Step 1 · Before

Start with the human reality

AI cannot know why this matters, who may be affected or what must not be lost unless you bring that context.

Human leads

Do not write for the machine.

Explain the situation as you would to a capable thinking partner who knows nothing about your life or work.

What are you asking for right now? *Start with the ordinary request. It does not need to sound clever.

What are you trying to achieve? *

Why does this matter?

What does AI need to understand about the situation? *

Who is this for, or who may be affected?

What matters and must not be lost?

Step 2 · Protect

Set the guardrails

Useful support depends on clear limits. Protect people, information, trust and the decisions that must remain human.

Human protects

✓

Safety pause

Use fictional or generalised details. Never enter private, identifying, confidential or commercially sensitive information without permission and safeguards.

What constraints or boundaries apply?

What must remain your decision?

What will need checking before you use the answer?

Add these responsible-use instructions

Ask what is missingIdentify important missing information before answering.Label assumptionsDo not let confident wording hide a guess.Do not decide for meOffer support and options, not the final choice.Show what to verifyEnd with facts, sources or consequences to check.

Step 4 · Make useful

Describe a useful result

Tell AI what good support looks like. Clear outputs reduce friction and make the answer easier for you to judge.

AI produces

What would a useful result look like? *

Who should be able to understand or use it?

Preferred format

Tone or style

Length or level of detail

Step 5 · After

Review, copy and keep ownership

Use the complete prompt anywhere, or separate the reusable working agreement from this specific task.

Human owns

Where do these parts go?

If your AI tool has a custom or system-instructions area, use "Working agreement" there and "Task brief" in the chat. Otherwise, copy "Complete prompt" into an ordinary chat.

Human ownership check

Before you act on AI's answer, pause here. Fast is useful only when the result remains responsible and human.

I will verifyCheck important facts, sources and calculations.I will consider peopleNotice bias, trust, permission and possible consequences.I will adaptMake the output fit my voice, values and reality.I will decide and ownChoose what to accept, reject, change and do next.

Bring the human. Hand AI the hard thinking.