Ask any general-purpose AI assistant to plan a five-day trip to Kyoto and you'll get something back within seconds: Day 1, Fushimi Inari in the morning, Gion in the afternoon. Day 2, Arashiyama bamboo grove, Tenryu-ji. Day 3... and so on. It looks complete. It covers the city. It is, in a technical sense, an itinerary.
It is also almost certainly wrong for you — because the AI produced it without knowing anything about you. It doesn't know whether you're a light walker who needs downtime, or someone who runs marathons and wants to hike to mountaintop shrines at dawn. It doesn't know if you've been to Kyoto before. It doesn't know if you care about temples at all, or if what you actually want is three days in the ceramics quarter talking to craftspeople.
What it did was answer a different question: "what are the canonical things to do in Kyoto, formatted as a schedule?" That's a useful thing to know. It's not a trip plan.
The Template Problem
Most AI travel tools make the same architectural mistake: they start with the destination and work outward. The destination becomes the prompt, the itinerary becomes the output, and the traveller becomes an implicit average. The resulting plans are technically accurate and generically useless — they'd serve the median tourist who wants to see the famous things, never miss a main attraction, and return home having been thorough.
There's nothing wrong with that if it matches what you want. But it's a narrow slice of how people actually travel. The slow traveller who wants to spend four days in one neighbourhood and skip the main sights entirely gets the same generic output as the first-time visitor who needs help navigating the must-sees. The solo woman travelling through the Middle East who wants to understand how safety considerations should shape her itinerary gets a list of the famous medinas. The couple in a long-term relationship where one person is a hiker and one person is a museum person gets an itinerary that serves neither.
Templates disguise themselves as plans. The difference is that a plan reflects the actual person taking the trip.
Why Asking Questions First Is the Whole Point
When we designed how Wanderglint works, the core decision was to make the first part of the conversation diagnostic, not generative. Before the system suggests anything, it tries to understand the shape of the trip the traveller actually wants.
What does a good day feel like to you? Early starts or late mornings? One long experience or several shorter ones? The kind of meal you linger over, or fuel-and-move? Are you comfortable with unplanned time, or does a day without structure feel like wasted opportunity? Have you been to this region before? What from previous trips landed, and what didn't?
These questions aren't small talk. They're the inputs that make a plan actually personal. The difference between "slow morning, one big hike, dinner worth booking in advance" and "eight sites a day, no hiking, primarily interested in architecture and food markets" produces completely different plans for the same destination — as it should. Those are different trips taken by different people, and they deserve different recommendations.
What Good AI Travel Planning Actually Does
There's a useful distinction between AI as information retrieval and AI as planning assistance. Retrieving information — what are the visa requirements for Morocco, what train connects Vienna to Budapest, what are the opening hours of the Prado — is something AI does well and that has been available in search form for a long time. It's genuinely useful, but it's not planning.
Planning is making tradeoffs. It's the recognition that a five-day trip to Lisbon cannot include every neighbourhood in the city, so you need to decide which three neighbourhoods to prioritise and why. It's understanding that if your flight arrives at 11pm and you're prone to jet lag, Day 1 should probably be light — not because that's what the guidebook says, but because you told the planner that your last trip to Europe started with a brutal first day and you never quite recovered.
Good planning also involves knowing what to leave unplanned. An itinerary that schedules every hour is a logistics document, not a travel plan. Some of the best things that happen on any trip happen in the spaces between the planned parts. A plan that respects that — that builds in deliberate unstructured time — is a better plan than one that treats free time as an inefficiency to be eliminated.
The Honest Limits of Any Tool
We'd be overstating things to claim that any AI system can fully replicate the judgment of a brilliant local friend who knows the city, knows you well, and can tailor recommendations to both simultaneously. That friend is irreplaceable, and if you have one for every destination you want to visit, you should use them.
What a well-designed AI travel planner can do is get meaningfully closer to that than a generic template. It can hold the constraints of your travel style across the conversation, surface the specific recommendations that match those constraints rather than the general recommendations that match the average traveller, and check the things that need checking — restaurant booking windows, accommodation logistics, activity timing — so you don't have to hold all of that yourself.
The bar isn't whether it's as good as a local expert who's your close friend. The bar is whether it's better than twelve browser tabs and a Reddit thread. And the gap between a question-first planning conversation and a destination-prompt itinerary generator is the entire distance between "tool that helps" and "output you have to rewrite before it's useful."
The Right First Question
The shift that matters most isn't technical — it's about what the first question is. If the first question is "where do you want to go?", you've already committed to treating the destination as the primary variable and the traveller as secondary. If the first question is "what does a good trip feel like to you?", you've inverted that relationship, and everything downstream of that first question becomes more useful.
This is why Wanderglint asks about travel style before it asks about destination, and why the conversation is structured to understand the person before suggesting the plan. The itinerary is the output. The travel style is the input. Confusing the two produces the template problem, and the template problem is why so many AI travel tools feel useful for thirty seconds and then get abandoned when it becomes clear that the plan they generated was built for someone else.
