the Merchant of Record. If a similar structure is applied to travel, hotels and travel companies would likely continue to own the financial transaction, while Google would orchestrate the experience around discovery, intent and execution.

The AI agent may initiate the process, but the payment itself would still need to move through existing commerce and payments infrastructure. In travel, that means merchant payment gateways, card networks, hotel systems, OTA platforms and the operational processes that sit behind every confirmed booking.

This means hotels and OTAs continue to manage taxation, invoicing, compliance and customer payments. Google orchestrates the experience. The industry processes the transaction.

This distinction matters because it reveals Google’s broader strategy. Rather than becoming another travel seller, Google appears to be building an orchestration layer that sits between discovery, decision-making and transaction execution.

What Could This Mean for Hotels?

In practice, many hotels still struggle with virtual cards, payment reconciliation and reservation modifications. Agentic bookings add another layer on top of an already fragmented ecosystem — one that is both financial and operational at the same time.

This is not only a payment challenge, but also a data and distribution challenge.

Hotel websites may therefore become more important as structured sources of information. They were originally built for human users, not machines. That distinction is starting to blur.

Structured content, inventory data and booking integrations are becoming just as important as traditional search visibility.

As AI agents increasingly shape the discovery and booking process, visibility will depend less on traditional marketing and more on how easily a hotel can be understood and booked by systems, not just people.

What Could This Mean for OTAs?

Much of the discussion around AI assumes that online travel agencies have the most to lose.

I am not convinced that the picture is so simple.

If AI agents become the primary interface for discovering and booking travel, the role of OTAs does not simply disappear. It shifts.

Instead of competing for user attention at the search and comparison stage, OTAs may increasingly operate behind the scenes as infrastructure providers — powering inventory aggregation, pricing logic, payment flows, customer service, and post-booking management.

In other words, the value may shift from front-end demand capture to back-end transaction execution.

The ability to present options to a traveler becomes less important than the ability to reliably fulfill a transaction initiated by an AI agent, across multiple suppliers, under different policies, currencies, and regulatory frameworks.

The Next Challenge: Trust and Responsibility

Technology may solve the authorization challenge. It may even solve the payment challenge. The bigger question could be responsibility.

What happens when an AI agent books a non-refundable room when the traveler expected a flexible rate? 

Who is responsible if a booking parameter is misunderstood? 

Who handles disputes when the transaction itself was technically valid, but the outcome was not what the traveler intended?

These questions extend beyond technology and into regulation, liability and consumer trust. And they may ultimately prove more difficult to solve than the booking process itself.

Conclusion

Travel is shifting from a search problem to an execution problem.

With UCP, AP2, and Universal Cart, AI is moving beyond recommendations into a model where agents can assemble and complete bookings within defined financial and operational boundaries.

This changes the core question for the industry: not just how hotels are discovered, but how intent becomes a valid, authorized and successfully settled transaction across a fragmented ecosystem.

Discovery becomes faster and more automated. The complexity moves underneath — into payments, authorization and responsibility.

What is still unfolding is: How quickly will the industry adapt to a world where AI agents do not just support decisions, but increasingly execute them?

Glossary

Agentic Commerce
A model in which AI agents can not only recommend products and services but also complete transactions on behalf of users within predefined limits and permissions.

AP2 (Agent Payments Protocol)
A payment authorization framework designed for AI-driven transactions. It enables users to delegate purchasing authority to AI agents through cryptographically secured mandates and transaction-specific permissions.

Universal Commerce Protocol (UCP)
An open standard that defines how AI agents communicate with commerce systems to discover products, build carts, and complete purchases on behalf of users.

Universal Cart
A persistent shopping cart designed to work across Google services and participating merchants, allowing users and AI agents to assemble and manage purchases across the shopping journey.

Merchant of Record (MoR)
The entity legally responsible for processing a transaction, collecting payment, handling taxes, issuing invoices, and managing payment disputes or chargebacks.

Online Travel Agency (OTA)

A digital marketplace that aggregates hotel and travel inventory from multiple suppliers and enables users to search, compare, and book accommodation. OTAs typically act as intermediaries between travelers and hotels, managing distribution, pricing visibility, and often parts of the booking and customer service lifecycle.

Sources

Google Cloud. “Powering AI commerce with the new Agent Payments Protocol (AP2).” Google Cloud Blog, September 16, 2025.
https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-to-payments-ap2-protocol

Google. “New ways to plan travel with AI in Search.” The Keyword, November 17, 2025.
https://blog.google/products-and-platforms/products/search/agentic-plans-booking-travel-canvas-ai-mode/

Mastercard. “Worldline, ING and Mastercard complete a live end-to-end European agentic payment in production.” Mastercard Newsroom, June 2, 2026.
https://www.mastercard.com/news/europe/en/newsroom/press-releases/en/2026/worldline-ing-and-mastercard-complete-a-live-end-to-end-european-agentic-payment-in-production/

Google. “Introducing the Universal Cart and more ways to help you shop.” The Keyword, May 19, 2026.
https://blog.google/products-and-platforms/products/shopping/google-shopping-cart/