AI & tools
Can AI plan a trip for you?
AI is reliable for the first draft and unreliable on details. It will produce a sensible shape for a week in Japan in seconds, and it will also invent a restaurant, underestimate a mountain drive, or schedule a museum on the day it closes. Use the output as a starting itinerary to verify, and check anything you plan to book.
What language models are good at here is structure. Knowing that Kyoto splits naturally into an eastern temple day and a western Arashiyama day, or that four days in Iceland means the Golden Circle and the South Coast, is pattern knowledge, and models have plenty of it. That saves the blank-page hour that stops most people from planning at all.
What they get wrong is anything specific and current. Opening hours, seasonal closures, whether a permit system exists this year, how long a drive takes on a mountain road, and whether a named restaurant is still open. A model producing text has no way to check any of it, and a confident wrong answer looks identical to a right one.
The practical fix is grounding. A planner that looks each suggested place up in real venue data can tell you the address exists, and can drop it on a map where a two-hour gap between stops becomes visible. Arriv builds plans from geocoded venues for that reason, and shows you what it found before anything is saved, so a bad match is one tap to replace.
Verify three things before booking: that the place exists and is open, that the travel time between consecutive stops is what the plan assumes, and that anything requiring a timed ticket is available on your date.
Do it in Arriv
Last reviewed