The vendor layer wins again

What it doesn't account for is that the chatbot converts badly at the point of purchase well before you get to hospitality-specific complexity. Walmart ran the clearest test we have: 200,000 products through OpenAI's Instant Checkout, and Daniel Danker, their EVP of product and design, reported conversion rates three times lower inside ChatGPT than when users clicked out to the website. Retail, with cleaner pricing and simpler inventory than any hotel booking, still couldn't close at a normal rate inside a chat window. That's an interface problem, not an industry problem.

Chat strips out the context that actually moves people toward a purchase decision - brand imagery, reviews surfaced at the right moment, the upsell nudge, the basic reassurance that they're not about to make a mistake. Working on UX and conversion for hotel websites, one pattern shows up consistently: any break in brand continuity has a measurable cost. When users land on an experience that feels inconsistent with what brought them there, trust drops and they default to whatever they already know. In hospitality, that's almost always an OTA. A chat interface that abstracts away the hotel's identity doesn't just flatten the brand, it actively redirects the booking.

This also isn't a new chokepoint for hotels. Most properties don't control their own checkout flow because it sits inside a white-label widget from a booking engine vendor, running on its own logic and its own integration roadmap. When Google tried to own the transaction with Book on Google, that vendor layer is exactly what prevented the connection from working cleanly, and it's sitting in the way now for the same reason.

OpenAI moving transactions to embedded merchant checkouts is probably an honest read of where conversation earns its place. The top of the funnel - research, narrowing options, resolving uncertainty - is where the chatbot is genuinely useful. The transaction is a different problem and needs to be treated as one.