Make It Ask You Questions
Invert the Direction
One of the most useful moves available is also the least used: instead of trying to write a complete prompt, hand over a rough one and say "before you answer, ask me up to five questions that would most improve your answer." What comes back is often uncomfortable, in a productive way. It asks who the audience is, what the budget is, what has already been tried, what would count as success — the exact things you left out, and frequently things you had not decided. Answering them takes two minutes and produces a far better result than any amount of solo prompt-polishing, because you were never going to guess which gaps mattered from the inside.
- Add: "Before answering, ask me up to five questions that would most improve your answer"
- The questions are a free audit of what your prompt was missing
- Cap the number, or you may get a long interview you did not want
- Best used at the start of anything long, expensive or hard to redo
Ask It to Restate the Task
A quicker variant for anything complicated: "Before you start, restate what you think I am asking for and list any assumptions you are making. Then wait." Misunderstandings become visible in ten seconds rather than after 1,500 words. It is remarkably common to read the restatement and realise the model has interpreted something reasonably but not the way you meant — usually a word that means something specific in your organisation and something general everywhere else. This is also a good habit for human work: if a task is worth a long answer, it is worth thirty seconds confirming the question. The assumptions list is often the more valuable half.
- "Restate the task and list your assumptions, then wait" catches misreads early
- Ambiguous internal jargon is the most common source of a wrong interpretation
- The assumptions list frequently surfaces decisions you had not actually made
- Cheap on anything where the full answer would be long or costly to redo
Interviews and Self-Critique
The same inversion works in other shapes. For anything where you hold the knowledge — a case study, a personal statement, a process nobody has documented — ask it to interview you one question at a time and then write from your answers. You get your own material, structured, instead of a generic piece that happens to be about your topic. And once you have a draft, "critique this against the brief, list the three weakest points, then rewrite only those" is a cheap quality upgrade. Note the limit: a model reviewing its own work catches structural and clarity problems well, but it is not a reliable check on whether the facts are true.
- "Interview me one question at a time, then write it from my answers" preserves your material
- "Critique this, list the three weakest points, then rewrite only those" is a cheap upgrade
- Self-critique is good at structure and clarity, weak at verifying facts
- One question at a time matters — a list of twelve questions gets skimmed
Try It Yourself
The highest-leverage habit in this module fits in one sentence: make the model interview you before it answers. Try it on something you genuinely need this week.
Pick a real task where the outcome depends on details only you know — a plan, a message, a decision. Use the prompt below and answer its questions honestly before letting it produce anything.
I want help with [your task]. Before you answer, ask me the three most useful questions — one at a time, waiting for my reply each time. Then give your best answer using what I told you.
- It asked at least one thing you had not thought to mention
- The final answer used your replies rather than generic filler
- The result beat what a direct one-shot request usually gets you
Prefer slides, quizzes, and saved progress? Read this lesson in the library — free, no sign-up.