The CRISP AI Prompt Formula: Turn Questions Into Useful Briefs

You ask AI to write a client email. It comes back polished, pleasant, and somehow still needs twenty minutes of repairs.

The sentences aren't the problem. The draft solves a different problem from yours.

People ask me for my prompt library, as if there's a vault of magic phrases that makes the machine behave. I don't have a vault. I use five moves.

A prompt is simply the instruction you give an AI tool such as ChatGPT. The useful ones work like a brief, not just a question.

Here's the shift: every important detail you leave out becomes a decision the AI might make for you. That includes your audience, your priorities, and sometimes promises you never intended to make. Fluent writing is not permission to negotiate on your behalf.

CRISP helps you keep those decisions where they belong.

The CRISP AI prompt formula

CRISP stands for Clear, Role-Specific, Inclusive of Context, Specific About Output, and Prescriptive About Process. Each part removes a different kind of guesswork.

I use these moves for emails, strategy drafts, hiring discussions, board memos, and landing-page rewrites. They shape the work; they don't replace your judgment.

You don't need five elaborate paragraphs every time. You need enough direction that the AI doesn't quietly choose the assignment for you.

C — Clear: name the problem you're actually solving

My test is simple: could two capable people read your prompt and reasonably produce opposite results? If so, make the goal clearer.

Before:

Help me write better emails.

After:

I send about 15 client update emails a week. Clients keep asking me to clarify next steps because I bury the ask in the third paragraph. Help me build a reusable template that puts the ask in the first two lines.

“Better” could mean warmer, shorter, or more formal. None of those necessarily fixes a buried ask.

The revised prompt names a visible problem and a testable result. You can look at the draft and check whether the ask appears where you need it.

That's useful beyond AI. If you can't say what would make the result better, reviewing it becomes a matter of taste.

R — Role-Specific: choose what deserves attention

Once the goal is clear, tell the AI which perspective to use. Then explain what that perspective should notice.

Before:

Review my resignation letter.

After:

Review this resignation letter from an employment-law perspective, focusing on someone leaving on bad terms. Flag wording that could create problems later, unnecessary admissions of fault, and overly emotional passages. Identify questions I should take to an employment lawyer.

A role is a filter, not a costume. This prompt directs attention toward potential legal concerns rather than commas and a warmer sign-off.

It does not turn the AI into a lawyer with a thousand resignation letters behind it. Typing a résumé into a prompt doesn't make it true.

Use roles freely for low-stakes drafting and critique. When a mistake could be dangerous, use the output to prepare for qualified review—not replace it.

The same applies to medical advice. “Act as a doctor” can produce doctor-shaped language. That isn't a diagnosis you should rely on.

I — Inclusive of Context: explain what changes the advice

A useful perspective still needs your actual situation. Otherwise, the AI is advising someone who happens to have your job title.

Before:

Write an email telling my VP the launch is delayed.

After:

I'm a project manager. I need to tell my VP that our launch is slipping two weeks. It's the second delay, and he's already frustrated. He's made clear that he wants solutions before explanations. He'll ask why this keeps happening, so the email needs to address the pattern.

“Second delay” isn't background color. It changes the assignment from announcing a schedule change to addressing damaged confidence.

That's the kind of context worth including. The AI can't account for a history you haven't supplied.

Ask yourself: would this fact change a smart colleague's advice? If yes, include it. Relevant history, audience preferences, constraints, and previous attempts usually qualify.

That doesn't mean pasting every document you own. Unrelated detail can hide the important facts. And context isn't free when it contains confidential information. Follow your organization's rules and check what you're allowed to share with the tool.

S — Specific About Output: tell it where the answer must fit

Now that you've explained the situation, describe the thing you actually need back.

Before:

Give me ideas for improving our onboarding process.

After:

Give me exactly five onboarding improvement ideas, each as one sentence I can paste into Slack for my team. Rank them by how quickly we could implement them. None should require new headcount or budget approval.

The first request could produce thirty ideas, a long essay, or a framework with quadrants. The quadrants haven't done anything wrong. They just weren't invited to the Slack conversation.

The revised version specifies the destination, length, ranking, and limits. You get something shaped for a decision, not another formatting assignment.

Format isn't just presentation. It determines what you can do with the answer. If you know it belongs in a one-page memo or a table in a deck, say so.

P — Prescriptive About Process: supply the judgment you've already made

Output instructions describe the finished item. Process instructions tell the AI how to approach it: what comes first, what to avoid, and when to stop for answers.

Before:

Write a message to the team about the layoffs.

After:

Use only the approved facts I provide. Lead with the decision and headcount number in the first sentence. Explain the business reason in two plain sentences. Then explain what happens next for affected people, including dates. Don't use “difficult decision” or “not taken lightly.” Keep it under 200 words and end with where questions can go. Ask me for any missing facts before drafting.

This gives the draft a deliberate sequence instead of leaving the AI to choose one. It also blocks familiar corporate padding that says very little.

Notice the instruction to ask for missing facts. Without it, a request for dates and next steps could invite plausible details you never approved.

For layoffs, legal language, or other high-stakes messages, the final draft still needs the appropriate human review. Better instructions don't remove that responsibility.

Put all five together—and look for the missing decision

Let's return to the delayed launch.

Before:

Help me write an email about a project delay.

After:

Act as a business writing coach helping an executive handle a hard conversation. I'm a project manager telling my VP that our launch is slipping two weeks. This is the second delay. He's frustrated and wants solutions before explanations. Write an email under 150 words. Lead with the fix, then the revised timeline, then the prevention step. Keep it direct, without corporate softening or an apology paragraph. Before drafting, ask me for the confirmed fix, revised launch date, and prevention step. Don't invent them.

The last two sentences matter more than they might appear.

“Lead with the fix” sounds like clear direction. But if you haven't supplied a fix, you've asked the AI to make an operational decision inside a writing assignment.

A writing prompt can quietly become a decision-making prompt. That's the hidden handoff to watch for. The same thing happens when you request a client apology with compensation but never approve the compensation.

CRISP doesn't make the underlying AI smarter. It makes your instructions harder to misread—and makes missing decisions easier to spot.

When to ask for questions instead of a draft

Sometimes you can't write a complete brief yet. Use the conversation to build one.

Before:

Write a pricing page for my new product.

After:

Before writing anything, ask me what you need to know about my product, customer, competitors, and pricing model. Focus on missing details that would change the page's claims, offer, or structure. Wait for my answers before drafting. Don't invent prices, features, or customer evidence.

Use this approach when an unknown could change a recommendation or commitment. Pricing, positioning, and legal language deserve more than a confident first guess.

The questions may be the most useful output. They can expose a decision you thought was settled but never actually made.

Don't assume the AI will ask everything that matters, though. You're still checking the brief, not just answering an interview.

Correct the draft, then judge the rescue work

You don't need a perfect first prompt. Once you see the result, tell the AI exactly what missed:

• “Too formal. Make it sound like a conversation with a colleague.”
• “You buried the ask. Move it into the first two lines.”
• “Cut this in half, but keep the deadline and next step.”
• “You assumed the client was unhappy. I never said that. Remove that assumption.”

Those corrections aren't a failure of prompting. They're part of it. But don't let endless revisions become their own afternoon activity.

Before you use the result, check four things:

1. Did it follow the brief? Check length, format, sequence, tone, and phrases you ruled out. Don't give it credit for instructions it ignored.
2. Is it supported by reality? Verify names, dates, numbers, policies, and commitments. Unsupported specifics can read just as smoothly as confirmed facts.
3. Can the intended person act? Your VP should know the plan. Your client should know the next step. Sounding thoughtful isn't enough.
4. How much rescue work remains? A useful draft needs adjustments. A poor fit needs rebuilding. Count the work, not the compliments you'd give the prose.

If repairs take longer than writing it yourself, something needs to change. The brief may be incomplete, or the tool may not suit the task.

CRISP gives you a way to diagnose the problem. It isn't a guarantee that every task belongs with AI.

Action Guide: use CRISP on one real task today

Choose something already waiting on your desk: a client delay email, hiring rejection, weekly update, or pricing objection response.

Use the copyable prompt below to supply the goal, perspective, context, output, and process. Leave a field marked “unknown” rather than filling it with a guess.

Then review the draft against the four checks above. Make one specific correction and see whether the next version needs less rescue work.

When a prompt works, save the actual text. Mark the audience, dates, and commitments that must change next time. Otherwise, yesterday's helpful context becomes today's wrong fact.

You don't need a thousand prompts. Start with five or six reusable briefs for work you repeat, then expand only where they earn their keep.

Your next move: take one vague request you've used before and name the decision you left unstated. Fix that gap before you press Enter.

Help me complete the task below. Treat anything marked “unknown” as missing information, not permission to guess.
CLEAR — The problem I need to solve: [Describe the actual problem.]
A useful result would: [Describe what should change or what the reader should do.]
ROLE-SPECIFIC — Use this perspective: [For example, a business writing coach.]
Focus on: [What this perspective should notice or prioritize.]
INCLUSIVE OF CONTEXT — Here are the relevant facts: [Audience, history, constraints, previous attempts, and confirmed decisions.]
Important unknowns: [List them, or write “none known.”]
SPECIFIC ABOUT OUTPUT — Produce: [An email, Slack message, one-page brief, table, or other format.]
Length and destination: [Specify both.]
Practical limits: [Budget, staffing, approved commitments, or other constraints.]
PRESCRIPTIVE ABOUT PROCESS — Approach it this way: [Sequence, tone, and phrases to avoid.]
Before drafting, ask focused questions if missing information could change the recommendation, a commitment, or the format. Ask one at a time and wait for each answer before the next. Don't invent facts, dates, numbers, policies, or promises.
If no important information is missing, produce the draft. Put any remaining uncertainties outside the draft so I can review them.

Act as a business writing coach helping an executive handle a hard conversation.

I'm a project manager telling my VP that our launch is slipping two weeks. This is the second delay. He's frustrated and wants solutions before explanations.
Write an email under 150 words. Lead with the fix, then the revised timeline, then the prevention step. Keep it direct, without corporate softening or an apology paragraph.
Before drafting, ask me for the confirmed fix, revised launch date, and prevention step. Wait for my answers. Don't invent them or make commitments I haven't approved.

Before writing anything, ask me what you need to know about my product, customer, competitors, and pricing model. Focus on missing details that would change the page's claims, offer, or structure. Wait for my answers before drafting the pricing page. Don't invent prices, features, or customer evidence.