Is “Book on ChatGPT” Hitting the Same Wall as “Book on Google”?

OpenAI just admitted what Google already learned with "Book on Google" back in 2022: people use AI to discover, not to buy.

I fully agree with Simone the reason is structural, not technical. A hotel booking is not a single transaction. It's a chain of responsibilities: payments, cancellations, refunds, tax compliance, inventory sync. OpenAI hadn't even built a system to collect sales taxes in the US when they launched their checkout. Nobody wants that operational liability. And users don't change buying habits built over decades just because the interface looks cleaner.

Benedict Evans shares it well: OpenAI has 900 million weekly users, but 80% send fewer than 4 messages a day. "A mile wide, an inch deep." No unique tech, no network effect, and a roadmap driven by the lab, not the customer. They tried to be the new OS, the store, the travel agent and the search engine all at once...

So what's next? The booking won't disappear from conversational platforms. It will be redistributed. The model that works is AI as orchestrator: discovey first, then handoff to the systems that already hold the inventory, the contract and the compliance layer.

The real question for hotel direct channels: is your inventory structured to be consumed by agents? Because that's the next battleground.

Conversational platforms will complete bookings, but not by simplifying hospitality's complexity. They will succeed when they can orchestrate it. Today a single reservation touches PMS, CRS, channel management, payment processing, cancellation logic etc. often different vendors, loosely integrated, running on incompatible data models. No conversational layer resolves that fragmentation from the surface.

But the deeper challenge is not booking. It is discovery. If the AI doesn't recommend your hotel, the booking question is irrelevant. This is the new distribution architecture: not search rankings or OTA sort order, but whether your property is the one the AI surfaces. That requires structured, machine-readable data most hotels don't yet provide.

What shifts the equation is connective infrastructure. Protocols like MCP allow AI agents to interact directly with hotel systems, creating discoverable, bookable inventory at dramatically lower acquisition costs. Large chains can build this now. For thousands of independents, AI-native aggregators will emerge as a critical new layer, absorbing the massive look-to-book query volume, caching inventory, filtering intent, and forwarding only high-conversion queries to hotel systems.

The question is not whether bookings happen inside a chat window. It is who builds the connective tissue between hotel systems and AI platforms, and who captures the value.

I’d agree. A retail transaction, even a high value one, is much simpler than a hotel or flight booking. Shopify is a top retail IBE, but its funnel is much simpler (view products, select product, pick colour, select size SMLXL, add to basket, add an upsell box of chocs, set delivery address, etc), than the well-converting IBEs in hospitality like Journey, Net Affinity, P3, Profit Room, Allora etc. Other than buying a car most retail purchases aren’t complex experiences. Amazon is deliberately one-click friendly. The Open AI team are struggling, especially given Claude Code was a code red for them, and Oracle is now regretting aligning with them on Datacentre openings. They have binned Sora and parked their PornBot. Allowing Oracle, On The Beach et al to embed their IBEs in Chat is much simpler and low risk. Google knows travel well. It bought ITA in 2011, which gave it a good understanding of the complexities of travel inventory. I think that Google understands that conversional interfaces are inherently limited, travel is visual, and it doesn’t want to alienate its PPC ad market buyers. Not yet at least. It can bide its time and try again if it wants.

Conversational commerce is the future. Full stop. Just no road comes without any bumps. Despite this recent pullback, the most likely path is one of embedded payments and reservation confirmation, whereby the chain of custody is kept by the hotel or travel agency but it's displayed as a screen within a screen. The option will then be pushed onto the user / traveler in terms of whether to embed the direct channel's booking engine or an OTA.

Secondly, consider how this pullback is being used as confirmation bias for those who have been skeptical about LLM-native bookings. One battle lost does not end a war. Focusing on the challenges of AI bookings should not overshadow the massive growth in top-of-funnel AI search. Just being the reservation flow isn't wholly native and frictionless doesn't preclude a massive number of potential customers from using the LLMs to discover hotels, wherever the booking link-out takes them. Every hotel now needs an AI search / AIO / GEO strategy.

Hotels should think of conversational platforms as discovery engines, not booking engines. The real opportunity sits earlier in the journey. Hotels need to be the source of truth in these environments. The content they publish across their website, booking engine, and digital channels must be consistent, structured, and owned. If a hotel is not surfaced as the authoritative answer in ChatGPT or similar platforms, the direct booking path is already lost before the transaction even begins.

At the same time, the experience is evolving. We are moving toward a more frictionless path to purchase, but not by collapsing everything into a single conversational interface. As mentioned in the original post, making a reservation today still involves multiple layers, interfaces, and chains of responsibility. Although the transaction of a credit card purchase may always remain separate, technology is advancing. Instead of a website-first and booking engine-second approach, both hoteliers and guests are gravitating toward a booking engine-first experience.

That shift only works if the foundation is strong. Content, imagery, and data will continue to be the source of truth that powers discovery and enables that booking-first strategy to succeed.

Yes—but with important caveats.

Hotel booking is operationally complex, but that is not the real barrier. Technology evolves fast, and integration challenges are ultimately solvable. The key unknown is user adoption. Past attempts like Book on Google or Tripadvisor Instant Booking failed because users never perceived booking there as natural—and adoption will likely be uneven across segments and age groups.

For conversational booking to work, assistants must build trust and clarity. Users need to understand payment, cancellation policies, and—critically—who the merchant of record is. The channel matters: booking via a known brand is not interchangeable with an unknown intermediary. AI may enable the transaction, but it acts as an interface, not the channel itself.

Loyalty is another major constraint. If users cannot clearly access and trust their benefits, they will default to familiar channels. Loyalty bookings may be the last to migrate—or remain hybrid.

Finally, a clear commercial model is essential. Suppliers need transparent economics before participating at scale.

That said, my view has evolved. Assistants already dominate discovery and are moving into consideration via structured data like MCP, making transaction a natural next step.

So the question is not whether this will happen, but when—and who will be ready.

We already see bookings completed within the AI platform (Perplexity launched it nearly a year ago and Google is releasing their new travel product soon).

OpenAI is lagging. They've shifted priorities.

As of right now, their behavior looks like a chaotic reaction to outside stimuli rather than a structured planned business strategy. They may pivot again tomorrow.

OpenAI backing down from handling bookings directly in ChatGPT doesn’t mean conversational booking has failed. It just means hospitality is more complex than outsiders assume.

A hotel booking depends on many moving parts – from live inventory, dynamic pricing, payment, policies, cancellations, refunds and guest communication – all staying aligned and in real time. This is why earlier efforts, like Book on Google, struggled. The interface may have been simple, but the operational reality underneath was anything but.

Direct booking inside conversational platforms will only work when the infrastructure underneath is open and connected enough to support the entire transaction, not just the search experience. Open APIs are what make that possible. They give conversational interfaces access to the real-time data systems that power the booking, from availability and rates to payments and guest data.

Built on Apaleo’s API-first platform and MCP, our ChatGPT Booking App is already being piloted with a hotel partner and fully tailored to its brand. Guests can search, select and complete their booking end-to-end within ChatGPT, without ever leaving the conversation. This is real proof that conversational booking can work when it is built on open, API-first foundations.

I believe OpenAI did not have any idea how a) complex travel inspiration, planning and booking is, and b) how complex travel and hotel tech infrastructure is. 

OpenAI did not take into account:

• The complexity of hospitality stakeholder universe (owners, operators, major hotel brands)
• The state of hotel tech stack (600,000 hotels have on-premises tech stacks not even cloud ones), 
• Travelers’ loyalty affiliations (Marriott has 260 million loyalty members, Hilton 225M, Booking 200M plus)
• Travelers’ entrenched travel inspiration, planning and booking habits, even inertia. 

The Digital Customer Journey is like a meandering river with dozens of digital touch points (Google claims 48 of those). In my view, LLMs will become just one of these touch points, same as what Google currently is after 25 years of trying hard to become a one-stop-travel booking platform (Remember the failed “Book on Google?”).

Google saw the light before ChatGPT. In Google’s own words: “To make agentic AI a reality, Google is working on building out the experience with industry partners such as Booking.com, Expedia, Marriott International, IHG Hotels & Resorts, Choice Hotels International and Wyndham Hotels & Resorts.”

What will get Google in return? Affiliate commissions! 

A thoughtful question — and one that deserves a grounded answer.

The parallel with "Book on Google" is instructive. Google had the distribution, the data, and the engineering depth — and still shuttered the initiative in 2022. The lesson was available. Yet here we are again.

OpenAI has quietly retreated from its ambition to turn ChatGPT into a one-stop shop for travel bookings, finding that users researched trips inside the chatbot but rarely completed purchases there.

I agree with the structural argument. A reservation sits on top of a fragmented stack: CRS, PMS, channel manager, payment gateway, rate logic, cancellation rules, tax handling, and post-booking support. The transaction is where operational, financial, and legal accountability must align. The closer you get to it, the less it is a search problem and the more it becomes an orchestration problem.

Conversational platforms will absolutely influence discovery, comparison, and booking intent. They may even support conversion in selected use cases. But owning the booking end-to-end at scale is a different challenge entirely.

Will it happen? Yes — but only where systems are tightly integrated and servicing responsibilities are clearly defined. Across the wider market, fragmentation remains the barrier.

Owning the interface is not the same as owning the transaction.

“Book on ChatGPT” will follow the path of “Book on Google” (or “Book on TripAdvisor", for that matter) because it does not solve a customer need. It's self-serving for the platform (Google / OpenAI) but it's not something travelers are looking for. It's not only a matter of industry fragmentation: it's user behavior. Planning and booking a trip is not done in a single session. A customer will explore possibilities, think about them, then book their trip, typically over several days. Therefore, they do not have any urgent need to book now, just when they first heard about a particular hotel.

The fact that something becomes possible does not mean that people are interested in it.

OpenAI stepping back from in-chat bookings isn't surprising. The structural complexity of hotel transactions was always going to be a problem, but there's a second issue that applies to any industry: the chatbot is a genuinely poor interface for completing purchases, and the data from Walmart's experiment makes that hard to ignore.

Yes, but your technology debt comes first. While booking via a generative AI agent is the vision, the industry faces a significant "implementation tax," trying to integrate 2026 AI technology with 2006 IT infrastructure and technologies. Early adopters hit a wall because the necessary integrations, data organisation, and fundamental operating structures simply weren't in place to utilise AI effectively.

For an AI to facilitate transactions as complex as accommodation reservations, it has to stay "wholly connected.” Ultimately, AI tools must communicate seamlessly and in real time with the hotel’s tech stack without fail. If it does fail, we’re just creating another high-fee middleware layer that erodes margins, leads to disengagement, and results in purchase abandonment. We view AI-bookings as another channel driven by guests' endless flexibility expectations. 

Hotels must be present in every conversational space at the exact moment a guest commits. Owning the interface is meaningless if you don't own the transaction. Another topic warranting urgency is ensuring that AI-powered or agentic bookings are facilitated through your brand's distribution systems rather than an OTA. At the end of the day, building unified data to accurately understand every customer’s context, with human-driven service in mind, is what wins with AI.

I think you’re right, this feels very similar to “Book on Google.”

We’ve already seen that simplifying the interface doesn’t solve the underlying complexity. Hospitality bookings aren’t a single action. They’re a chain of dependencies — payments, policies, inventory, operations — all needing to align.

A conversational interface doesn’t remove that.

But maybe the more interesting shift isn’t where the booking happens.

It’s how people discover.

And I’m not convinced that completing a booking inside a chat is even that valuable.

Because the nature of shopping isn’t linear. It’s about comparing, exploring, being surprised.

There’s a reason we have window shoppers, not every interaction is meant to convert instantly.

The real question is:

👉 Will you be found?

👉 And will you be found with the right product in the right context?

Because there’s a big difference between:

“Standard Room” vs “High-floor, quiet corner room, away from the elevator — perfect for a 1-night business stay.”

One gets lost. The other gets chosen.

Conversational AI in Hospitality: Evolution, Not Revolution

The recent shift away from “Book on ChatGPT” toward third-party transactions should not be seen as a failure of conversational platforms, but rather as a clarification of their role within hospitality distribution.

The comparison with “Book on Google” is valid: simplifying the interface does not resolve the structural complexity of hotel bookings. Transactions in our industry remain deeply interconnected across systems, responsibilities, and legal frameworks. As highlighted in broader technology discussions, a modern interface often hides incomplete backend integration —and this is precisely where many visions fall short.

However, the real distinction lies elsewhere: between B2B and B2C.

B2B bookings are rational, rule-based, and constrained—price, availability, and policy dominate. These processes can be streamlined and, to some extent, automated.

B2C bookings, in contrast, are emotional decisions. Guests are not just purchasing a room; they are buying anticipation, trust, and experience. This cannot be reduced to a single conversational flow.

Therefore, conversational AI will not replace the booking process—it will enhance it. It will answer questions, reduce uncertainty, and guide decisions, ultimately strengthening direct relationships.

The future is not “booking through AI,” but “booking with AI support.” This is evolution, not disruption—and understanding this difference is critical for our industry.

Bookings can happen inside tools like ChatGPT—but not on their own.

To actually complete a booking, these platforms still need to rely on existing systems (like OTAs, hotel booking engines, and payment providers) that handle the hard parts: payments, availability, cancellations, and customer support.

That’s where the challenge is.

In travel, making a booking isn’t just one click—it’s a chain of connected steps across different systems. If a conversational platform tries to handle everything itself, it quickly runs into complexity and risk.

So what we’re likely to see is:

AI helps you search, compare, and decide faster

But when it’s time to pay and confirm, you’re passed to a platform that already manages bookings properly

For conversational booking to fully happen in one place, someone would need to take control of all those moving parts behind the scenes. Until then, these tools will guide the decision—but not fully own the transaction.

We are only in the early days of AI capability.

More recently we've seen introduction of AI based process tools that can be programmed to complete the necessary steps in a process. I see no reason why in the future this would not be possible for a hospitality or travel style bookings. In fact, I see incentive for providers to deliver that tool when it becomes possible. In reality, how different is it to an app with a booking engine and link? Just the next step in the logical transition of the process. I think it will come.

The remaining question is the price-value relationship with travel products in general. Not cheap. I do feel (certainly from my personal travel behaviour), that the traveller will want to 'see' the product. If this is not true, why does a market exist for reviews from unknown people? Therefore, I believe the first step in the 'seeing is believing' phase will remain.

This begs the question which party will be best to provide it? It's always been clear to me that the only somewhat trustworthy party is the provider of the end product. One day that penny will drop for the broader traveling community.

We will see bookings completed directly within both chatbots and agents, and the recent announcement from OpenAI really has nothing to do with travel. If you go back to the original announcement they made, it never mentioned travel once, not even indirectly. They were simply talking about stepping back from their previously stated direction of driving retail purchases within ChatGPT. So if you want to buy a new tent, a pair of running shoes, or get a great deal on an umbrella, you can have ChatGPT do all the research and make personalized recommendations. Based on those recommendations, you can click through to the retailer's website (either directly or via an intermediary) to complete the purchase.

There's also been speculation that chatbots/agents won't ever be able to book travel because it's too complex. That's also highly unlikely. More like a bedtime story than reflecting a serious grip on reality. Present-day models are capable of rendering medical diagnoses, solving complex mathematical problems, assisting in drug discovery and drug trials, and other truly hard and complex things. Travel just hasn't hit the top of the list yet. But it's coming!

I wrote an article about this recently. See here for more details.

The deeper truth about our industry is that control of demand is not the same as controlling fulfillment. Hospitality is not structurally broken, it is structurally distributed. The friction isn’t technical, it’s institutional. Ownership of guest experience spans brands, operators, payment providers, regulators, and legacy infrastructure. No single interface layer—whether Google yesterday or ChatGPT now—can compress that into a seamless, end-to-end transaction without assuming risk the ecosystem isn’t ready to offload.

That said, this this is not a ceiling, it’s a transition point.

Conversational platforms will not replace OTAs or core systems, they will re-orchestrate them. The advantage will go to those who expose their capabilities (pricing, inventory, servicing) as composable, real-time services that can plug into any interface, conversational or otherwise.

The strategic question isn’t whether bookings happen inside AI, but who owns the service layer that AI replies on. If hospitality providers remain fragmented, intermediaries will continue to capture value. If service layer becomes unified, hospitality shifts from simply being bookable to truly programmable—and that’s where lasting control emerges.