My Secret AI Prompt Formula — The One I Actually Use

People ask me for my prompt library. They assume I have a vault somewhere — a thousand magic phrases I paste in to make the machine behave.

I don’t.

I have five moves.

I use them for email, strategy, hiring decisions, and rewriting a page that isn’t landing. Same five moves every time. They take about twenty extra seconds to type, and they are the difference between AI giving you something generic and giving you something you would actually use.

Here’s the thing nobody tells you when you start: a prompt is not just a question. A prompt is direction given to intelligence. If you hand a brilliant new hire a one-sentence assignment with no context, no audience, and no idea what “done” looks like, you should expect something generic. That is not a talent problem. It is a briefing problem.

AI does not reward fancy wording. It rewards clarity.

The formula is CRISP

C — Clear

Remove the ambiguity before you hit enter.

“Help me write better emails” is not clear. Better how? Shorter? Warmer? Less apologetic? To whom?

Vague in, generic out. Most bad AI results are not the model failing. They are the result of a person asking for “help” and getting exactly that: help, in general, for anyone.

R — Role-Specific

Tell it what perspective to use.

“Act as a business writing expert who coaches executives on hard conversations.”

“Act as a hiring manager who has screened a thousand resumes for this role.”

The point is not role-play for its own sake. The point is to define the lens the AI should use so it notices the right things.

I — Inclusive of Context

Ground it in your actual situation.

This is where most people underinvest. The AI does not know your industry, your boss, your last quarter, or the fact that this is the third time you have had to push the timeline.

Give it the mess.

“I’m a project manager. I have to tell my VP that a launch is slipping two weeks. It’s the second delay. He is already frustrated and he tends to want solutions before explanations.”

That paragraph is worth more than any clever phrase you could add later.

S — Specific About Output

Say what you want back.

A plan? A draft? Three options? A one-pager you can forward? A checklist?

If you do not say, the AI has to choose for you. Format is not decoration. Format is what turns information into action.

P — Prescriptive About Process

Tell it how you want the work approached.

“Lead with the solution, not the apology. Then the revised timeline. Then what I’m doing to prevent it next time. Keep it under 150 words. No corporate language.”

This is the step that separates people who get useful work from people who stay frustrated. You are not only asking for an outcome. You are specifying the path.

What this looks like in practice

Before:

Help me write an email about a project delay.

After:

Act as a business writing expert who coaches executives on hard conversations. I’m a project manager telling my VP that our launch is slipping two weeks — second delay, he is already frustrated, and he wants solutions before explanations. Write the email. Lead with the fix, then the revised timeline, then the prevention step. Under 150 words. Direct, no corporate softening, no apology paragraph.

The second prompt takes a little longer to type. But the difference is that the output is much closer to something you can actually use.

Nothing about the AI changed between those two prompts. The direction changed.

The move almost nobody makes

Make the AI ask you questions first.

Before it gives you anything personalized, tell it:

Start by asking me the questions you need answered about my situation, my constraints, and what I have already tried. Wait for my answers before you write anything.

Now it stops acting like a vending machine and starts acting more like a collaborator.

That is useful because the questions often expose the missing context before the AI starts confidently filling in the blanks.

Start messy

You do not have to get the prompt perfect on the first pass.

This is a conversation. Start rough, look at what comes back, then say what is wrong with it.

“Too formal.”

“You buried the ask.”

“Cut this in half.”

“You assumed something I never said.”

That refinement loop is not a failure of prompting. That is prompting.

Save what works

When a prompt finally gives you something great, save it.

Over time, you are not building a giant library of magic phrases. You are building a small set of instructions that work for the things you do again and again.

Five moves: Clear. Role. Context. Output. Process.

Ask questions first. Start messy. Save what works.

That is the whole secret. It was never a magic phrase.

The deeper lesson is that the best prompt keeps the decisions that need your judgment with you, while giving AI enough direction to do the production work well.

Use the Action Guide below to put CRISP to work on one real task today.