How ChatGPT Stopped Hallucinating

Six months ago, our demo was bulletproof. We'd ask a potential customer to pick a hotel in their portfolio and then open ChatGPT and ask for rates in that market. It would confidently invent numbers—wrong prices, misidentified hotel brands, rates from OTAs that didn't even list the property. The AI wasn't just wrong—it was creatively wrong. Hallucinating with conviction. We'd all laugh.

We can’t play “Laugh at the LLM” anymore.

Today: ask ChatGPT for a rate and it answers perfectly. "$289 per night via Booking.com." An OTA link. An accurate price. Real-time availability.

What changed?

ChatGPT stopped guessing. It started quoting. And the source it's quoting from—reliably, consistently, almost exclusively—is Booking.com's live inventory feed. Same story with Google AI Mode. When it recommends hotels with booking options, it's pulling from Google Hotel Center—the platform where hotels send their rates, availability, and property data that powers Google's travel products.

This is a fundamental shift in how travelers discover and book hotels. AI-based search is creating immersive, conversational planning environments where the entire journey from inspiration to booking happens in a single interface. The data infrastructure powering these experiences needs to be more technical, more structured, and more comprehensive than ever before. If you're not feeding these systems properly, you've already lost the game before it starts.

Why AI Platforms Converged on Feeds

At The Phocuswright Conference 2025 in San Diego, Google's James Byers (Group Product Manager for Travel in Google Search) made the economics explicit. When asked about the commercial model for agentic booking in AI Mode, his answer was revealing:

"At the moment, we see this being a similar model that we have today. We'll adapt it if we need to, but this is really meant for every partner, for every way to buy and have as open an ecosystem as we can have."

AI Mode isn't building a new commercial framework. It's using the same infrastructure as Google Hotel Ads—CPC, CPA, and the same Hotel Center feeds. Byers announced a few partners: Booking.com, Expedia, Marriott, IHG, Choice Hotels, and Wyndham. Accurate and bookable data (combined with the opportunity to advertise) will remain core to Google's travel search experience.

Byers was emphatic about what makes this model work:

"Preparation is in two parts. One is: don't abandon the foundations you have today. Feed data, prices, and availability in these domains are still incredibly important because that's how you anchor the conversation."

Notice the language: "anchor the conversation." If your feed is broken or inaccurate, the AI can't anchor anything. It can't recommend you. It can't complete a transaction. And if the price in your feed doesn't match your checkout price, trust is broken immediately.

This is the new technical bar: your first-party data needs to be more structured, more accurate, and more comprehensive than ever before. If your Hotel Center feeds aren't pristine, AI Mode will default to showing an OTA listing. Not out of preference, but out of necessity.

The message is clear: AI platforms will default to whoever provides the most reliable, consistent data. Right now, that's the OTAs. The "boring" work of feed management and ARI (Availability, Rates, Inventory) is now more critical, not less.

But here's the opportunity: these AI systems want to do more with your data. They're trying to answer increasingly complex queries like "Find me a hotel in Napa with a quiet reading nook and vegan breakfast options." This is precisely the detailed information that a check-the-box, fill-in-a-form, aggregated OTA feed lacks. This is how you compete.

Does Your Direct Channel Match Your Feeds?

Here's where most hotels fail without realizing it. Most of you are already sending structured data to Google Hotel Center—rate feeds, availability, room types. And if you aren't, you have to start. But when a potential guest lands on your website, does that same precision carry through?

Google's developer documentation is explicit. Your hotel must ensure alignment across three layers:

Your Hotel Center feed (what you're sending Google about availability and pricing)

Your on-site structured data markup (JSON-LD on your landing pages)

Your booking engine display (what travelers actually see and can book)

According to Google's Hotel Price Structured Data Reference:

"It is important to ensure that you have updated structured data values in your website that match with the values shown to users in your UI in order to provide reliable information and also to conform with Google's Price Accuracy Policy."

Google's documentation on structured data for Hotel Ads price accuracy sets a hard threshold:

"Google requires a 98-100% accuracy rate for SDM implementation. This means that the price in the markup data on a webpage must always match the price displayed to the shopper. If the accuracy rate falls below 98%, the markups will not be used for price accuracy checks until the issue is resolved."

If these three layers don't align, you fail the test. And AI Mode will default to showing an OTA listing instead of yours.

This is where the traditional hotel org chart breaks down. You've got Sales managing OTA relationships, Marketing running your website, Revenue optimizing pricing, CIOs building out the booking and technical infrastructure. The bigger brand companies can and will organize across these functions to meet the emerging distribution landscape. But few small hotels have the ability to coordinate efforts across disparate systems to make their hotel more bookable in an integrated world where guests employ an AI at the controls.

The Opportunity: Help AI Deliver Better Experiences

Being OTA-level good at feeds gets you into the game. But here's where hotels can actually win.

At Phocuswright, Byers made a crucial distinction: while price feeds anchor the transaction, descriptive content anchors the discovery. AI Mode is building immersive planning environments where travelers ask complex questions like "weekend trip to a European city with Christmas markets" or "hotel near Napa with a quiet reading nook."

These AI systems want to answer these questions well. Byers specifically advised travel companies:

"If you know the niche bits of your business about what travelers care about that maybe never really surfaced before in Search... there's a new opportunity to surface that content. So use human voices, use factual content, get all those little bits of nuance out there because users will find them."

This is the shift: your content infrastructure needs to be as technical as your ARI feeds. Not buried in blog posts or generic property descriptions, but structured, indexed, and tied directly to specific offerings:

Amenity details tied to specific rooms: "Deluxe King with heated floors and blackout shades"

Calendar events exposed as structured data: "Live jazz every Friday at 7pm in the lobby bar"

Third-party authority linked to offerings: Press mentions, awards, influencer content tied to specific room types

Niche features AI can cite: Vegan breakfast options, specific gym equipment, pet-friendly policies with weight limits

OTAs have pristine feeds, but they're generic. They can't tell travelers about your rooftop pool heated year-round, your Saturday kids' craft hour, or your chef sourcing ingredients from local farms. You can if you structure it properly.

The AI Storefront: Website + Feeds + Content Indices

The practical answer is what we call an AI Storefront—not just a website or a feed, but a combination of all three: your website content, your data feeds, and structured content indices that AI systems can reliably pull from. Think of it as your property's canonical surface for AI assistants.

It serves two functions: first, it mirrors your feeds with perfect accuracy—every price and availability window in your Hotel Center feed is also exposed through Hotel and HotelRoom JSON-LD. When AI Mode validates your pricing, it finds perfect parity across feeds, structured data, and visible UI. Second, it enhances feeds with structured content—the rich, niche details that help AI answer complex queries.

This is where Kismet enters. We build AI Storefronts that normalize your ARI data in OTA-grade formats while structuring your existing content into AI-readable indices. We keep your Hotel Center structured data in sync with your feeds and booking engine automatically—a living system that AI assistants can reliably pull from while still handing human visitors to your brand.com for the booking experience.

If you read my previous post—"How to Beat the OTAs at AI Search"—think of this as Part 2. That piece explained why OTAs win by default. This one lays out the technical infrastructure hotels need to compete.

The New Bar for Being "Official" in AI

Let me distill this down to one principle that Google AI Mode and ChatGPT are converging on:

"Trust whoever gives us accurate feeds and matching landing pages. Delight users with whoever has the best content."

Right now, OTAs are winning because they nail the first part. They have industrial-grade feed infrastructure and pristine data hygiene. Hotels have been losing because they've treated feeds as an afterthought and relied on outdated SEO tactics.

But hotels have a structural advantage on content—if they're willing to structure it properly. The stories, the local expertise, the brand voice. That's all sitting in your CMS and social feeds right now. The challenge is making it machine-readable and feed-integrated so AI can use it.

To win in AI, your direct channel has to look like an OTA on the back end and feel like your hotel on the front end. No shortcuts.