AI has officially entered the booking funnel
For years, hotels have been adjusting to search engine optimization (SEO), online travel agencies (OTAs), and metasearch. But today’s reality is different: AI assistants like ChatGPT, Claude, Gemini, Copilot, and Perplexity are becoming the new travel planners.
Travelers are asking conversational questions (“Find me a boutique hotel in Barcelona under $200”) and increasingly searching for hotels and getting answers without clicking websites.
But let’s be crystal clear: search and booking are not the same thing.
- Search is discovery. It’s the process of an AI finding, interpreting, and recommending hotels. That’s what this course focuses on.
- Booking is a transaction. It’s when the AI actually books the hotel, handling payments, confirmations, and all the messy details.
Booking capabilities by AI platform today
Here are the three main booking models AI platforms are experimenting with today:
1. API Model (already adopted by Perplexity, Google Gemini, also some OTAs’ direct integrations via GPTs)
- How it works: The AI connects directly to the source via API feeds.
- What’s exchanged: Structured ARI data.
- Strengths: Fast & reliable: No scraping, no guesswork; Always accurate: Updates instantly with ARI (Availability, Rates, Inventory); Already widely in use.
- Weaknesses: Requires tech integration (channel manager, CRS, or booking engine provider); Hotels need to give API access, which means that smaller independents may not get exposure.
- Example: When you ask Google Gemini “Find me a boutique hotel in Lisbon this weekend”, it taps Google Travel APIs, which are powered by OTAs/hotel APIs, not scraping web pages.
- How to leverage: Ensure your hotel is present in Google’s hotel ecosystem: this means maintaining an updated Google Business Profile and supplying your rates/inventory to Google Hotels (via a free Google Hotel Center feed or through a connectivity partner/channel manager).
2. Web Model (currently adopted by ChatGPT Agent Mode)
- How it works: The AI acts like a human browsing the web – it opens Booking.com, Expedia, or your direct site, reads the content, clicks through, and can even fill forms.
- What’s exchanged: HTML content + simulated user actions.
- Strengths: Doesn’t need special integration: Any hotel website that’s usable by a human can be used by an agent; Can combine multiple sources (official site + OTA + reviews) on the fly.
- Weaknesses: Slower than APIs; Risk of errors (AI “clicks the wrong thing” or misreads); Payment & confirmation workflows are trickier (AI must securely handle credit cards).
- Example: In Agent Mode, ChatGPT could open your hotel’s booking engine, pick dates, compare to Booking.com, and decide where to complete the reservation.
- How to leverage: To be prepared, ensure your direct booking process is AI-friendly: e.g. no hard CAPTCHA on your booking page, mobile-friendly design, and simple checkout flows. While many travelers may be cautious about entrusting an AI with their login or payment info, this capability is growing. If your hotel’s own website is straightforward to book, an AI assistant might successfully book guests directly with you (especially if the user instructs it to use your site). Likewise, a seamless presence on OTAs pays off here – if the AI decides to book via, say, Expedia or Booking.com (perhaps using stored loyalty or payment details), having availability and competitive rates on those platforms means the AI can complete the reservation for the guest.
3. MCP Model (Model Context Protocol – the new shiny and most obscure object, not yet in use but full of promise)
- How it works: MCP (being pushed by OpenAI, Anthropic, etc.) would allow AI agents to directly talk to a hotel’s system in a standardized way – like a universal API connector but designed for LLMs.
- What’s exchanged: Structured, LLM-ready context (availability, pricing, policies) — richer than raw APIs, easier for multiple models to plug in.
- Strengths: Could make hotel direct bookings seamless without OTAs in the middle. Standardized: less custom dev per platform; Supports contextual queries (“What’s your quietest ocean-view room under $300 with free breakfast?”).
- Weaknesses: Not implemented yet, nobody knows how hard it will hit us; Requires hotel tech vendors (CRSs, booking engines and others) to adopt MCP and connect to AI platforms; Adoption will likely be slow, just like with past hospitality standards.
- Example (future): You ask Anthropic Claude: “Book me a pet-friendly suite in your hotel this Saturday under $400.” Claude queries your MCP-enabled booking engine → gets structured results → books directly.
- How to leverage: We don’t really know yet. Just sit and wait. Unless you’re a tech vendor or hotel chain working with developers on backend systems, achieving direct MCP-supported bookability is not yet feasible.
What exactly is MCP?
Think of MCP as the USB-C for AI agents.
- It was introduced by Anthropic in late 2024, and quickly embraced by OpenAI, Microsoft, and Google in early 2025.
- The idea is that instead of building a separate “plugin” for each AI, a company (a tech provider building a hotel booking platform for example) can build one MCP server.
- Any AI platform that supports MCP can then “mount” that server, instantly gaining access to the tools and data it provides.
An MCP server is basically a connector: it exposes functions (tools) like getAvailability, getRates, createBooking, and data resources like hotel descriptions, amenities, and photos.
The key idea: one integration, many AIs.
What this could look like for a guest
Imagine the near future. A traveler opens ChatGPT and types: “Book me a king room in Chicago near Millennium Park, Sept 18–20, under $250 a night.”
Here’s what happens:
- ChatGPT mounts a Hotel Booking MCP connector (provided by an OTA, CRS, or GDS).
- It queries availability and rates across N hotels.
- ChatGPT summarizes 3–5 options, complete with price, photos, cancellation rules, and review highlights.
- The guest clicks “Book now” (within the ChatGPT interface).
- A secure authentication step (OAuth or tokenized payment) kicks in.
- Within the same chat window, the booking is confirmed and a confirmation number is displayed.
No redirects. No website search. Booking completed inside the AI assistant. BOOM. But… This won’t happen for a while. Not with ChatGPT at least. OpenAI will likely implement it within custom GPTs, not the main chat window, for various reasons. To understand these reasons – let’s review the timeline of OpenAIs behaviour related to travel search.
From plugins to GPTs to MCP
To understand why this matters, it helps to look at the evolution:
- 2023–2024: Plugins – Expedia, Kayak, Instacart, and Klarna built ChatGPT plugins. They worked, but only inside ChatGPT and each required custom code.
- 2024: GPTs & Actions – OpenAI phased out the plugin store, introducing custom GPTs with “Actions” that could call APIs. Better, but still siloed per platform.
- 2025: MCP – A universal, open connector standard adopted by multiple AI companies. The idea is that a vendor builds once, and that connector can be used by ChatGPT, Claude, Gemini, Copilot, and beyond.
For hotels, MCP might potentially represent the shift from experimental add-ons to mainstream, scalable distribution channels.
Public vs. private MCP connectors
Here’s a key nuance: how MCP connectors surface to end users.
- Private (Enterprise Mount): In enterprise ChatGPT/Teams/Copilot, admins can mount a hotel MCP connector directly. Staff could then ask, “Check tonight’s availability in our Paris property” and ChatGPT calls the MCP directly.
- Public (Consumer Mount): In the future, consumers in the main ChatGPT window could potentially use hotel connectors, either through branded custom GPTs (e.g. “Expedia GPT” or “Bonvoy GPT”) or natively in ChatGPT’s default assistant.
Do I need to explain the huge difference between the two? If the answer is yes for some of you – ask yourself about the difference in public adoption rates in each of these scenarios. And then ask yourself how OpenAI will be making a decision of selecting THE preferred partner for default ARI and bookings within the platform.
So what will likely happen:
- Short term: Expect branded and third-party GPTs wrapping MCP connectors for discoverability and trust.
- Long term: OpenAI (and others) might eventually allow public connectors (maybe even more than one) natively in the main chat. That means a user wouldn’t need to “switch GPTs” – booking would just work inside the main assistant.
What it takes to be “Mounted” (not what you thought)
It’s important to note: MCP doesn’t magically make your connector visible to every AI. Each AI platform (OpenAI, Google, Microsoft, Anthropic) still has to:
- Approve your connector (security review, compliance, business terms).
- Mount it (register the endpoint in their runtime so the model can call it).
So if you’re a vendor with 100,000 hotels in your MCP, you don’t need to build five separate APIs — but you do need each AI company to approve and mount you. That leads us to the next point.
Who will become preferred hotel booking partners?
Logically (as we agreed above) OpenAI and peers will be very selective about which connectors they mount for public consumer use. Their decision criteria will include:
- Coverage: Can you satisfy most queries globally? (Advantage: OTAs and GDSs.)
- Data quality: Do you offer live ARI, rich content, low error rates?
- Trust & reliability: Are you PCI-compliant? Do travelers trust your brand?
- Business alignment: Are you willing to share revenue or pay CPC fees?
- Liability & support: Can you handle customer disputes, refunds, and compliance globally?
That means the early winners will likely be OTAs (Expedia, Booking.com) but maybe even GDSs.
It could be a centralized platform that books direct… but wait… we don’t have one. Oh well.
So what will likely happen is (drum rolls)… fragmentation. OTAs will have their own MCP, hotel companies will have theirs, large CRS/channel managers could follow. And Independent hotels will (again) be screwed.
Unless we build a centralized platform that books direct… but that’s too good to be true.
The business models emerging
MCP obviously defines the connection, not the commercial terms. But based on precedent here are potential business models that could be adapted by AI platforms:
- Commission / Rev-share (dominant today): Each booking completed via ChatGPT to OTA or vendor shares % with OpenAI.
- CPC (cost-per-click / per-call): Hotels or OTAs pay per invocation, similar to Google Hotel Ads.
- Hybrid: Free organic presence + paid boosted placement.
- Enterprise SaaS: Vendors pay OpenAI a fee for default mounting inside enterprise versions of ChatGPT.
For hotels, this likely means OTA-style commissions will extend into AI channels… Unless we build a centralized platform that books direct… but no… too good to be true.
What hotels should do now
- Ensure your hotel is present in Google’s hotel ecosystem.
- Make sure you’re listed on all major OTAs (don’t love the commission part but we have to prepare for the worst).
- Ensure your direct booking process is AI-friendly so that AI agents can book on your website.
- Take Ira’s new online course about how to adapt your web content for AI search for better discoverability (course launching Sept 2025, more info at www.iravouk.com).
- Monitor industry trends on the evolution of MCP. We don’t know where it’s going but it might be big.
Conclusion: a distribution revolution in the making
The Model Context Protocol could potentially become the universal booking connector for AI. Instead of one-off plugins, hotels could be bookable inside AI platforms at scale, with a single connector powering ChatGPT, Gemini, Claude, and Copilot.
The stakes are high:
- If OTAs dominate this layer, hotels risk another round of intermediation.
- If hotel tech vendors step up (centralized platform for direct bookings people, please!!!), both branded and independent properties could retain more direct control in the AI era.
The opportunity is simple: be ready, or be invisible. Because when a guest says to ChatGPT, “Find me a cozy boutique in San Diego with a killer spa and a bar that knows how to pour a proper margarita,” you want your property topping that list.
Ira Vouk
Ira Vouk Hospitality 2.0 Consulting
View source
Please visit:
Our Sponsor