Prompt Design > Prompt Engineering

"Prompt engineering" implies there's a right answer — a magic incantation that unlocks the AI. There isn't.

Prompt design is about building repeatable systems that produce consistently good output. It's design thinking applied to AI interaction.

Here's the framework.

The RICE Framework for Prompts

Every effective prompt has four components:

R — Role

Tell the AI who it is. This sets the knowledge domain, vocabulary, and perspective.

Weak: "Help me write an email."
Strong: "You are a senior account manager at a B2B SaaS company. You're known for concise, friendly communication that respects the client's time."

I — Input

Give it everything it needs. Context, examples, constraints, data.

Weak: "Here's my product, write about it."
Strong: "Here's our product description, our ICP (ideal customer profile), our competitor's positioning, and 3 examples of emails that got responses."

C — Constraints

Tell it what NOT to do. Boundaries improve output.

Weak: (no constraints)
Strong: "Do not use jargon. Keep under 150 words. Don't start with 'I hope this email finds you well.' Don't use exclamation marks."

E — Expected Output

Show it the format you want. Be explicit.

Weak: "Send me the result."
Strong: "Format your response as:

  • Subject line (under 50 characters)
  • Email body (under 150 words)
  • Follow-up note (one sentence)"

Putting It Together

Here's a real prompt using RICE:

ROLE: You are a senior content strategist at a top-tier marketing agency.
You specialize in B2B SaaS content that drives signups, not just traffic.

INPUT:
- Our product: [description]
- Target audience: [ICP]
- Competitor content: [top 3 links]
- Our best-performing post: [link + metrics]

CONSTRAINTS:
- No listicles
- No "ultimate guide" titles
- Every piece must have a clear CTA
- Aim for 1,200-1,500 words
- Use data and examples, not opinions

EXPECTED OUTPUT:
For each content idea, provide:
1. Title (under 70 characters)
2. One-line hook
3. Key argument in 2 sentences
4. Target keyword
5. Why this will outperform competitor content

Testing Your Prompts

Good prompt design includes testing:

  1. Run it 3 times. If the outputs vary wildly, your prompt is too vague.
  2. Change one variable. Swap the role or add a constraint. See what improves.
  3. Test edge cases. What happens with unusual input? Does it break gracefully?
  4. Get a second opinion. Show someone else the output without context. Does it make sense?

Common Patterns That Work

The "Before and After"
Show the AI a bad example and a good example. Ask it to make your input look like the good one.

The "Critique First"
Ask the AI to critique your draft before rewriting it. This produces significantly better rewrites.

The "Step-by-Step"
For complex tasks, break the prompt into steps. "First, analyze X. Then, based on your analysis, do Y."

The "Persona Switch"
Ask the AI to respond from multiple perspectives. "How would a CFO react to this? How about a junior developer?"

The Takeaway

Stop looking for magic prompts. Start designing prompt systems.

A well-designed prompt:

  • Works consistently across different inputs
  • Produces output you can use with minimal editing
  • Can be taught to your team
  • Gets better as you refine it

That's design, not engineering.

Part of the FOMA methodology series. These aren't tips — they're frameworks.