Sometimes a helpful-sounding answer arrives before the situation is clear.
You ask for an idea for a weekend activity. The reply assumes a car, a full afternoon, good weather, and an enthusiasm for hiking you never mentioned.
For a low-stakes planning task, try asking for the questions first.
A small prompt
“Before suggesting anything, ask me up to three questions whose answers would materially change your recommendation. Use what I've already told you. If you have enough information, say so instead of inventing questions.”
The limit matters. You asked for help, not an application form.
Imagine planning an indoor afternoon of crafts using supplies you already own. Useful questions might concern how much time you have, what materials are available, and whether the activity needs to be quiet. Asking for your complete five-year creative vision probably won't help.
This is an illustrative example, not a recorded exchange or a measured result.
Check the questions, too
You can skip a question that feels irrelevant. You can also answer, “I haven't decided,” or set a boundary: “No purchases, no travel, and no more than an hour.”
Once you have supplied the missing details, ask for two options that fit. Have the assistant flag any assumption it still needs to make.
Read the suggestions against your actual constraints. If the “no purchases” activity somehow requires a new printer, it needs another pass.
Know when to move on
A question is useful when its answer changes what happens next. If another round would only make the plan longer, you can stop and choose.
Try this with something small and reversible. Leave private information out, and don't rely on a chatbot's questions to identify every risk in a consequential decision.
The aim is a better fit with fewer unsupported assumptions. If a direct answer or a conversation with a person would be easier, use that instead.
