AI Moves The Front Door

More than half of US leisure travelers have used AI somewhere in their travel journey. And earlier this month, the industry got its 1st real test of a fully agentic booking. This is the state of travel search, discovery, and distribution in 2026. Before we get into the episode, a quick note on the format of today's show. This is the 1st episode of a 5-part foundational series. These episodes are longer and cover a lot more information than the usual AITB briefings. The goal is to build a shared map of where AI and travel actually stand right now, so that every episode after this has somewhere to land. Think of it as the context layer. Once it's in place, we move faster. Regular episodes will begin as soon as the series wraps. Now, let's start with the number that reframes everything else. Phocuswright published a report in March called the AI Surge: Travel's Fastest Behavioral Shift in a Decade. They surveyed 1,570 US leisure travelers in February of this year. The finding, 56% of travelers used AI for planning, booking, or in-destination assistance for at least 1 trip in the past 12 months. That's up from 43% in late 2025 and more than double where the industry was in 2024. Pete Comeau, Managing Director of Phocuswright, put it this way: AI has crossed the threshold from curiosity to utility. Travelers are past the point of experimenting now. They're integrating AI into the core of how they research and shape their trips. That's the important word, integrating. AI is not replacing the whole journey, at least not yet, but it's entering the journey. And once it enters, it starts influencing what gets considered, what gets compared, and eventually, of course, what gets booked. Now, here is the detail that tells you where the shift is actually landing. Phocuswright found that generative AI platforms, ChatGPT, Gemini, Claude, and others, have reached 33% usage for trip research. Traditional search engines sit at 35%. They're nearly level. The front door of travel discovery is no longer Google by default. It's not the OTAs either. And these are not low-value users. The same Phocuswright study found that travelers who use AI for trip planning have a median household income of $129,200 versus $104,000 for non-users. They take 3.8 leisure trips per year instead of 2.9. They spend $4,500 annually on travel, not $3,000 It turns out that the AI-assisted traveler is the industry's highest value customer. That is not a niche. That is the customer everybody's fighting for. So who positioned themselves inside that new front door 1st? Let's go back to October 6, 2025, OpenAI's DevDay, a moment that for me personally indicated we are on the verge of a major shift in how travel gets booked. 1 week after announcing the Agentic Commerce Protocol, a tool that simply put enables AI to process payments for anything, including bookings, OpenAI launched its Apps SDK. SDK is a software development kit, basically the toolkit that lets 3rd parties build applications that run inside ChatGPT conversations. A few weeks after that DevDay, Sam Altman, OpenAI CEO, revealed that in the future, users may be able to book hotels in 1 click, I quote, directly through ChatGPT. Here's what actually happened. The 2 most important travel launch partners for ChatGPT Apps were Expedia and Booking.com. Overnight, those 2 OTAs got a direct presence inside ChatGPT's massive user base. Clayton Nelson, VP of Strategic Partnerships for Expedia Group, told PhocusWire at the time that traffic from generative AI search was still small but growing quickly and converting to bookings at higher rates than any other traffic. And that last part, that is the signal. Higher conversion. That is not a traffic story. It's a quality story. Now, what didn't happen on October 6 matters just as much as what did happen. No airline got the same direct ChatGPT position. No independent hotel chain was the default path. No vacation rental manager had a direct route. They were mostly accessible through intermediaries. The new front door opened, but Expedia and Booking.com were already standing in the doorway. That changed, at least partially, on March 4, 2026. Lighthouse, the AI commercial platform for hospitality, launched the Hotels Network app inside ChatGPT, the 1st direct booking app for hotels on the platform Now, to be precise, March 4 is the date when OpenAI approved the app. It was submitted by the Lighthouse team a while back. The listing of the app on the ChatGPT marketplace coincided with ITB Berlin, and since we were all there, we did a short interview with the Lighthouse team, which we will air as a bonus episode on the AI Travel Brief. So what is this app? It is built around MCP, Model Context Protocol. MCP is a way for AI systems to connect to external tools and data sources, so they can pull live information instead of relying only on static web content. Something like API, but for AI. In this case, it creates what Lighthouse calls a live data bridge between hotels and AI platforms, verified property content, real-time rates, direct booking links, and no commission model. Of course, you have to be a Lighthouse client to use that. Juanjo Rodriguez, head of direct booking at Lighthouse, told PhocusWire that this matters because OTAs already made the leap to AI discovery, and he was explicit about the business model. It's a core infrastructure play. We don't want to become a cheaper OTA. That's an important distinction. Lighthouse is not trying to own the guest. It is trying to give hotels a way to appear inside AI discovery without sending that demand back through an OTA. Then a few weeks later, DirectBooker pushed in the same direction, and you're very likely to hear from DirectBooker CEO and co-founder Sanjay Vakil as well on this show. DirectBooker launched a ChatGPT app and a Claude connector with supply agreements from major hotel groups, including BWH Hotels and Radisson, and integrations with hotel connectivity providers like Cendyn, eviivo, Mirai, SiteMinder, and Roiback. Their model is slightly different, but the strategic point is the same. For vacation rentals, I'm sorry to say, but the distribution layer is still missing. That's where Onseason, the AI direct distribution engine I'm building for the vacation rentals industry, is heading. Hotels now have early direct booking infrastructure for AI discovery. Vacation rental managers, well, they don't. And that's a problem because if AI assistants can see Airbnb, Booking.com, and Expedia, but cannot reliably see professional property managers directly, then the AI era simply recreates the same dependency problem with a new interface. Different screen, same gatekeeper, so more of the same. Now, here's where the story gets complicated in exactly the way that's worth sitting with. On March 5, 2026, Skift reported with OpenAI spokesperson confirmation that the company was pulling back from enabling direct in-chat transactions inside ChatGPT. The company's message was clear. ChatGPT would prioritize search and product discovery, while commerce and checkout would move through 3rd-party apps and protocols. Expedia and Booking Holdings shares jumped on the news. The market had priced in the risk of AI disintermediating the OTAs, and well, that risk for now was off the table. But I think the market read was incomplete. The booking threat may have eased, but the discovery threat did not Because the most valuable part of the journey is not always the payment, it's the shortlist. The shortlist is where preference gets shaped. It is where a traveler decides which destinations, which brands, which hotels, which neighborhoods, which property types, and which platforms are even worth considering. If AI owns the shortlist, AI owns the beginning of the commercial journey. Even if Expedia, Booking, Google, Uber, TikTok, a hotel website or a direct booking engine still closes the sale. Coney Dongre, a research manager at Phocuswright, made a related point on LinkedIn. Travel commerce is hard. The closer AI gets to the messy core of travel, real-time pricing, fare rules, cancellation policies, post-booking servicing, chargebacks, refunds, disruption management, you name it, the more valuable experienced intermediaries become. Travel is not buying a pair of shoes inside a chatbot. The transaction is never just the transaction. The hardest part of agentic travel is not finding a flight. It's what happens after the flight changes, the guest cancels, the card fails, the rate changes, the hotel overbooks, the traveler wants a refund, or the supplier just simply says no. That is why incumbents still matter. They don't just own supply, they own accountability. And that framing maps directly onto the most important data point in travel AI this year. Skift Research and McKinsey, in their joint report, Remapping Travel with Agentic AI, found that more than 90% of consumers trust AI-generated travel information. However, only 2% currently would allow AI to book on their behalf. And while 80% of travel executives plan to deploy agentic AI at scale within the next 3 to 5 years, the consumer has not moved to match them. Gareth Williams, founder of Skyscanner, told Skift, "I have been really struck by how negative the public is towards AI compared to people inside the industry." This is what I call the 80-2 gap. 80% of executives building for a consumer that 2% of the market has become. Most of them don't really have a choice. If you're not building for AI, you will be penalized by the markets. Every story in this series sits inside that gap. Expedia's own AI Trust Gap report says almost the same thing from another angle. Travelers are comfortable letting AI suggest options. They're not yet comfortable letting AI press play Expedia surveyed more than 5,700 adults across the US, UK, and India. The report found that 53% are comfortable letting AI suggest travel options, but only 8% feel comfortable booking through an AI platform. And 68% still prefer to book via a trusted travel brand even when AI booking is available. So the picture is becoming quite clear. AI has permission to influence the choice. It does not yet have permission to own the transaction. Discovery moved 1st. Booking is lagging. Infrastructure is still catching up. And here is the nuance that keeps this honest. Mike Coletta, Senior Manager of Research and Innovation at Phocuswright, made an important point about the March 2026 data. Half of travelers who used AI in search engines still click through to source websites after seeing AI answers. That violates the simple zero-click world narrative. AI is reducing click-through in search overall, but travel is more resilient because it is higher stakes and verification-heavy. So AI is the 1st filter. Traditional channels are still the confirmation layer. The front door moved, but the checkout counter did not. Now, we also have to talk about Google because this episode could easily sound like ChatGPT versus OTAs, but the bigger distribution story may be Google turning search into an AI travel workspace before ChatGPT becomes a true travel commerce platform. ChatGPT created the headline moment. Google still owns the habit. Google's AI Mode is already moving from answers into planning. With Canvas, travelers can describe a trip and generate an itinerary that pulls together flights, hotels, maps data, reviews, photos, and web information, then refine it through follow-up questions. Google has also said it is working on agentic booking for flights and hotels through partners including Expedia, Booking.com, Marriott, and Wyndham. That matters, and that matters because Google does not need to invent travel discovery. It already sits on top of it. So the question is not just who wins inside ChatGPT or inside Claude or Perplexity. The question becomes, does search become conversational before ChatGPT and all these other players become transactional? And then there is TikTok out of the blue. On May 12, TikTok introduced TikTok GO in the US, letting users discover and book hotels, attractions, and tours directly from the app. Launch partners include Booking.com, Expedia, Trip.com, Viator, GetYourGuide, and Tiqets. Now, this is not exactly the same as AI search, but it is part of the same distribution shift and narrative. TikTok is turning inspiration into booking. The traveler sees a video, taps into the place, checks availability, compares options, and books without beginning the journey on Google or an OTA homepage or on ChatGPT. So the new travel funnel is not linear anymore. It is not search, compare, and then book. It is see something, ask something, save something, tap something, generate something, compare something, then book wherever trust and convenience meet. And then there is Uber. On April 29, Uber announced hotel bookings inside its app powered by Expedia. US users can now book hotels from Expedia's supply, with more than 700,000 hotels available and Vrbo vacation rentals expected to be added later. Uber also introduced Travel Mode, a more concierge-like experience that guides travelers through airports, local recommendations, restaurant reservations powered by OpenTable, and trip-related services. It also announced AI-powered voice booking. Again, this is not just a simple hotel tab inside a ride-hailing app. This is Uber saying, "We already know where you're going when you land, where you sleep, where you eat, how you move, and how you pay." Uber is making a play to be a super app. And to be a super app, you need travel distribution. And that is travel distribution, maybe not in the old OTA sense, but absolutely in the new interface sense. And I keep saying this, we still don't know what kind of interfaces we will use to interact with the digital environment in the future. And that is the broader pattern here. So the future of distribution is not 1 new front door. It is every high-intent surface becoming bookable, every interface. AI chats, Google Search, Maps, TikTok videos, Uber, messaging apps, fintech apps, creator content, loyalty apps, corporate booking tools. The front door didn't move to 1 place. It fragmented into every interface where intent appears. AIVO Research published a travel AI visibility study in April 2026 testing 1,500 queries across 5 AI engines and 18 travel and hospitality brands. Their finding? Each platform has developed a distinct recommendation philosophy. Every door is different. Perplexity behaves differently from OpenAI. Google AI Mode behaves differently from both. Different AI surfaces are building different versions of the travel consideration set, and this is not neutral. That is a visibility question every travel brand needs to be asking right now. When a traveler asks, "Where should I stay in Lisbon with a toddler?" Or, "What are the best boutique hotels in Rome near restaurants?" Or, "Beach villas in Croatia cheaper than Airbnb?" Who gets mentioned? Is it your brand, your direct booking website, your OTA listing, your competitor, Airbnb, Booking.com, Vrbo, a blog, a Reddit thread, a media article from 2019? This is why AEO and GEO matter. Even if the language is still messy and the playbook is still immature and half of the people don't know what they're talking about, AEO means optimizing for answer engines. GEO means optimizing for generative engines. Different names, but basically the same goal, making your business understandable, quotable, and trustworthy to AI systems. But this is not SEO with AI keywords. AI visibility is about structured data, clean inventory, consistent policies, reviews, entity clarity, live rates, availability, direct links, and trustworthy sources. That's what makes travel so complicated. The AI has to know what you offer, when it is available, how much it costs, who is it right for, what the rules are, and whether the information can be trusted. For hotels, that work is already starting through Lighthouse, DirectBooker, Google, OTAs brand apps, and AI visibility efforts. For vacation rentals, the problem is, hmm, sharper. Hotels have brands, chains, loyalty programs, CRS systems, and decades of distribution muscle. Most professional property managers don't have that. Inside an AI answer, Airbnb is visible, Booking.com is visible, Vrbo is visible, but the long tail of professional operators, that is still not visible, at least not for the majority. For vacation rentals, machine-readable inventory is not a technical detail. It's the difference between being part of the AI consideration set and disappearing behind the platforms again. I can talk all day about the lack of GDS in the short-term rental space, but that's a topic for another episode. Anyhow, that brings me to a part of the story I think has been underreported in the past months, and I'm calling 2026 the pipeline year. Not because consumer adoption arrived on the schedule the industry projected or hoped for. It didn't. But because this is the year the actual infrastructure, real inventory, real payments, real conversations started getting wired together. And here is the trigger event. On February 12, 2026, Sabre, PayPal, and Mindtrip announced what OAG described in its March radar as the travel industry's 1st end-to-end agentic AI booking pipeline. The announced launch was Q2, and it went live on May 6. Now, I haven't tested it myself, but at launch, reports say that it is focused on flights. A traveler describes their trip in natural language inside Mindtrip's conversational interface. Mindtrip queries Sabre's live travel inventory. The AI surfaces curated personalized flight options, and PayPal's commerce infrastructure handles payment inside the conversation. Sounds pretty straightforward, but it's fairly complex. Important to note, no redirect to a separate website like ChatGPT Apps. No handoff to a traditional booking engine, again, like ChatGPT Apps or Claude connectors. The search-select-pay loop happens in a single chat, and that is the most important part. Travel has had plenty of AI chatbots that can talk about travel, plan travel, et cetera, and travel has had plenty of booking engines that can process transactions. What's new is connecting real-time inventory, conversational AI, and payment into 1 continuous flow. That is the agentic North Star the industry keeps talking about. But it is also worth saying that this is not magic; it's infrastructure and a lot of hard work. Underneath all these announcements are competing pipes, MCP, GDS rails, NDC, and commerce protocols for payment and checkout. Different acronyms, same kind of question. Can an AI system access accurate inventory, price it, book it, pay for it, and service it? That last part, service it, is the part people keep underestimating. While Sabre was building that pipeline, the industry was simultaneously arguing about what the standards underneath any of this should look like. The argument happened in public at the Airline Distribution 2026 conference in Barcelona in March. The question was, does NDC still matter when AI agents can reason across complexity directly? Guido Becher, Global Head of Travel and Loyalty at Rappi, put the provocation plainly. "I've seen integrations that took maybe 2 or 3 months being done in 1 day. Maybe on search and distribution, many of the standards are not even that important anymore." Riccardo Vittoria, Co-founder and CEO of Acai, pushed back with the constraint the agentic AI conversation keeps stepping over. In servicing, AI can be a disruptor. In bookings, we need to see because agentic AI has a lot of cognitive power but is still expensive. Cost, not trust, not technology. Simple, pure economics. That's the word nobody in the agentic AI pitch is pricing in clearly. Now, my take is that the new hill isn't NDC versus EDIFACT. It is whether your inventory is machine-readable at all in any format. The standard matters less than the data underneath it. Now, 1 more thread worth pulling before we close, and it challenges the conventional wisdom I've heard in a lot of industry conversations this year. The assumption is that corporate travel will be the last segment disrupted by agentic AI. Too much policy complexity, too much compliance, too many approval layers, too many exceptions. Well, my personal take is that that may be wrong. Business travel may actually be 1 of the 1st places agentic booking works. Why? The reason is structural. The trust problem that keeps leisure consumers from letting AI book on their behalf is different in managed travel. In corporate travel, the policy is the authorization layer. The online booking tool enforces the guardrails. The travel management company handles the exceptions. The company defines what is allowed, what is preferred, what is out of policy, and when approval is needed. The AI doesn't need to win the traveler's full personal trust from scratch. It simply needs to operate inside a predefined company policy. It can be guardrailed, and that's 1 of the biggest problems with AI these days, how to harness and guardrail AI. That is much, much more of a solvable problem. And then on May 4, Amex GBT agreed to be acquired by Long Lake Management for $6.3 billion, a 60% premium to its pre-announcement share price. Now, Long Lake is a firm focused on acquiring service businesses and rebuilding them around applied AI. The press release did not lead with cost savings or headcount reduction. It led with applied AI, traveler experience, and making business travel faster, smarter, and more seamless. That is a $6.3 billion bet that the TMC layer is not simply being disrupted. It's being rebuilt as the agentic layer The place where AI sits between the traveler and the supplier operating at corporate scale with policy as the guardrail and human agents handling the edge cases. That is a very different story than the corporate travel is the last to move. Corporate travel may be the 1st segment where agentic booking actually works because the conditions that make autonomous AI viable in consumer markets are quite hard to build. In managed travel, many of them already exist. We'll go deeper in corporate travel in a later episode, but the Amex GBT deal alone is worth knowing about right now. Now, the full story of what all of this means for OTAs, for data moats, for who controls the customer relationship in the agentic era, that is its own episode, because OTAs are not a subplot here. They're actually the main event in travel AI distribution, and they deserve the full treatment. That's the next foundation episode, which will air right after Airbnb's summer release on May 20. So here's where we are. More than half of active US leisure travelers have now used AI somewhere in the journey. Discovery moved faster than booking. ChatGPT opened the door, OTAs stepped in 1st, and hotels are now building direct paths through Lighthouse and DirectBooker. Vacation rentals still need theirs. The 1st real agentic flight booking pipeline went live this month with Mindtrip, Sabre, and PayPal. The standards debate is giving way to a simpler question: Is your data machine readable at all? And corporate travel, long assumed to be the last segment to move, just had a $6.3 billion bet placed on that it may move 1st. That's the state of travel distribution in 2026. Discovery has moved. Booking is lagging. The pipes are being built, and the battle is not just for the transaction, it's also for the shortlist. That was episode 1 of the foundational series of the AI Travel Brief. Thank you for listening or watching. Cheers.