Ask AI to ask you useful questions before it starts

Example result.

Reader:      Tom Okafor, operations lead
Purpose:     choose one option for a small pilot
Facts:       calendar = less setup; booking page = captures more detail; neither chosen
Format:      half-page email, comparison then one decision question
Unresolved:  cost of either option

1. Describe what you know, then ask for questions

Write down the job, the reader if you know it, the facts you have and where the result will be used. An empty request invites a questionnaire about everything.

Help me prepare a brief for the task below. Before drafting, ask up to
three questions whose answers would change the result, and say briefly
why each matters. Don't ask me to repeat what I've given you. If
something isn't essential, suggest a sensible assumption and label it.

Task: [what I need to make]
Reader: [who, or "unknown"]
Facts and sources: [what I have]
Use: [where it will go]

Do not draft the result yet.

Fill in each bracket in a few words; “unknown” is a fine answer for the reader.

Why this works: the three-question limit keeps it short. “Why each matters” lets you judge whether a question is worth answering. A labelled assumption is easier to spot than one made silently.

What you should see: two or three questions, each with a one-line reason.

2. Answer the important ones, redirect the rest

Answer with facts. “We haven’t agreed a date” is a useful answer: it tells the AI to leave the date open. If it asks for confidential detail it doesn’t need, give a substitute; see what you can put into AI at work.

If a question is about a preference, redirect it. Weak: a reply asking “What font would you like? Should it have a logo? What tone? How long? Who’s the reader?” Only the last one changes what gets written.

Focus on purpose, reader and evidence. Formatting can use a labelled
assumption. The important open point is [missing fact]. Ask one
question about that, then restate the brief without inventing an answer.

[missing fact] is the gap that matters, such as “whether either option is chosen”.

Why this works: it tells the model which question earns its place, so the next reply is one question, not five.

What you should see: a single focused question, or a restated brief.

3. Confirm the brief, then ask for the draft

Restate the agreed brief in five short bullets: reader, purpose, facts
to use, format and unresolved information. Then draft [the result]
from that brief. Keep unresolved facts visible and don't turn proposed
dates or decisions into commitments.

[the result] is the thing you’re making, such as “the email”.

Read the five bullets before the draft. If a bullet is wrong, fix it there.

Why this works: a wrong brief is a two-word fix. A wrong draft is a rewrite.

What you should see: five bullets like the example result, then a draft that follows them.

Check the result

• The draft is written for the reader in the brief and answers their question.
• It uses only the facts you gave.
• Every unresolved point is still visibly unresolved.
• No labelled assumption has turned into a statement of fact.

The usual failure: the questions were good, but the draft fills a gap anyway. An option becomes “the choice”, or a date appears that nobody gave. Spot it by checking each sentence that sounds decided against the “Unresolved” bullet.

The draft says [statement], but the brief lists that as unresolved.
Rewrite the affected sentence to keep it open, and check the rest of
the draft for anything else that goes beyond the brief.

[statement] is the sentence that went too far.

Fictional example. You need a short note about two options for booking shared equipment.

Task:   short note about equipment booking options
Reader: unknown
Facts:  a shared calendar needs less setup; a booking page could capture
        more details (item, purpose, return date). Neither is chosen.
        No cost estimate yet.
Use:    email

The step 1 prompt brought back three questions:

1. Who will read this, and what do they need to do with it? A decision note and a staff announcement are different documents.
2. Has one option been chosen, or should the note compare them? This decides whether it recommends or announces.
3. Is there any cost or access information? Without it, cost has to stay open.

Your answers: Tom Okafor, the operations lead, needs to pick one for a small pilot; nothing is chosen; no cost yet. The restated brief is the example result.

• The failure. The draft’s last line read: “The booking page will launch next month.” Nothing in the brief says so.
• The repair. The repair prompt replaced it with: “Once you’ve chosen an option for the pilot, we can work out cost and timing.” It found nothing else beyond the brief.
• What the questions changed. Without question 1, the first draft would have been an announcement to staff about a system nobody has picked.

Choose the best equipment booking option and announce it.

Fictional practice source:
Fictional equipment-booking options: a calendar needs less setup; a booking page could capture more details. Neither option is approved. Costs are unknown. The operations lead needs to choose a small pilot.

Compare the calendar and booking page in a short table.

Fictional practice source:
Fictional equipment-booking options: a calendar needs less setup; a booking page could capture more details. Neither option is approved. Costs are unknown. The operations lead needs to choose a small pilot.

Before recommending, ask the operations lead which booking details are required, what approved setup/access is available, and who decides the pilot. Keep costs unknown and both options unapproved. Draft only; do not announce a launch.

Fictional practice source:
Fictional equipment-booking options: a calendar needs less setup; a booking page could capture more details. Neither option is approved. Costs are unknown. The operations lead needs to choose a small pilot.

Task kit for this job: Download (ZIP) · Reusable instructions

Go deeper