KAYAK's Ask AI and the Boundary of Conversational Commerce: AI-Led Discovery Grounded in Live Inventory

Live Search That Keeps Moving Next to the Chat

KAYAK Launches Ask AI to End the Chat-to-Search Disconnect

KAYAK: How behavioral data from 100,000+ monthly AI conversations shaped Ask AI, which combines conversational planning with live search

www.cxtoday.com

KAYAK, one of the largest travel metasearch platforms, announced Ask AI on April 23, 2026, a conversational travel planning tool. As users describe what they want in plain language, the flight, hotel and rental car results on the same screen update in real time with the conversation. What appears is drawn from actual prices and bookable options across hundreds of travel partners, with no redirects to another page and no restarting the search.

This article reads the launch not as a travel-industry feature release but as an implementation case study in AI commerce, meaning selling in which conversational AI carries the journey from product discovery toward purchase. The further stage, where AI agents execute comparison and transactions on the user's behalf, is called agentic commerce. Metasearch is a business built entirely on discovery and comparison across many sellers' inventory and prices. Where such a specialist chose to insert conversational AI, and where it deliberately did not, speaks to design decisions facing any business that wants AI to sell its products.

The timing coincides with search demand around the World Cup, held across three North American countries in the summer of 2026. KAYAK's data shows flight searches to U.S. host cities up 23% from last summer, with Kansas City surging 168%. Hotel prices have climbed 36% in the U.S., 55% in Canada and 119% in Mexico. Demand for complex multi-city itineraries built inside a single conversation is the feature's first proving ground.

The Boundary Line Drawn by 100,000 Monthly Conversations

Ask AI's design rests on operating data from a predecessor. In October 2025, KAYAK launched AI Mode, a natural language search combining ChatGPT with KAYAK's travel data, and the first feature to graduate from the company's internal sandbox, KAYAK.ai, to the main site. Typing a request in everyday language returns candidates drawn from the latest prices across hundreds of providers.

Running it surfaced the boundary of where AI actually helps. According to Chief Product Officer Matthias Keller, analysis of more than 100,000 monthly AI conversations showed that AI is especially useful in the exploratory phase, when the trip has no fixed shape, but travelers do not make decisions from a handful of AI suggestions alone. They shape the trip through conversation while destinations and dates are still vague, then return to the familiar results pages and filters when it is time to compare and commit.

Mapped onto a purchase funnel, what AI has reliably captured is the entry point where demand gets put into words. The moment money moves still belongs to the comparison UI. KAYAK concluded the disconnect was a problem of experience design and arrived at chat and search cooperating on one screen, rather than chat replacing search. Keller put it plainly in his interview with CX Today.

"

The right model is not chat instead of search, but chat working alongside search.

Source: Matthias Keller, Chief Product Officer at KAYAK

Visions of AI shopping tend to assume conversation will replace form-based UIs outright. Measuring AI's actual reach against first-party data at the scale of 100,000 conversations a month, and choosing coexistence over replacement, makes this a useful reference point for gauging the distance between conversational commerce and agentic commerce.

Designing Trust by Grounding Suggestions in Live Inventory

The wall conversational commerce has to climb is trust. Booking a flight or hotel involves real money and plans that cannot afford to fail. When a chatbot returns a few options with no surrounding context, choosing one becomes a leap of faith. The same structure applies to any e-commerce category with high prices or date constraints.

Ask AI's answer was to ground the AI's suggestions in live, bookable search results rather than curated recommendations. Each time the user refines the brief in chat, the results list on the same screen updates instantly. Travelers keep the ease of conversation while comparing prices, timings and cancellation policies with the transparency they are used to. Keller notes that travelers now take longer to decide, track prices closely and scrutinize cancellation policies earlier in the process, and the design takes that iterative adjust-and-reconsider behavior as its premise.

What should not be missed is that what actually runs beneath the conversation is live search itself. Whether a product appears on an AI's suggestion surface depends on inventory and prices being supplied in real time, in machine-readable form. For KAYAK, feeds from hundreds of travel partners are that substrate. Translated to physical e-commerce, it corresponds to a data foundation where product attributes, stock and prices are structured and reflected without lag. The design is a clear demonstration that conversational selling stands on the unglamorous groundwork of product data readiness.

Where the Conversation Lives, and What External AI Dependence Costs

There is a second design decision inside Ask AI: where the conversation lives. While Expedia and Booking.com shipped apps that run inside ChatGPT, KAYAK embedded the conversational experience in its own site and apps. Opening a storefront inside an external AI brings new traffic, but the dialogue data depends on the platform's design. Keeping the conversation on an owned surface lets the company accumulate chat logs as first-party data, meaning customer data a business collects and holds itself. That Ask AI was itself derived from 100,000 monthly conversation logs is proof of what that accumulation is worth.

Behind this choice lies the experience of paying the cost of external dependence first. In October 2025, Booking Holdings took a $457 million writedown on KAYAK. KAYAK CEO Steve Hafner attributed it to changes on the search surface, including Google's AI Overviews, that thinned free traffic and inflated paid advertising spend. When the entry point of discovery moves into someone else's AI layer, the cost of appearing there gets passed on to the business. A company that grew on search traffic came to feel the owned-channel versus external-AI-channel structure earlier than almost any industry.

Booking Holdings is also laying groundwork for conversation as the entry point of transactions. In May 2026, Skift reported that KAYAK co-founders Steve Hafner and Paul English are preparing Lola, a conversational AI venture for the group. An owned conversational surface, a measured distance from external AI layers, and a next conversational business: the posture toward AI-led discovery is being assembled across several layers at once.

An Evolution of Metasearch, Not Automated Transactions

It matters just as much not to overread Ask AI. KAYAK itself frames it as an evolution of metasearch, not a replacement for the core tools travelers trust when it is time to book. Ask AI conversationalizes discovery and comparison; it is not an autonomous agent that books or pays on the user's behalf. No announcement confirms KAYAK adopting payment delegation, identity verification or agent-facing checkout at this point.

The line KAYAK drew coincides with the split between discovery and checkout that runs across AI commerce today. Even when conversation captures demand, automating inventory holds, delegated payment authority, and post-purchase cancellations and refunds requires a different set of foundations. KAYAK's behavioral data adds that users are not ready either: for now, they trust a decision made with full comparison transparency more than an experience that moves from a few suggestions straight to purchase.

Brought back to retail and e-commerce, the case yields three things worth keeping. An experience design that runs AI assistance alongside the existing search-and-comparison UI instead of closing everything inside chat. A supply of real-time, machine-readable inventory and price data as the price of admission to AI suggestion surfaces. And the judgment to treat where the conversation lives, on an owned surface or inside an external AI, as a trade-off between data ownership and traffic costs. What gives the case its weight is that each lesson is backed by first-party data and an actual profit-and-loss consequence.

Conclusion

KAYAK's Ask AI demonstrates, with 100,000 monthly conversations behind it, a design in which conversational AI and traditional search cooperate on a single screen. AI wins the entry point of a purchase, where demand gets put into words; decisions still demand the transparency of comparison. That empirically measured boundary is a reference point for any business trying to make conversation the entry point of its transactions. Grounding suggestions in live inventory, and keeping the conversational surface and its data under its own roof, shows concretely how a company holds onto customer touchpoints as discovery moves to AI. Precisely because transaction automation still lies ahead, starting with machine-readable inventory and price data and an experience that pairs chat with comparison is the realistic preparation for retail and e-commerce operators.