Part 1 of our AI for Families series — practical takes on how AI tools are changing the way we travel and live as a family.
I asked ChatGPT to plan a week in Lisbon for a family with three kids — ages 10, 8, and a very mobile 3-year-old who cannot be trusted near anything breakable. The itinerary it produced looked solid. It suggested a fado museum on day two. I don’t know what it was thinking.
We’ve been using AI tools for travel planning for about two years now, across trips to Portugal, Mexico, Japan, and a few shorter hops from Mauritius. Here’s what we’ve found.
What AI gets right
The logistics side, mostly. If you need to figure out the least painful routing from Mauritius to a city with no direct flight, or you want to know whether a 6-hour layover in Dubai is enough time to leave the airport with kids, AI is faster and more patient than any travel forum. It doesn’t get annoyed when you ask the same question four different ways.
It’s also good at building a rough shape of a trip. Ask it to give you a 10-day itinerary for a family with mixed ages and a reasonable budget, and it’ll produce something workable within seconds. Not finished. But workable — a skeleton you can pull apart and rebuild rather than starting from nothing.
We used Claude to map out our Portugal trip before we committed to anything. It split the time sensibly between Lisbon, Sintra, and the Algarve, flagged the driving distances, and reminded us that road-tripping with young kids means you need more buffer days than the math suggests. All accurate. All things we knew but hadn’t written down in one place.
The restaurant discovery piece has improved too. Tools like ChatGPT with browsing, or Perplexity, will pull local-ish recommendations if you push them — ask for “restaurants near Alfama that have high chairs and aren’t in every tourist guide” and you’ll get something more useful than TripAdvisor’s top 10. Sometimes.
Where it goes wrong
It hallucinates details with total confidence. This is the problem. It’ll tell you a specific museum opens at 9am when it actually opens at 10am. It’ll list a restaurant that closed two years ago. It’ll describe a “family-friendly beach” that turns out to have a 40-minute hike and no shade.
The fado museum suggestion was actually fine in isolation — I looked it up and it would have been interesting for Indra. But it wasn’t filtered for a 3-year-old with a 45-minute patience window. This is the core limitation: AI doesn’t know your kids. It knows a generic family. Simmi at 3 has strong opinions about stairs. Krishav at 8 will walk himself into the ground if there are things to see. Indra at 10 mainly wants to know if there’s anything he can climb. A generic itinerary doesn’t account for any of that.
You have to give it a lot before it becomes useful. The more specific you are — ages, mobility, how long your kids last in museums, whether they eat anything other than pasta — the better the output. The generic prompt gets you the generic itinerary.
It also has no feel for pace. An AI-generated day might look perfectly reasonable on paper: two attractions, a meal, an afternoon activity. In practice that’s exhausting with young children. You need the dead afternoon where you do nothing except find a playground and let them run until they’re done. AI doesn’t plan for that, because it doesn’t know what it costs to keep three kids regulated across a full travel day.
How we actually use it
Research and structure, not decisions. The AI builds the scaffold. We pull it apart.
I’ll ask for a full itinerary, then cut about 30% of it. I’ll ask for every family-friendly activity in a city, then filter based on where the kids are right now — Indra is into history at the moment, Krishav wants anything with water. The AI gives me a long list quickly; the judgment about what makes the cut is still ours.
Lisa tends to use it for accommodation research. She’ll describe exactly what we need — ground floor or elevator, cot available, near a supermarket, not on a main road — and get a shortlist to investigate further. Better than an hour on Booking.com.
We’ve also started defaulting to Perplexity for destination-specific questions because it cites sources. If it says a particular beach is good for toddlers, we can click through and check when that was written. That’s not a small thing.
The actual bottom line
AI trip planning is useful the way a well-read assistant is useful — someone who’s been everywhere but never with your specific family, and who occasionally makes things up. You wouldn’t hand them the trip and walk away. But you’d use them as a starting point.
The families who find it frustrating are usually expecting it to replace the planning. It doesn’t. It changes the shape of the work. The judgment calls about what your kids can handle, what the family actually needs, what makes a trip worth the effort — those are still yours.
We’re planning Southeast Asia later this year. Indra has already started asking ChatGPT about temples. The kid has a process.
Next in the series: The family AI toolkit — apps and tools we actually use every week while travelling.