Let AI Ask You Questions Before It Answers

When you don't know what to specify, flip the roles: the prompt that makes AI interview you, its settings, and how to get out of an endless interrogation.

3 min read

Large reception hall of a building
Photo: MiracleMiles · CC BY 4.0, cropped

The president of a neighborhood association has to organize the September flea market. She opens an AI and writes: "Help me organize our flea market." She gets a twelve-step plan, perfectly generic: set the budget, choose a date, publicize it. She already knew all that. The AI didn't know how many vendors she was hoping for, whether city hall had already given permission, or who would be there to run the refreshment stand.

When you don't know what to specify, the simplest move is to turn the situation around: instead of guessing what needs to be said, you let the AI ask you the questions.

The basic prompt

Prompt
I need to organize the annual flea market for my neighborhood association.

Before suggesting anything, ask me the questions you need so that your plan fits my situation. Ask them one at a time, waiting for my answer between each. Five questions maximum.

When you have enough information, give the plan and list the assumptions you made on my behalf.

Three details matter. "One at a time" avoids the wall of ten questions that you only half answer. The cap of five prevents the interrogation. And the last sentence shows you what the AI assumed for lack of an answer, which is often as instructive as the plan itself.

The variants that help

Offer a default answer. If answering tires you out, add: "For each question, offer your most likely answer for me to simply confirm." You answer "yes" or you correct it.

Skip what you've already said. When you've pasted a text or a description beforehand, specify: "Don't ask any question whose answer is already above." Without that sentence, AIs sometimes ask again for what you just wrote.

Ask only for the questions. Sometimes you'd rather answer at your own pace, in a document. "Don't write anything. Just give me the list of questions to ask me, ranked from most important to least important." You answer them outside the conversation, then paste everything back.

Stop on your signal. "Keep asking questions until I say 'enough.'" Handy for a fuzzy project, as long as you don't forget to say it.

When it goes wrong

The AI asks questions that are too obvious ("what's your budget?") or too numerous: start over by framing it, "only ask the questions that would really change your plan, not the ones whose answer is obvious."

You don't know how to answer one of them. That's valuable information. Answer "I don't know, pick whatever is most common and tell me it was an assumption." If the AI makes a choice, reread it, because you're the one who will be responsible for the plan.

The AI chains questions without ever concluding. Write "you have enough information, move on to the answer." Setting a maximum from the start avoids this.

What it teaches you

The questions asked form a list of what an AI needs to know for your task. Write them down. Next time, you'll put them straight into your prompt, and you won't need the interview anymore. That's exactly what the article on the context to give AI describes, but learned on your own case.

For tasks you do often, keep the final version in your reusable templates with the answers as variables. The president in our example will use it next year, changing only the date.

To try

Pick a project that your head is full of but that you have nothing on paper for: a trip, a move, an event. Paste the basic prompt, replacing the first line, answer five questions, then read the assumptions at the end. You'll see right away what a one-line request couldn't contain.

Getting started

Imposing a Response Format on the AI: Lists, Tables, Length

Want a short answer and the AI delivers three pages? How to set the length, the structure and a table that pastes cleanly into a spreadsheet.

4 min read

Getting started

Improving a Prompt Through Iteration: The Step-by-Step Method

Four rounds of correction on a real case, how to phrase each piece of feedback, when to start over and how to lock in the final prompt for reuse.

4 min read

Getting started

Prompt glossary: the words you need to know

Token, hallucination, few-shot, system prompt, RAG, agent: 35 words from the AI vocabulary explained simply, with an example and the article to read next.

6 min read