Apps & saving placesBy ArrivLast checked
How to use ChatGPT to plan a trip
Ask for a day-by-day itinerary with named venues, then verify each place and put the list on a map. Give it your trip length, your pace, and the season, because a vague prompt returns a vague list. The step most people skip is turning that text into addresses you can navigate to and share. Check hours and travel times before you book.
Prompt with constraints
Trip length, who is coming, how much walking you want, the season, and one or two things you already know you want all change the answer. "5 days in Lisbon, two of us, slow mornings, one day trip, no museums before 11" returns something you can use, while "Plan Lisbon" returns a list you could have guessed yourself.
Ask for venue names and neighborhoods, in those words, because if you skip that you get categories: a traditional tavern, a nice lunch spot, a viewpoint. Those hand the research straight back to you, so the check is worth running: Montreal's planning page or Seoul's will tell you whether what came back is real.
Get it off the page
A text itinerary cannot tell you that two of its Tuesday stops are forty minutes apart, and it cannot be shared with six people in a form that stays current. Paste the list into a mapping tool and the geography problems surface immediately, which is the same move as organizing any itinerary by day. Only the source of the list has changed, so the check is identical.
A prompt that produces something checkable
"Four days in Montreal in October, two of us, slow mornings, one day trip to a nearby town, no museums before 11. Named restaurants and neighborhoods, walking time between consecutive stops, and flag anything that needs a timed ticket." That prompt is enough, whereas "Plan Montreal" is a coin flip.
If you get categories back, ask again and insist on names, and then map them. A model that puts Old Montreal, the Plateau, and a market in the same morning has never done that walk, so the map will tell you in seconds.
When the hours are wrong
ChatGPT will put a museum on its closed day and a dinner at a place that takes no walk-ins, because neither fact reaches it in a form it can trust. Treat every timed thing as unverified until you have opened the venue's own page. The expensive version of this is booking straight out of the chat: a table at a restaurant that moved last spring, or a hike that needs a permit nobody mentioned.
Keep the chat as research, and keep the trip in a map with addresses on it. Whether AI can plan a trip is the longer version of this warning, while this page is only the procedure.
One trick pays for itself: ask for walking time between consecutive stops, because models will happily produce those numbers and the wrong ones are easy to spot. A ten-minute walk that is really forty shows up as soon as you check two of them against a map.
Geography problems only appear on a map, because two Tuesday stops forty minutes apart read as perfectly reasonable in a paragraph. That is why the paste step is the part of this you should never skip.
- Prompt with length, pace, season, and named places, because a vague ask returns a vague list.
- Map the list before you book anything, since the gaps between stops only show up there.
- Check hours and timed tickets on the venue's own page, because that is the only source that updates.
Step by step
- 01
Give it the constraints
Length, group, pace, season, and any fixed plans, because vague prompts return generic itineraries.
- 02
Ask for named places
Specific venues and neighborhoods, so the result is checkable against a maps search.
- 03
Verify the bookable parts
Opening days, timed tickets, and whether each place still exists, which is a ten-minute job.
- 04
Map the list
Geocode each place, and then look at the gaps between consecutive stops.