Prompt Engineering with OpenAI ChatGPT

Prompt Engineering: A Simple Guide to Better Results

Many people use AI with vague prompts and hope for the best. Something like: "Write a postmortem update about yesterday's outage." often produces inconsistent results. Instead of playing Prompt Roulette, use a structured approach.

A Better Workflow

1. Contract – Define what success looks like. Ask: What must be included?

2. Scaffold – Use a reusable template. For example:

• Summary
• Impact
• Timeline
• Known Facts

3. Levers – Change only one thing at a time:

• Add more detail
• Shorten the response
• Change the tone
• Adjust the audience

4. Loop – Review and improve:

Generate → Review → Find one issue → Improve the prompt → Run again.

Think of a Prompt as a Contract

A strong prompt usually contains five parts:

• Instructions – What should AI do?
• Context – Facts and background information.
• Constraints – Rules and limitations.
• Examples – Optional style or format samples.
• Output Contract – Exact structure of the response.

Use Roles and Output Contracts

Instead of saying: "Analyze these tickets and recommend what to do."

Provide more context:

• Role: Support Operations Analyst
• Audience: Product Manager

Then define the output structure:

• Summary
• Key Themes
• Recommendations
• Next Steps
• Unknowns

Ground Responses with Evidence

To avoid unsupported claims:

• Use only the provided source.
• No guessing.
• Label claims as Evidence, Assumption, or Unknown.

If a claim cannot be traced back to the source, treat it as an assumption or unknown.

Debug the Prompt, Not the Output

When AI-generated code has issues, improve the prompt rather than immediately editing the code.

Use this cycle: Draft → Evaluate → Diagnose → Fix Prompt → Retest

Use Behavior Knobs

Control the response by adjusting:

• Detail: Short or Detailed
• Tone: Formal, Casual, Teaching
• Audience: Developer, Junior Developer, PM, Customer

For example:

• Developer → Code only
• Junior Developer → Code + Explanation
• PM → Summary + Recommendations

Key Takeaway

The biggest lesson in prompt engineering is simple: Change the Output Contract, not the task.

Clear instructions, good context, strong constraints, and a defined output structure will consistently produce better results. Prompt engineering is not about magic tricks. It's about giving clear directions and improving them step by step.