--- title: "How Expedia, Booking.com, and Google Flights Failed the AI Test" description: Legacy OTAs added AI chatbots to the same search pages. The core paradigm is unchanged — and that is exactly the opportunity. canonical: https://nowah.xyz/blog/expedia-booking-google-ai-test lastModified: "2026-08-07T07:55:43.175Z" --- # How Expedia, Booking.com, and Google Flights Failed the AI Test Legacy OTAs added AI chatbots to the same search pages. The core paradigm is unchanged — and that is exactly the opportunity. Every major travel company announced an AI strategy in 2023. By mid-2024, most of them had shipped something. Expedia launched an AI chatbot. Booking.com built an AI trip planner. Google wove AI summaries into flight and hotel results. Kayak added a conversational layer. Hopper kept refining its price prediction algorithms. Three years later, we can evaluate the results. And the results are not good. Not because the AI itself is bad. These are well-resourced companies with access to the same large language models and the same engineering talent. The AI works fine. The problem is where they put it and what they let it do. Every single incumbent took their existing product, a product built around search forms and results lists and filter sidebars, and attached an AI chatbot to the side of it. The chatbot can answer questions. It can suggest destinations. It can summarize reviews. But when you actually want to book something, you get redirected right back to the same 2003-era interface you were using before. That is not an AI travel product. That is a travel product with an AI sticker on it. ## Expedia: the chatbot that hands you off ![Illustration for this section](https://pics.nowah.xyz/website-media/product-032-img-1.webp) Expedia was one of the first movers. They partnered with OpenAI and launched a ChatGPT-powered feature inside the Expedia app in April 2023. The pitch was compelling: ask the AI about destinations, get personalized suggestions, plan your trip through conversation. The reality is more modest. Expedia's AI chatbot sits in a separate tab within the app. You can ask it things like "where should I go for a beach vacation in March?" and it will give you a reasonable answer. It can suggest destinations. It can talk about different neighborhoods in a city. It can help with general trip planning questions. Here is what it cannot do: book you a flight. Search live inventory in real time. Compare hotel prices. Execute a payment. Handle the actual transaction that is the entire reason you opened a travel app. When you're ready to go from planning to booking, the chatbot hands you off to Expedia's standard [search interface](/blog/no-search-bar-booking-interface). Same form. Same results page. Same 347 flights sorted by price with a sidebar full of filters you'll barely touch. The AI conversation you just had? It doesn't carry over. You're starting from scratch in a completely different interface. Expedia has over 3 million properties and massive flight inventory. The AI has no meaningful access to any of it during the conversation. It's a concierge desk that can tell you about the hotel but can't give you a room key. The gap between the AI marketing and the AI reality is wide. Expedia spent millions telling consumers they could "plan trips with AI." What they built was a travel FAQ chatbot that sits next to the same product they've had for twenty years. ## Booking.com: the planner that can't execute Booking.com took a different angle. Instead of a general chatbot, they built an "AI Trip Planner" that generates day-by-day itineraries for your destination. Tell it you're going to Rome for five days and it will produce a detailed itinerary with morning activities, afternoon sightseeing, and restaurant suggestions. This is genuinely useful content. The itineraries are reasonable. The suggestions show some understanding of geography and pacing. For someone who has already booked their flights and hotel and just needs help filling the days, it's a decent tool. But the AI Trip Planner doesn't book anything. It generates an itinerary, and then you have to manually search for and book each component yourself through Booking.com's standard interface. Want to book that hotel the AI suggested? Go to the search bar. Type it in. Scroll through results. Compare prices. Select your room. Enter your details. Complete payment. Then go back to the itinerary and do the same thing for the restaurant reservation, and again for the museum tickets. The disconnect is almost comedic. You have an AI that is smart enough to plan an entire trip but not empowered enough to book a single hotel room. Booking.com's core product is the hotel listing page. It is one of the most optimized conversion machines in e-commerce history. Every element on that page, the urgency signals ("Only 2 rooms left at this price!"), the social proof ("147 people booked this in the last 24 hours"), the comparison tools, has been A/B tested thousands of times. The AI Trip Planner doesn't threaten any of this because it sits in a completely separate section of the product. The booking revenue still flows through the traditional listing pages. This is the classic [innovator's dilemma](/blog/innovators-dilemma-travel-expedia) in action. Booking.com's most profitable feature is the thing that AI should replace. So they built the AI to complement it instead. ## Google Flights: smarter search, same paradigm Google's approach to AI in travel is the most sophisticated and the most constrained. Google Flights now includes AI-generated summaries that appear above search results. Search for flights to Tokyo and you might see a paragraph explaining that prices are currently lower than average, that cherry blossom season starts in late March, and that direct flights from your origin are available on three airlines. Google Hotels shows similar AI enhancements, with neighborhood descriptions and review summaries. These additions are genuinely helpful. Google has extraordinary data about pricing trends, and their AI can synthesize that into useful advice. The price tracking and prediction features are legitimately good. But Google Flights is not a booking platform. It is a referral engine. When you find a flight you want, Google sends you to the airline's website or to an OTA to complete the booking. Google makes money from those referral clicks and from the ads displayed alongside search results. An AI agent that just books the flight for you would eliminate both revenue streams. So Google's AI makes the search better without changing [what happens after](/blog/what-happens-after-you-book) the search. You still get a list of results. You still have to compare them yourself. You still click through to a different website to buy. The AI helps you understand the results, but it doesn't help you avoid the results page entirely. There is also the personalization gap. Google has enormous amounts of data about you, but Google Flights' personalization is limited to your search history and location. It doesn't know that you prefer aisle seats, or that you have a connecting flight phobia, or that you always travel with a carry-on only. That kind of individual preference data requires a different product relationship than a search engine provides. ## The common pattern: AI as a layer Step back from the individual implementations and a pattern is clear. Every incumbent followed the same playbook: 1. Take the existing product (search form, results page, booking funnel) 2. Add an AI layer on top (chatbot, planner, summary) 3. Keep the existing product as the primary booking path 4. Use the AI for discovery, inspiration, or Q&A only 5. Route users back to the traditional interface for transactions This pattern makes perfect business sense for the incumbents. Their existing products generate billions in revenue. Those products have been optimized over decades. Replacing them with something unproven is an enormous risk. Adding AI as a supplementary feature lets them claim innovation while protecting core revenue. But this pattern also means the fundamental booking experience hasn't changed. The user still fills out a search form. Still scrolls through hundreds of results. Still compares across tabs. Still loses context when switching between flights and hotels. Still abandons 85% of the time before completing a purchase. The AI layer hasn't solved any of the structural problems because the structural problems live in the architecture, not in the absence of a chatbot. ## Why incumbents can't go AI-native Understanding why they failed the AI test requires understanding why going further is structurally difficult for large travel companies. **Legacy code and architecture.** Expedia's technology platform was built in the late 1990s and has been modified continuously since. Booking.com's stack evolved over more than two decades. These are massive systems with millions of lines of code, thousands of engineers, and decades of accumulated technical decisions. You cannot rebuild them around a conversational AI paradigm without essentially starting over. And starting over while running a business that processes billions of dollars in bookings is not something boards of directors approve lightly. **Organizational structure.** Large OTAs have teams organized around the existing product. There's a search team, a results team, a booking funnel team, a payments team, a hotels team, a flights team. Each team optimizes their piece. An AI-native product doesn't have those pieces. It's a conversation. Where does the conversation team sit in an org chart built around a search-results-booking funnel? The organizational restructuring required to truly go AI-native is as hard as the technical restructuring. **Revenue model dependency.** Expedia makes money through commission on bookings facilitated by their search and display infrastructure. The value they provide to suppliers (hotels, airlines) is traffic and visibility. An AI agent that picks the best option and books it removes the visibility that suppliers pay for. Booking.com's advertising revenue depends on users seeing listing pages. Google's entire business is search and ads. AI-native booking threatens the core revenue mechanism of every single incumbent. **Existing user base.** Expedia has hundreds of millions of registered users who know how the product works. Redesigning the entire experience risks alienating people who are comfortable with the current one. Change aversion is real. Incumbents must thread a needle between innovation and familiarity that startups don't face. ## What Hopper and Kayak are doing differently Not every travel company is stuck in the same rut. Hopper and Kayak deserve separate mention for trying harder than most. Hopper built its business around AI-powered price prediction. Their core feature tells you whether to buy now or wait for prices to drop. This is genuinely useful and genuinely AI-powered. It addresses a real user anxiety (price uncertainty) with real data science. Hopper has also moved into fintech with price freeze features and cancellation protection, creating revenue streams that don't depend on search advertising. But Hopper's AI is narrow. It predicts prices. It doesn't have a [conversational booking](/blog/ai-travel-booking-conversation-first) flow. It doesn't remember your preferences across trips. It doesn't handle the full trip lifecycle. The core booking experience is still a search form with results and filters. Hopper solved one problem with AI and left the rest untouched. Kayak added a conversational feature and it's better than most. You can ask Kayak's AI about flights and it does return some real results within the conversation. But Kayak is fundamentally a meta-search engine. It aggregates results from other sites and sends you there to book. The AI conversation ultimately leads to a redirect. Kayak's business model depends on those redirects, so the AI can enhance the search but can't replace the hand-off. Both companies show that incremental AI adoption is possible. But incremental adoption is not transformation. Using AI for one feature (price prediction) or one step (search) while leaving the rest of the experience unchanged doesn't address the fundamental problem: booking travel is still a fragmented, exhausting, multi-tool process. ## The window of opportunity Here is the honest assessment of where things stand. About 65% of consumers under 35 say they're comfortable with the idea of AI booking travel for them. That number has been climbing steadily. User readiness is not the bottleneck. Technology capability is not the bottleneck. The bottleneck is that the companies with the most travel distribution power are structurally unable to rebuild around AI. This creates a real window. Not a permanent one, because eventually the incumbents will figure it out or acquire someone who has. But a window. The window exists because: **AI capabilities are advancing faster than incumbents can adopt them.** Every quarter, the underlying models get better at handling ambiguity, maintaining context, and executing multi-step tasks. An AI-native product benefits from these improvements automatically. An AI feature bolted onto a legacy product benefits minimally because the feature's scope is limited by the surrounding architecture. **Users are training themselves.** Every person who uses ChatGPT, every person who sends a voice note instead of a text, every person who asks Alexa a question is becoming more comfortable with conversational AI. The behavioral shift is happening independently of any travel company's efforts. When those users encounter an AI travel booking experience that actually works end-to-end, they'll adopt it fast. **The incumbents are moving slowly.** It has been three years since the initial AI announcements. Expedia's chatbot is functionally similar to what it was at launch. Booking.com's trip planner still doesn't book. Google Flights still redirects. The iteration pace on AI features within legacy products is glacial compared to what an AI-native company can achieve. **The conversion gap remains.** OTA [conversion rates](/blog/low-conversion-rates-ai-fix) are still stuck in the low single digits. Cart abandonment is still above 80%. The fundamental user pain, too many options, too much complexity, too little help, is completely unaddressed by the AI features the incumbents have shipped. This isn't a small gap to exploit. It's a canyon. ## What AI-native actually means We talk about "AI-native" a lot, and it's worth being precise about what it means in contrast to what the incumbents built. AI-native means the AI agent is the product, not a feature of the product. There is no search form. There is no results page. There is no filter sidebar. There is a conversation. You tell the agent what you want. The agent searches live inventory, compares options, applies your preferences, and presents a curated selection. You choose. The agent handles payment and confirmation. Your trip is created, your itinerary generated, your documents stored. All in one flow, all in one conversation. AI-native means the agent has memory. It knows your seat preference, your airline loyalty, your budget comfort zone, your hotel style. It learned these from your conversations and your bookings. The tenth time you use it is radically different from the first time. This is not a recommendation algorithm saying "users who booked this also booked that." This is an agent that actually knows you. AI-native means the architecture was designed for conversation from the ground up. The streaming infrastructure, the tool orchestration, the [state management](/blog/state-management-ai-conversations), the payment integration, all of it built assuming the primary interface is a chat, not a web page. You can't retrofit this. Bolting a chatbot onto Expedia's architecture gives you Expedia with a chatbot. Building from scratch gives you something different in kind, not just degree. AI-native means end-to-end. Not just search. Not just planning. Not just one part of the trip. Flights, hotels, trip management, documents, tools, ongoing assistance throughout the journey. One product. One conversation. One agent that handles it all. That's the test the incumbents failed. Not because their AI is bad, but because they used good AI to enhance a bad paradigm instead of replacing it. ## Where the industry goes from here My prediction is that the travel industry is headed for a split. The incumbents will continue adding AI features to their existing products. Those features will get incrementally better. Expedia's chatbot will eventually be able to do more. Booking.com's planner might gain some booking capabilities. Google will weave AI more deeply into search results. But the fundamental architecture of those products, the search-results-booking funnel, will persist because it's too valuable and too entrenched to abandon. The incumbents will get maybe 30% of the way to AI-native before their organizational and economic constraints stop them. Meanwhile, AI-native products will capture the users who want something fundamentally different. Not a better search engine. Not a smarter filter. An agent that does the work for them. These users will skew younger at first (the 65% under 35 who are ready for this), but the demographic will widen as the products prove themselves. The market is large enough for both to coexist for a while. Global travel is a trillion-dollar industry. Even capturing a small percentage with a fundamentally better experience represents an enormous business. But the trajectory is clear. Every year, AI gets better and users get more comfortable. Every year, the argument for the old paradigm gets weaker. The question isn't whether AI-native travel booking will become the default. It's when. And the incumbents' half-hearted AI adoption is accelerating that timeline, because it's showing millions of users what AI could do for travel while simultaneously demonstrating that the incumbents aren't the ones who will deliver on the promise. The best thing Expedia, Booking.com, and Google did for AI-native travel startups was launch their AI features. They validated the market. They trained users to expect AI in travel. And then they under-delivered just enough to leave the door wide open. --- Nowah is an AI travel agent that searches and books real flights and hotels through conversation — no filters, no thirty open tabs. [Plan your next trip](https://app.nowah.xyz).