DEV Community

yaya806798160-ai
yaya806798160-ai

Posted on

The 5-section prompt framework that finally made my LLM output "ship it" instead of "almost right"

#ai #promptengineering #llm #productivity

Most "prompt engineering tips" articles are useless. They say "be specific" without showing you HOW.

Here is the template I now use for every serious prompt. It has five sections, and each one exists because leaving it out causes a specific, predictable failure.

The 5-Section Structure 

[Role]
You are a {role} with {years} years of experience in {domain}.

[Context]
Working on {project}. Target audience: {audience}.
Goal: {goal}.

[Output Schema]
Return a JSON object with fields:
- {field1}: {description}
- {field2}: {description}

[Constraints]
- Length: {n} words max
- Tone: {tone}
- Must include: {items}
- Avoid: {forbidden}

[Examples]
Input: {example_input}
Output: {example_output}
Enter fullscreen mode Exit fullscreen mode

Real Example: Writing a Product Intro

Without the framework:

Write a product intro for my SaaS.

Output (GPT, no framework):

Welcome to AcmeCloud, the leading cloud management platform
that helps businesses streamline their operations. Our
cutting-edge solution offers a wide range of features
designed to meet your needs...
Enter fullscreen mode Exit fullscreen mode

Generic. Useless. Could be any SaaS.

With the framework:

[Role]
You are a B2B SaaS marketer with 10 years experience
writing for ops directors at 200-person companies.

[Context]
Product: AcmeCloud (Kubernetes cost optimization)
Audience: Ops directors at Series B SaaS companies
Goal: Generate qualified demo bookings

[Output Schema]
Return JSON with: headline (10 words max),
subheadline (20 words max), 3 bullet benefits
(15 words each), CTA button text (3 words)

[Constraints]
- Tone: confident but not salesy
- Must include: 1 specific number (cost savings %),
  1 specific time saving (hours/week)
- Avoid: "leverage", "synergy", "revolutionize"

[Examples]
Good headline: "Cut your K8s bill by 40% in 30 days"
Bad headline: "Unlock the power of cloud optimization"
Enter fullscreen mode Exit fullscreen mode

Output (GPT, with framework):

{
  "headline": "Cut your K8s bill by 40% in 30 days",
  "subheadline": "AcmeCloud auto-rightsizes your clusters. Ops teams save 12 hrs/week.",
  "bullets": [
    "Average customer saves $18k/year",
    "Setup takes 15 minutes, no code required",
    "SOC 2 compliant, no data leaves your VPC"
  ],
  "cta": "See your savings"
}
Enter fullscreen mode Exit fullscreen mode

Same model, same product, vastly different output.

Why This Works

  1. Role gives the model a "persona" to write from, which calibrates tone and vocabulary.
  2. Output Schema forces structured thinking — the model can't ramble when it has to fill fields.
  3. Constraints prevent common failure modes — banning "leverage" eliminates 80% of corporate-speak.
  4. Examples anchor the quality bar — one good example beats ten rules.

I Built a Tool That Does This Automatically

I got tired of writing this scaffolding manually, so I built PromptOpt.

Paste a rough idea. Get back a structured prompt in this exact format. Feed it to any LLM. Free tier includes 10 optimizations, no card required.

Try It Yourself

Pick a recent prompt you wrote. Add the 5 sections above. Re-run it on the same model. The output difference is usually dramatic.

What's your "always include" section that I missed?

Top comments (0)