Curiosity got the better of us. With AI trip planning tools everywhere lately, we decided to run a real test: give a popular AI assistant the exact brief a client might give us, a ten day family trip to Italy, five star service, kids in tow, no detail spared, and see what came back.
The itinerary looked great. In theory.
Here’s what happened when we looked closer.

The AI produced a clean day by day outline: fly into Rome, three nights in Rome, train to Florence, three nights there, drive to the coast, four nights in a seaside town. Hotels were name brand and easy to find. Restaurants were the ones that show up on every “best of Italy” list.
What it didn’t do was ask a single question about our family. Ages of the kids. Whether anyone had a food allergy. Whether we wanted a slower pace or a packed one. It handed back the same itinerary it would hand anyone who typed in the same three cities.
We asked it to book a specific cooking class in Florence that we happen to know is fully reserved months out unless you have a relationship with the host. The AI cheerfully told us it was available. It wasn’t. It had no way of knowing that, because it doesn’t have a person on the ground who’s actually called and asked.
Same story with a restaurant reservation in Rome that requires a phone call from someone the restaurant recognizes. The AI suggested walking in. Anyone who has tried to walk into a busy Roman restaurant on a Friday night knows how that ends. Most Americans do not realize that Italy doesn’t turn a table over two or three times a night like we do in the States.
The itinerary had no room built in for the unexpected good stuff, the vineyard visit that runs long because the owner wants to show you one more thing, the gelato spot that’s better because a local mentioned his cousin owns it, the villa staff arranging a private grape harvest activity for the kids because they know the family is arriving that week. None of that comes from a search engine. It comes from relationships built over years of actually being there.

Rain forecast for the coastal leg. A hotel that looks stunning in photos but is a twenty minute walk from anywhere useful with young kids in tow. The AI had no way to flag any of this, because it wasn’t building the trip with judgment. It was assembling one from information. This reminds us of when the OTA’s first came into play and everyone was a DIY travel planner. It’s just information, without context.
This isn’t just our experiment. It’s something we see constantly from clients before they ever reach out. Someone will spend a weekend going back and forth with an AI tool, typing in what they think they want, and getting back paragraph after paragraph of options. Three hotels in Florence, all described beautifully. Four possible routes through the Amalfi Coast. A list of “top rated” villas with no way to tell which one actually suits their family.
Then they’ll copy and paste that same wall of text to us, hoping it’s a starting point.
It rarely is. The AI never had to actually decide anything. It just generated more options, because that’s what it’s built to do when you ask it a broad question. Nobody picked the hotel. Nobody weighed the tradeoffs. Nobody said “given what you told me about your kids and your trip, here’s the one that’s actually right, and here’s why the other three aren’t.”
That’s the part AI can’t do. It can describe. It can’t decide. And a trip with ten days on the clock doesn’t need more descriptions, it needs someone to make the call.

AI is a genuinely useful research tool. It’s fast, and it’s a fine starting point if you’re daydreaming about where to go next. But planning a real trip, one where every day is accounted for, every reservation actually holds, and every detail reflects your specific family, still takes a person who has been there, knows the players, and picks up the phone when something needs fixing at 9pm on a Tuesday.
We’ll happily use AI to help brainstorm. We won’t hand your trip over to it.
Want a trip built by someone who’s actually stood in that Roman piazza? Start here and let’s build something an algorithm never could.