Listen to the podcast episode
Or listen now on Apple Podcasts or Spotify.
Key Takeaways
Try the tool before repeating Chesky’s take: Brian Chesky says a chat interface is not the solution for travel, but Odessia isn’t plain chat: it pulls in property previews, images, and descriptions, and builds an itinerary panel as you plan. That critique lands on version-one ChatGPT’s wall of text and links, not on a richer visual layer. Conrad’s read is that Chesky may not have actually used the thing he’s dismissing, and being early-and-wrong here is a risk for Airbnb too.
Recommendations without preference data are just guesses: Conrad asked Odessia to plan a Myrtle Beach staycation and its top pick was a 4.29-rated two-bedroom with 21 reviews. It never asked who was traveling, and he has four kids. The winning version knows your preferences, so the right ten listings for you aren’t the right ten for anyone else, and it can nudge you up to a nicer stay or down to better value from there.
Watch what the OTAs do, not what they say about AI: Despite billion-dollar AI budgets, Expedia and Vrbo’s booking interfaces look the way they did three years ago: destination, dates, guest count, search. Pull up archive.org and compare if you doubt it. All the chatter hasn’t changed the core flow, which tells you conversion data hasn’t demanded it yet.
The only question that matters is whether it converts better: If a new AI interface doesn’t meaningfully lift conversion rate, you’re chasing your tail so you can say you have AI. Paul’s heat maps on sites that bolted on chat widgets show barely any usage, mostly rudimentary discount-hunting chats. Bolting a chatbot onto an existing experience doesn’t automatically improve anything, and that cuts against vendors and Chesky’s critics alike.
Five minutes of talking beats a hundred clicks: The old-school travel agent’s superpower was listening before booking. Telling an AI about the four kids, the beach, and the budget passes far more context than tapping dates and a guest count ever will. An AI that interviews you, then builds the itinerary from that rich profile, might be the actual winning interface, and voice is the piece nobody has shipped yet.
Planning isn’t booking, and the gap is where trust breaks: Nearly 30% of travelers reportedly use AI for full trip planning, but Odessia still hands most bookings off to third-party sites to complete. When the trip disappoints, someone has to own the recommendation, and the AI won’t. Platforms spent years earning traveler trust by siding with guests when things went wrong, and AI booking will have to earn that same trust from scratch.
AI’s inconsistency makes it impossible to scale a playbook on: Ask the same question ten times and you’ll get seven different answers, and nobody can show you why a property was chosen. Paul sees similar sites with identical setups getting wildly different AI-referral traffic, hundreds of visits a week on one and dozens on another. Until there’s some predictability, treat AI visibility as a channel to watch, not one to bet the plan on.
The winners-win algorithm problem hasn’t gone away: A decade of algorithmic travel keeps pushing whatever already gets booked, leaving zombie listings with great reviews and no volume. In theory an AI could be fairer, surfacing the hidden-gem property with strong reviews and little activity. Nobody has demonstrated it actually works that way yet, so file it under hope, not strategy.
10x efficiency claims haven’t shown up as 10x revenue: If AI were truly 10x-ing code output and ticket processing, you’d see 10x the profit somewhere, and you don’t. Specific tasks genuinely get dramatically faster, but the business-level transformation remains unproven, including for travel platforms weighing AI interfaces. That’s a reason to test cheaply, not to rebuild everything on faith.
What We Cover In This Episode
Sparked by Brian Chesky dismissing chat as the interface for travel, Conrad and Paul test-drive Odessia’s AI trip concierge and debate what an AI planning interface actually needs: real personalization, proof it converts, accountability when a recommendation flops, and maybe voice. Along the way they check what Expedia, Vrbo, and Airbnb have actually changed, which is not much.
