The Anatomy of a Perfect Prompt

The 6 Building Blocks Every Power User Knows

After analyzing thousands of prompts that consistently produce great output, a clear pattern emerges. The best prompts aren't longer — they're more structured. They contain the same 6 components arranged in a way that gives the AI everything it needs to do excellent work.

I call this the CRISP-F framework: Context, Role, Instructions, Specifics, Persona (output), and Format.

Component 1: Context — The "Why" Behind Your Request

Context is the background information the AI needs to understand your situation. Without it, every answer is generic.

WITHOUT CONTEXT:
Write a product description for headphones.

WITH CONTEXT:
I'm launching noise-cancelling headphones on Amazon, competing
against Sony WH-1000XM5 and Bose QC Ultra. My product's edge:
40-hour battery (vs Sony's 30) and 20% lighter weight.
Target buyer: remote workers who take calls all day and
listen to music between meetings. Price point: $249.

Write a product description for these headphones.

Notice how context transforms the task from "describe generic headphones" to "write a competitive product listing with specific differentiators." The AI now knows your market position, target audience, and key selling points.

Component 2: Role — Activate the Right "Expert"

Assigning a role primes the AI to draw on a specific knowledge domain and writing style. It's the difference between asking a random person and asking a specialist.

WITHOUT ROLE:
Review this Python code for issues.

WITH ROLE:
You are a senior Python developer with 15 years of experience
who specializes in code review at a FAANG company. You're known
for catching subtle bugs, security vulnerabilities, and
performance issues that junior reviewers miss.

Review this Python code for issues.

The role doesn't just change the tone — it changes what the AI looks for. A "senior security engineer" will flag different things than a "Python performance specialist." Pick the role that matches what you actually need.

Roles that work exceptionally well:

• "You are a senior [X] engineer at [well-known company]" — activates high-standard professional thinking
• "You are an expert [X] with [N] years of experience" — pulls from deeper domain knowledge
• "You are a [role] who is known for [specific trait]" — shapes the style (e.g., "known for explaining complex topics simply")

Component 3: Instructions — The Actual Task

This is what you're asking the AI to do. The key principle: be imperative and unambiguous. Don't hint — command.

WEAK INSTRUCTIONS:
It would be great if you could maybe help me
improve this email.

STRONG INSTRUCTIONS:
Rewrite this email to:
1. Cut the word count by 50%
2. Lead with the key ask in the first sentence
3. Remove all passive voice
4. Add a clear call-to-action with a specific deadline

Numbered lists work better than paragraphs for instructions. The AI can check each item like a checklist, reducing the chance it misses something.

Component 4: Specifics — Constraints and Boundaries

Specifics tell the AI what to include, what to exclude, and the boundaries it should operate within. This is where you prevent the "close but not quite right" problem.

SPECIFICS EXAMPLE:
- Length: 150-200 words (not a word more)
- Tone: professional but warm, like a friendly colleague (not corporate)
- DO NOT use: "synergy", "leverage", "innovative", "cutting-edge"
- MUST include: one specific customer pain point and one statistic
- Language: plain English, 8th-grade reading level
- DO NOT start with a question

Negative constraints ("do NOT") are just as important as positive ones. They prevent the AI from falling into its default patterns — which are often the generic, overused phrases you're trying to avoid.

Component 5: Persona (Output Voice)

This controls how the output sounds. It's different from Role (which controls expertise). Persona shapes the voice, style, and energy of the writing.

PERSONA EXAMPLES:
- "Write as if you're Paul Graham writing a blog post"
- "Match the tone of Stripe's documentation — clear, precise, no fluff"
- "Write like a tech journalist at The Verge — engaging but informative"
- "Explain this like a patient college professor who uses analogies"

If you have an existing piece of writing whose style you want to match, paste it in and say "Match the tone and style of this writing sample." That's more effective than describing the style in words.

Component 6: Format — The Shape of the Output

Define exactly how you want the response structured. This is the most underused component and often the highest-impact change you can make.

FORMAT EXAMPLES:

"Respond in this exact structure:
## Summary
[2-3 sentences]

## Key Findings
[Bullet list, max 5 items, each under 20 words]

## Recommendation
[1 paragraph, max 75 words]

## Confidence Level
[High/Medium/Low with one-sentence justification]"

"Return your answer as a JSON object with these fields:
- title (string, max 60 chars)
- summary (string, max 160 chars)
- tags (array of 3-5 strings)
- sentiment (positive|negative|neutral)"

Putting CRISP-F Together

Here's a complete prompt using all 6 components:

[ROLE] You are a senior content strategist who specializes
in B2B SaaS email marketing.

[CONTEXT] We sell a $49/month project management tool for
freelancers. Our free trial converts at 12% (industry avg
is 15%). We've identified that users who don't create their
first project within 24 hours almost never convert.

[INSTRUCTIONS] Write an onboarding email that:
1. Gets sent 2 hours after signup
2. Motivates the user to create their first project TODAY
3. Includes one specific use case they can start with

[SPECIFICS]
- Max 150 words
- Subject line included (max 45 chars)
- No corporate jargon
- DO NOT mention pricing or the trial ending
- Include exactly one CTA button text

[PERSONA] Write like a helpful friend who happens to be
a productivity nerd — enthusiastic but not pushy.

[FORMAT] Structure:
Subject: [subject line]
Preview text: [40 chars max]
Body: [email content]
CTA Button: [button text]

You don't need to label the sections with [BRACKETS] — that's just for learning. In practice, you can write it as flowing text. What matters is that all 6 components are present.

When to Use All 6 vs. a Subset

Not every prompt needs all 6 components. Here's when to use what:

• Quick tasks (summarize, translate, reformat): Instructions + Format is usually enough
• Creative work (writing, brainstorming): All 6 components for best results
• Technical tasks (code, analysis): Role + Context + Instructions + Format
• Conversations (brainstorming, Q&A): Context + Role, then iterate