How to use AI to plan an itinerary properly

What to tell it

The difference between a generic list of top attractions and a genuinely useful plan is entirely in the constraints you give.

The seven things worth stating every time: exact dates and arrival and departure times, the city and where you are staying, the group size and anything relevant about it, the per-person budget, the pace you want, what you have already booked, and what the group has explicitly ruled out.

That last one is the most underused. Telling it that nobody wants to queue for two hours, that one person cannot walk more than 5km, or that the group has no interest in churches removes most of what would otherwise fill the plan.

• Exact dates, and the arrival and departure times
• Where you are staying, so the geography is anchored
• Group size and anything that constrains it
• Budget per person for the whole trip, not per day
• Pace: how many things per day you actually want
• What is already booked and fixed
• What the group will not do, stated explicitly

Ask for the right shape

Ask for time blocks, morning, afternoon, evening, rather than exact times. Exact times in a generated itinerary are false precision and produce a plan that feels broken by the second morning.

Ask for two or three items per day, not six. Generated itineraries are systematically overpacked because a longer list looks more valuable. A real day in a city holds one major activity, one meal that matters, and one or two smaller things.

Ask for the day grouped by neighbourhood. This is the single most useful instruction, because it forces the geographic logic that makes a day work.

Ask it to flag which items need booking in advance and how far ahead. This is genuinely useful and rarely volunteered.

Iterate, do not regenerate

The instinct when an itinerary is 70 percent right is to ask for a new one. That is the wrong move, because the new one will lose the good 70 percent and introduce different problems.

Instead correct specifically: keep days one and two, replace the day three museum with something outdoors, move dinner later because the group eats at 21:00, and remove anything more than 20 minutes from the centre.

Two or three rounds of specific corrections gets to something genuinely good. Five regenerations gets to five generic itineraries.

The verification pass

This is not optional and it is where the actual reliability comes from. Go through the plan and check every item that has an address, an opening time, or a price.

What to check, in order of how often it is wrong: opening days, particularly Monday and Tuesday museum closures. Seasonal closures, particularly ferries, mountain lifts, and anything coastal outside summer. Whether the venue still exists. Prices. And whether a restaurant can take your group size.

For a week-long trip this takes about twenty minutes and it is the difference between a plan and a document.

• Museum and gallery closing days, usually Monday or Tuesday
• Seasonal closures for ferries, lifts, and coastal businesses
• Whether named restaurants and bars actually exist
• Current ticket prices and whether entry is timed
• Group capacity, which almost always requires asking directly

Useful things to ask for that people do not

A rainy day version of each day. Takes one prompt and saves the trip when it rains.

The reverse route. Asking whether the day would be better done in the opposite order sometimes reveals that it would, because of light, crowds, or opening times.

What to skip if running late. Knowing in advance which item is the one to drop prevents a rushed decision on the street.

The one thing most visitors get wrong here. This tends to surface genuinely useful practical information, though like everything else it needs checking.

Turning it into a real plan

Once verified, the itinerary needs to live somewhere the group will open on the trip: shared, editable, and readable on a phone. A chat message or a wall of text in a note is not that.

Attach the bookings to the plan. Flight numbers, accommodation address, confirmation codes, and who has the tickets, so that the person who did not do the booking can find them.

And leave gaps deliberately. The best trips have unplanned afternoons, and a plan with no slack means that the first delay cascades through the whole day.