Using ChatGPT to plan a trip: where it stops
Ask any general-purpose chatbot for a ten-day itinerary and you will get one, fast, and it will be broadly sensible. That is genuinely useful, and worth doing. But most people hit the same wall a few days later, and it is worth knowing where it is before you build a trip on top of it.
What it does well
- The blank page. Turning "somewhere warm in May, about two weeks" into a concrete starting point is the hardest part of planning, and a model is very good at it.
- Local knowledge at breadth. It knows that Óbidos is small enough for an afternoon and that Lisbon deserves more days than people give it.
- Rewriting on request. Ask for a cheaper version and you get one.
Where it stops
1. The plan is a message, not a state.
The itinerary exists as text in a conversation. Change one thing and you get a new itinerary — a fresh message, regenerated. The bits you had already decided come back subtly different: the hotel you liked has changed neighbourhood, the restaurant you picked has vanished, the order is different.
This is the real problem. Not that it cannot change the plan, but that it cannot change part of it.
2. There is no record of what you rejected.
You ruled out Seville on day three. On day nine you cannot remember whether you ruled it out or never considered it. The conversation holds the answer somewhere, under everything since.
3. You cannot edit it yourself.
If one line is wrong, your only tool is to describe the correction in prose and hope. There is no field to fix.
4. Nothing is structured.
You cannot total the costs, see the stops on a map, or check whether Tuesday has anything in it, because it is prose. It looks like a plan and behaves like an essay.
5. It will state things confidently that are not true.
Opening hours, prices and whether somewhere still exists are exactly the facts a model is most likely to get wrong and least likely to flag. A plan is a document you turn up at a door with.
What to do instead
Use a model for the first draft — genuinely, it is the best tool for that. Then move it somewhere that holds structure: something where a place is a thing with a status and a cost, not a sentence in a paragraph.
The test for whether a planning tool is worth using is simple. Change one thing and see what happens to everything else. If the answer is "it gets regenerated", you will lose work every time you change your mind. And you will change your mind constantly, because that is what planning a trip is.
Doing this in heytripp
heytripp is built around that test. Tripp proposes changes to a board that already exists — add a town, drop a hotel, rework a day — and everything you have not touched stays exactly as it was. Ask for something destructive, like deleting a stop or rewriting your dates, and it shows you what will change before it does it.
Every field is editable by hand, so when the assistant gets something slightly wrong you fix it in a form rather than arguing with it. And when Tripp needs a fact it does not know, it looks it up rather than guessing.