Ask ChatGPT to plan a weekend trip for you and the result is almost always the same: a tidy list of famous sights, a couple of generic restaurants, and the vague sense that the itinerary could have been written for literally anyone. You scroll, correct half of it, and start again.
The answer wasn’t bad because the chatbot is bad. It was bad because your question was missing half the information it needed — budget, dates, how you like to travel — and the chatbot filled in the blanks with guesses instead of asking you. There is a small habit that fixes this, and it takes one sentence: tell the chatbot to ask you questions first.
The real problem is the gaps in your question, not the question itself
Most prompts are full of things you know and never say. “Plan a weekend trip” means something very different if you’re travelling with a toddler, a £200 budget, or a bad knee. A human travel agent would ask. A chatbot, by default, does not. It picks the most likely interpretation, answers confidently, and leaves you to spot the mismatch.
This isn’t a bug you need to work around with clever phrasing. Chatbots are trained to be helpful immediately, so guessing feels like good service to them — even when the guess is wrong. If you’ve ever wondered why an AI answer sounds confident while missing the point, that confident-but-wrong habit is worth understanding. Asking for questions first simply redirects the instinct: instead of guessing, the chatbot investigates.
The fix: make the chatbot interview you first
Before the chatbot answers, add one line to your request:
“Before you answer, ask me the clarifying questions you need to give a useful answer.”
That’s it. The wording isn’t magic — “ask me questions first”, “what do you need to know before you answer?” all work. What matters is that you’re explicitly giving the chatbot permission to do what it normally avoids: admit it doesn’t have enough information yet.
This works on ChatGPT, Claude and Gemini alike. None of them has a built-in “interview mode” switch for regular chats, but all three follow a direct instruction like this reliably. It also works in the free versions — there’s no paid feature hiding behind it.
What it looks like in practice
Take the trip-planning example. Instead of the generic itinerary, the chatbot comes back with something like: “What city are you travelling from? What’s your budget? Are you travelling with children? Do you prefer museums, outdoors, or food?” You answer in plain sentences — “London, about £300, just me and my partner, mostly food and walking” — and then the answer arrives, built around your actual situation.
The difference is most dramatic on requests with lots of hidden variables: trip plans, buying advice, job applications, lesson plans, business ideas, event planning. A vague request plus clarifying questions consistently beats a vague request alone, because the chatbot stops designing for the average person and starts designing for you.
A practical detail: you can limit the interrogation. “Ask me up to three questions, then answer” keeps things moving when you’re in a hurry. Without a limit, some chatbots get enthusiastic and fire off ten.
When this trick earns its keep — and when it’s pointless
It helps most when:
- The request is open-ended. “Help me redecorate my living room” has dozens of valid answers; “what’s the capital of Peru” has one.
- Your situation is unusual. The more your case differs from average, the more the chatbot’s default guesses miss.
- You’d accept corrections anyway. If you’d normally fix the answer in two or three follow-ups, answering three questions upfront gets you there faster.
It’s not worth it when:
- The question is simple and factual. Asking for questions before “what year was the Eiffel Tower built” just adds a pointless round trip.
- You’re using Deep Research mode. ChatGPT’s research mode already asks clarifying questions automatically before it starts — that research mode is worth knowing about — so adding the instruction there is redundant.
- You already gave all the context. If your prompt is long and specific, there may be nothing left to ask.
One workflow tweak, if the back-and-forth annoys you
Some people find answering questions in the chat messy, especially on long projects: your answers scatter down the conversation and later replies can lose track of them. There’s a cleaner alternative — let the chatbot ask its questions, then go back and edit your original message to include the answers, and resubmit. The whole conversation then proceeds from one complete, well-formed request.
And if you find yourself typing “ask me clarifying questions first” every single day, you can stop: put it in ChatGPT’s custom instructions instead, so it becomes the default behaviour for every new chat. Here’s how custom instructions work, with examples worth copying.
The broader lesson is a simple one. Chatbots are strongest not as answer machines but as collaborators — and collaboration starts with questions. The next time an answer comes back suspiciously generic, resist the urge to rewrite your prompt from scratch. Just add the sentence, answer what it asks, and watch the second answer do what the first one only pretended to.
