When the trip no longer starts with a date: what Kayak is telling us about AI and the future of travel search

For years we took one thing for granted: a trip begins when the traveller knows where they want to go and when they want to leave. That was the implicit assumption behind almost every travel search engine, every booking engine, every metasearch, every OTA and, when you get down to it, behind plenty of hotel and destination websites too.

Origin, destination, dates. Then the comparison begins.

The interview published by Travel Weekly with Kayak's leadership puts its finger on a shift that is far more interesting than it looks at first glance: artificial intelligence is bringing back to the centre a phase of travel that the old digital tools handled badly — the one where desire is still fuzzy, incomplete, unformed.

And perhaps the most striking data point is this: according to Kayak, users of Kayak AI are ten times more likely than the average user to start a search without a specific destination or defined dates. They don't ask "Milan–New York, 10 to 14 October." They ask: "I need to escape this awful weather — where can I go?"

For anyone working in tourism, that is not a technical footnote. It is a paradigm shift.

The old travel search was precise, but rigid

The travel engines we know were built for a very rational type of query. They worked well when the user had already decided almost everything:

• where they were departing from;
• where they wanted to go;
• when;
• with what constraints on price or duration.

It was a logic that was perfect for comparison and booking — far less so for the inspiration phase.

And that is exactly where AI is carving out new territory. Phocuswright has said it clearly on more than one occasion: travel search is becoming less and less a sequence of filters and more and more a natural-language conversation, especially in the early stages of the funnel.

In practice, the traveller is no longer forced to adapt to the logic of the tool. It is the tool that tries to adapt to the way the traveller actually thinks.

And anyone working in a hotel or a DMO (destination management organisation) knows how true that is. People don't think in fields to be filled in. They think in desires, moods, constraints, contexts. They want sunshine, quiet, ease, somewhere that works with the kids — or maybe just two days to breathe.

AI widens the top of the funnel and changes the value of discovery

The central point of the Kayak interview is not so much the volume of conversations — over 100,000 a month — which is still relatively small compared to overall traffic. The point is the quality of those conversations and what they reveal about emerging behaviour.

If the user arrives without dates and without a destination, it means the AI interface is claiming space in the most delicate part of the customer journey: the part where desire is shaped.

And that phase matters far more than we sometimes think, because whoever controls discovery also controls a share of the future value.

We already see it with Google, which is working increasingly explicitly on agentic features for travel and commerce. We see it with ChatGPT, Perplexity and the recommendation systems embedded in major platforms. And now we see it in Kayak, which is trying not to get squeezed between the classic OTAs and the new AI-native players.

Why Kayak thinks it still has an advantage

In the interview, Peer Bueller and Matthias Keller defend a thesis I find very clear-headed: in a world where inventory risks becoming a commodity, competitive advantage lies increasingly in the quality of the user experience and in the ability to make the complexity of travel comprehensible.

Bueller puts it very concretely: price comparison and product comparison don't work well in pure long-form text. To genuinely grasp the differences between a flight, a fare or a service class you still need a visual representation — a matrix, an interface that simplifies the chaos.

And here Kayak starts from a very strong asset: years of experience comparing prices, classes, bundles, fare rules, filters and variables that in travel are anything but trivial.

Translation: language models may be similar for many players, but not all of them have the same know-how when it comes to turning the confusion of travel supply into a legible decision.

The real issue: leisure travellers are still cautious, corporate ones are not

Another very important passage in the interview concerns the difference between leisure and corporate.

According to Keller, agentic booking makes more sense today in the more constrained travel spaces — especially corporate travel, where the traveller already knows what needs to be booked and simply wants to do it faster and with less friction. In leisure, by contrast, the stakes are still too high for the user to delegate everything comfortably to an AI agent.

This is very consistent with other market signals. In business travel, where policies, budgets, routes and preferences are more predictable, agentic automation has a much more immediate ROI. In leisure, the trip remains an emotional, identity-laden purchase — far less something you hand over to someone else.

And this is where tourism faces a very interesting hybrid zone: not a world where AI books everything on its own, but a world where AI accompanies us much earlier, helping us formulate our needs more clearly.

The big lesson for hotels and destinations: travellers no longer search the way they used to

This, in my view, is the real news.

Not so much that Kayak has an AI feature. But that the grammar of travel demand is changing.

If a trip increasingly begins with vague, natural, open-ended phrases, then hotels and destinations need to stop thinking exclusively in terms of rigid keywords and start making themselves findable within more nuanced searches, such as:

• where to go if I need to disconnect for two days;
• easy European cities to reach without too much stress;
• quiet Italian spots with good food and a station nearby;
• places that aren't too hot in August where I can work and rest.

These are no longer questions from niche travel blogs. They are increasingly questions asked directly at the main interface.

What this means for tourism content

From where I stand, anyone working in tourism marketing needs to make a fairly sharp leap.

It is no longer enough to build content to rank on pure hard-and-fast keywords. You need to write, organise and structure content capable of responding to more complex, more human needs — closer to a real conversation.

For hotels this means, for example:

• clearly explaining who they are really talking to;
• spelling out not just the amenities, but the type of experience they offer;
• using a language that is closer to the guest's real needs — less technical, less standardised.

For destinations, it means telling their story not only through attractions, but also through use cases.

A person isn't just searching for "Levanto" or "Lago di Garda." They might be looking for a place to do some remote working for a few days, a car-free weekend, an autumn escape with thermal baths and walks, or a holiday with children that doesn't feel like an obstacle course.

Whoever learns to be read well in this new grammar will have a growing advantage.

The hallucination problem hasn't gone away — but neither has the potential

Bueller is fairly direct on this too: AI has improved enormously, but it is still full of hallucinations. His view, however, is that these are preventable problems and that quality can improve substantially.

Here we need to be very pragmatic.

In tourism, mistakes are never neutral. A wrong answer about a baggage policy, a visa requirement, a transfer or a hotel can ruin a real experience — not just a digital interaction.

For this reason, the travel sector will probably continue to move more slowly than others toward full automation. But this does not reduce AI's potential in the exploration and orientation phases. If anything, it makes it more interesting: it is a space where users accept help but still want to keep control.

A broader reflection: demand is becoming more open, but supply is still too rigid

And here, in my view, a very sharp contradiction in our industry comes to the surface.

On one side the traveller is starting to think in more fluid, less linear terms — less tied to specific dates and destinations. On the other side, the vast majority of hotel websites, booking engines and tourism portals are still built on rigid models that require the user to have already figured everything out.

It is as if the customer arrives saying "help me work this out" and we reply "come back once you've filled in all the fields properly."

This applies to hotel websites, but it applies even more to many Italian destinations, which often have information portals that are abundant but not very dialogic — rich in content but poor in real guidance.

The risk is that the discovery phase gets almost entirely captured by third parties — from Google to Kayak, from ChatGPT to the OTAs — while official sites remain tools for confirmation or service, and no longer for inspiration.

What anyone working in tourism should be doing now

1. Rethink content for open-ended questions

We need to produce content that responds to needs and situations, not just rigid keywords. Travel content will have to be more empathetic, more useful, more readable by AI and by humans alike.

2. Work on clarity and consistency

If a conversational engine has to "understand" who we are, we cannot afford vague descriptions, standard promises and incoherent content spread across the website, OTAs, Google Business Profile and social media.

3. Accept that the inspiration phase is changing

We can no longer assume the customer will arrive already ready to book. More and more often they will arrive with an idea that has been generated or filtered by an AI system. And we will need to be good at confirming it, refining it or correcting it.

4. For corporate, watch closely

The business segment will probably be the first where we see genuine agentic booking automation in the wild. Anyone working in corporate travel would do well to monitor this shift very carefully.

All told, the Kayak interview gives us a very useful snapshot of the moment we are living through. AI is not just speeding up travel search. It is changing the point from which that search begins.

If the traveller no longer starts with a date and a destination, but with a mood, a problem or a desire, then the entire tourism ecosystem needs to adapt.

For Italian hotels and destinations this is an enormous challenge, but also an opportunity. Because whoever learns to tell their story well inside this new conversational language will be able to intercept the trip long before it becomes a booking.

And, as always, whoever gets there first in the traveller's head has already done half the work.