Ask before you write

Why this matters: If the ask is fuzzy, AI still writes something that looks finished. It just guesses what you need.

The prompt (copy and paste)

Paste this into your AI chat. Put your own request at the end. Use a made-up or safe one.

Help me understand the request below before I do any work.
Use only what the request says. Don't fill in gaps.

Give me:
1. What we know for sure
2. What we don't know
3. Up to 5 questions I should ask the person (most important first)
4. A suggested thing to make, marked "suggestion"
5. What "finished" could look like, marked "to confirm"

Don't do the task yet. Don't make up facts, options or advice.

Request:
"Can you look into whether we should expand the partner program and put something together for next week's management meeting?"

Before and after

Before: "Write something about expanding the partner program for management."

The AI writes a full plan. It has made-up numbers, a made-up region and a firm "we recommend". None of it came from your boss. You spend Friday polishing guesses.

After: You use the prompt above. The AI gives you questions instead:

• What should management decide at the meeting?
• Do they want a short discussion paper or a full recommendation?
• One page, or a few slides?
• What's already fixed, like budget or timing?

You send two of those questions to your boss. Now you know what to make.

Check it (3 things)

1. Known vs guessed. Can you tell what the request said from what the AI assumed?
2. Useful questions. Would the answers change the work? If not, ask for better ones.
3. Still a suggestion. Is the plan marked "suggestion" or "to confirm"? It isn't agreed until a person says yes.

If the AI starts doing the task anyway, run it again. Add: "Stop. Only ask questions."

Do this now

Find one fuzzy request from this week. Remove anything private first (see Think before you paste). Run the prompt. Send your boss the single most useful question it gave you.