TL;DR Booking.com's AI strategy puts only 20% of the effort into being visible in AI tools like ChatGPT. The other 80% goes to building value AI platforms can't copy and improving its own booking funnel, because Google sees intent but Booking.com sees the transaction. The biggest lesson so far: every safe bet aged poorly.
Around 60% of searches on Google and AI chat tools now end without a click. People read the AI summary, get their answer and never visit a website. For an online travel platform built on that visit, this zero-click shift is more than a trend to keep an eye on.
At NimbleFest 2026, Pranav Pathak shared how Booking.com deals with that shift. The approach is less a list of AI features and more a way of deciding where AI money should go.
How is AI changing search and e-commerce?
AI brings clear gains and new pressure at the same time. Across the industry, companies adopting AI report around 30% more developer productivity, 25 to 45% less time spent per support ticket, 60 to 80% better content quality and 20 to 40% more conversion in AI-enhanced areas. AI-driven traffic to e-commerce grew 758% year over year.
The other side of the ledger is less comfortable. LLM providers make money through ads and subscriptions, so they have every reason to keep users inside their own chat window. Customers expect AI-native experiences. Competitors are leaner. And three incentive models now pull against each other: Google wants to protect its paid clicks, LLMs want users to stay in the conversation, and businesses want to own the customer relationship themselves.
How does Booking.com split its AI effort?
Booking.com works in three tracks: about 20% of the effort goes to staying visible on AI platforms, 30% to building defensible value and 50% to improving the existing funnel.
Stay visible (about 20%). Integrate early with every new AI surface. Booking.com is live on OpenAI, Anthropic and Google. In ChatGPT, users can search Booking.com inventory directly through an API.
Build defensible value (about 30%). Create AI-powered assets only you can offer. One example is an AI-powered car rental help board built on authentic customer reviews and details like pickup times. OpenAI can't replicate that data. OpenAI and Google solve horizontal e-commerce. They don't solve travel. Domain expertise is the advantage.
Enhance the existing funnel (about 50% to start). Apply AI to the traffic and experiences you already have. When Booking.com added free-text search filters, one request stood out by a wide margin: hot tubs. The filter was built quickly. Conversion, personalization, pricing, content velocity and cost savings through support deflection all live in this track.
The split says a lot. The ChatGPT integration gets the headlines, but most of the work happens on Booking.com's own ground.
What can't AI platforms replicate?
AI platforms can't copy what only you own: unique user-generated content like reviews and listings, proprietary transactional data, exclusive supplier relationships, brand trust and distribution, operational complexity such as customer service at scale, and legal and regulatory expertise.
Transactional data is the one that sticks. Google sees intent. Booking.com sees the transaction. A search engine knows what people say they want. A marketplace knows what they actually booked.
That also shapes where Booking.com draws the line with AI platforms. The rule is simple: discovery can happen in ChatGPT, but conversion, payments and servicing the reservation stay on Booking.com.
How does Booking.com decide which AI projects to build?
Every AI initiative at Booking.com passes through four screening questions:
- What proprietary asset does it use?
- Can a user get the same outcome via ChatGPT? If yes, they will.
- Does this make us harder to replace, or just easier to use?
- Is the ROI measurable and meaningful in the short term?
Next to that, Booking.com avoids one-way doors: decisions about who owns user intent, who owns ranking and prioritization logic, and who owns the memory of past outcomes. If a decision can't be reversed, avoid it or only take it with an exit path in place.
How does Booking.com measure AI ROI?
Every AI initiative has to justify itself in business metrics like conversion or shipping speed. Cycle times, PR counts or feature usage don't count.
The calculation doesn't need to be exact, though. What matters is the order of magnitude. The difference between $100M and $1M is meaningful. The difference between $100M and $105M isn't. Everything launches as an A/B or multivariate experiment across markets and languages. What works expands, what doesn't gets cut.
That discipline pays off. Early on, 5 out of 100 ideas made it to production. Today Booking.com runs 50 to 60 AI initiatives a year, and 75 to 80% of them reach production.
What would Booking.com do differently?
Booking.com would take more risks. Looking back, it was overprotective of publicly available data, like TripAdvisor data that ended up with OpenAI anyway. It spent too much effort optimizing for LLM costs, while prices dropped fast. And it played too safe: every risk Booking.com took paid off massively and the safe bets aged poorly.
What did work: anticipating demand, integrating early with new AI providers, and saying yes to partnerships whenever they were a two-way road.
What Booking.com learned about chatbots, new AI models and AI discoverability
On chatbots, Booking.com moved away from an open "talk about travel" assistant toward contextual entry points at specific moments in the journey. That brought more engagement and better outcomes.
On keeping up with every new model, the answer was to stop trying. Booking.com works from core LLM capabilities like intent detection, summarization, dialogue and moderation, and adds its domain expertise on top.
AI discoverability (also called LLM SEO or GEO) works a lot like SEO, with different parameters. It's still an unsolved problem.
What to take home
Run your current AI projects past the four questions. Check which ones use an asset only you have, which ones a user could replace with a ChatGPT prompt, and which decisions you can't undo. Then look at how your effort is split. If most of it goes to being visible on someone else's platform, you may be building on their ground instead of yours.
