When AI Keeps Missing the Point, Ask It to Interview You First
A simple questions-first prompt can help you turn an unfinished idea into a clearer draft without pretending you already know every detail.
Sometimes the hardest part of using AI is not writing the final document. It is figuring out what you actually need to tell the tool. You may know the situation, the people, and the desired result, but those details are still in your head. A short request such as “write a notice for me” forces the AI to guess. A better approach is to ask the AI to interview you before it drafts.
Why questions can improve the first useful draft
Official prompting guidance from OpenAI recommends clear, specific instructions with enough context, followed by review and refinement. Anthropic’s guidance makes a similar point: explain the desired output, its constraints, and the reason behind important instructions. You do not have to organize all of that information perfectly on your own. You can make gathering the information part of the prompt.
Copy this questions-first prompt
Try this: “I need help creating [type of document or plan]. Before you draft it, ask me up to five short questions to learn the audience, goal, essential facts, timing, and preferred tone. Ask one question at a time. Do not invent missing details. After I answer, summarize what you understood and ask me to correct anything important. Then create the draft in [desired format].” Replace the bracketed parts with your task. The question limit keeps the conversation focused, while the summary gives you a checkpoint before the AI starts writing.
A practical example
Imagine a hotel manager needs a notice about a planned water interruption. “Write a water notice” leaves many gaps. Who will receive it? Which areas are affected? When will the interruption begin and end? Is the time confirmed? What should guests do if service does not return? In a questions-first conversation, the AI might ask for those facts before drafting. The manager can answer that the interruption is scheduled from 10:00 a.m. to noon, affects one building, and should be explained in a calm, apologetic tone. This is a fictional example, but it shows why the questions matter: the quality of the notice depends on facts the AI could not safely assume.
Add two guardrails
First, tell the AI not to invent facts. It may still make a mistake, so check names, dates, times, prices, policies, and instructions yourself. Second, avoid pasting information that the task does not require. A maintenance notice normally does not need guest names, payment details, identification numbers, or private account information. Use placeholders when possible, such as “[property name]” or “[contact number],” and add the final details after you have reviewed the draft.
Know when this method is worth using
The interview approach is most useful when the task has several decisions hiding inside it: a customer update, a training outline, an event plan, a procedure, a difficult email, or a comparison of options. It is unnecessary for a simple request such as correcting one sentence. It also should not replace professional judgment for medical, legal, financial, safety, or policy decisions. In those cases, use AI to help organize questions or create a draft for qualified review—not to make the decision for you.
Try one small experiment
Choose a document you have been postponing because the details feel scattered. Paste only the questions-first prompt, name the type of document, and let the AI ask its first question. Answer briefly. When it summarizes the situation, correct at least one detail or add one constraint before asking for the draft. That correction step is not a failure; it is the normal process of turning a vague idea into clear instructions.
Sources and further reading
Interested in patient, practical AI training? Contact Brent.