The 5 Core Principles
- Be specific — vague inputs produce vague outputs
- Assign a role — "Act as a..." dramatically improves quality
- Provide context — the AI knows nothing about your situation
- Specify format — tell it exactly how you want the output
- Iterate — the first response is rarely the final answer
The Role Technique — Most Powerful Single Improvement
Assigning a role to the AI is the single highest-impact change you can make to any prompt. Compare these two prompts:
Without role: "Give me advice on negotiating my salary."
With role: "You are an experienced HR director who has hired for hundreds of roles across tech, finance, and marketing. Give me specific, tactical advice on negotiating my salary for a senior marketing manager role at a Series B startup."
The second prompt gets advice calibrated to a specific context from a specific perspective. The output quality improvement is dramatic.
The CRISP Framework for Consistent Results
- Context — Background information the AI needs. Your industry, audience, situation, constraints.
- Role — Who should the AI be? An expert copywriter? A senior developer? A career coach?
- Instruction — What exactly do you want it to do? Be precise and unambiguous.
- Specifics — Word count, tone, format, style, examples to follow or avoid.
- Purpose — Why are you creating this? Who is the audience? What action should it drive?
Advanced Techniques
Chain-of-thought prompting: Add "Think through this step by step before giving your answer" to any reasoning task. This forces the AI to work through the problem rather than pattern-matching to a quick answer, significantly improving accuracy.
Few-shot examples: Show the AI 2-3 examples of what you want before asking it to produce something. Example: "Here are three product descriptions I love: [examples]. Now write one in the same style for [product]."
Constraint setting: Limits often improve quality. "Write this in exactly 150 words" or "Use only data from 2024 or later" gives the AI a tighter target to hit.
Persona adoption: Ask the AI to take on a specific character's perspective. "How would a sceptical CFO respond to this business proposal?" produces more useful critical feedback than "What are the weaknesses of this proposal?"
Prompts for Different Purposes
For editing your own writing:
"Edit the following text. Improve clarity and flow. Reduce word count by 20% without losing meaning. Keep my voice — don't make it sound more formal. [paste text]"
For brainstorming:
"Give me 20 unconventional marketing ideas for a local bakery trying to reach customers aged 18-30. Rank them by estimated cost to implement. Be genuinely creative — avoid clichés like 'social media contest.'"
For research summaries:
"Summarise the key arguments for and against [topic] in a balanced way. Present five points on each side. Cite specific facts or figures where possible. End with the strongest argument on each side."
For analysis:
"Analyse the following business situation and identify: (1) the three core problems, (2) the root cause of each, (3) a practical solution for each. Be direct and specific. [situation description]"
The Iteration Mindset
Professional prompt engineers rarely get perfect outputs on the first try. They treat prompting as a conversation — each response teaches you how to refine the next prompt. After any AI response, ask yourself:
- What did it get right that I want to keep?
- What did it miss that I need to add to the next prompt?
- Was the format right, or does it need restructuring?
A typical workflow for important outputs: 3-5 iterations, each refining a different dimension of the response.