Most weak AI output starts before the prompt. The person asking has skipped the foundation underneath the task.

AI can help draft, organize, explain, compare, plan, and build. But it does not know your standards, your context, your audience, or the process behind the result you want. You supply those. When those pieces are fuzzy, the output gets fuzzy too.

Set a timer for ten minutes and answer these questions before writing the prompt.

1. What process are you asking AI to support?

Name the work in plain language. Do not write “help me with marketing” if the real task is “turn three customer questions into a short email sequence.”

Ask:

  • What is the actual job?
  • What does a good result need to do?
  • Who is it for?
  • What decision will I make after seeing the output?

2. What do you already know?

AI needs your source material. That may be a policy, a resume, a sales call transcript, a team standard, a customer complaint, or your own rough thinking.

Before prompting, gather the material you would hand to a smart assistant. If you would not trust a person to answer without that context, do not expect AI to do it well either.

3. What does good look like?

Write two or three quality standards before asking for output.

Examples:

  • “The answer should be short enough to send in an email.”
  • “The tone should be warm but direct.”
  • “Do not invent metrics, legal claims, or client results.”
  • “Flag uncertainty instead of smoothing it over.”

These standards help you judge the answer instead of merely reacting to whether it sounds polished.

4. What should AI not decide?

Good AI use protects human judgment. Name the boundaries.

AI can suggest language. You decide whether the promise is true. AI can organize a plan. You decide whether the priorities match the business. AI can explain a risk. You decide who needs to review it.

5. How will you check the result?

Every useful prompt needs a checking plan. That can be a source comparison, a second-pass critique, a checklist, or your own expert review.

Try this final instruction:

Before giving me the final answer, list the assumptions you are making and the parts I should verify.

That one sentence changes the relationship. You are no longer asking AI to sound done. You are asking it to help you think.

The self-check: If you cannot describe the task, source material, quality standard, boundaries, and review plan, the next step is not a better prompt. It is clearer thinking.