The RCTF Framework
Every effective prompt has four components. Nail these and you'll consistently get better outputs from any AI model.
| Component | What It Does | Example |
|---|
| R — Role | Sets the AI's expertise and perspective | "You are a senior data analyst with 10 years of experience in retail analytics." |
| C — Context | Provides background the AI needs | "Our company sells organic skincare products online. Revenue grew 30% last year but customer retention dropped to 62%." |
| T — Task | Defines exactly what you want | "Analyse the attached customer data and identify the top 3 drivers of churn." |
| F — Format | Specifies how the output should look | "Present findings as a table with columns: Driver, Evidence, Recommended Action, Expected Impact." |
5 Core Techniques
1. Zero-Shot Prompting
Ask directly without examples. Best for straightforward tasks where the AI's training is sufficient.
Zero-Shot
Classify the following customer review as Positive, Negative, or Neutral. Return only the classification.
Review: "The product arrived on time but the packaging was damaged. The item itself works fine."
2. Few-Shot Prompting
Provide examples so the AI learns your desired pattern. Essential for custom formats or nuanced tasks.
Few-Shot
Convert these feature descriptions into user-friendly benefit statements.
Feature: "256GB SSD storage"
Benefit: "Store thousands of photos and apps without ever worrying about running out of space."
Feature: "IP68 water resistance"
Benefit: "Take it to the pool, the beach, or out in the rain — your phone is fully protected."
Feature: "[YOUR FEATURE]"
Benefit:
3. Chain-of-Thought
Ask the AI to reason step-by-step. Dramatically improves accuracy on complex, multi-step problems.
Chain-of-Thought
A company's revenue was $2.4M in Q1 and $2.8M in Q2. Operating costs were 68% of revenue in Q1 and 71% in Q2. Marketing spend increased from $180K to $240K between quarters.
Think through this step-by-step:
1. Calculate the profit for each quarter
2. Determine the profit margin change
3. Assess whether the increased marketing spend was justified by the revenue growth
4. Provide your recommendation
4. System Prompts
Set persistent instructions that shape every response in a conversation. Think of it as configuring the AI's operating mode.
System Prompt
You are a UK-based employment law advisor. Always cite relevant UK legislation. Flag when advice differs between England/Wales and Scotland. Use plain English — avoid legal jargon where possible. When uncertain, clearly state the limitation and recommend consulting a solicitor.
5. Iterative Refinement
Build on outputs through follow-up prompts. The most underused technique — treat AI as a conversation, not a slot machine.
Iterative Refinement
Round 1: "Draft a project proposal for [X]."
Round 2: "Good structure. Now make the benefits section more specific — add estimated time savings in hours per week."
Round 3: "The tone is too formal for our team. Rewrite in a more conversational style while keeping the data points."
Round 4: "Add a section on risks and mitigations. Include at least 4 risks."
Output Formatting Guide
| Format Instruction | When to Use | Example Phrase |
|---|
| Bullet points | Lists, quick scanning | "Present as bullet points, max 8 items" |
| Numbered steps | Processes, sequences | "Provide step-by-step instructions, numbered 1-N" |
| Table | Comparisons, structured data | "Format as a table with columns: X, Y, Z" |
| JSON / CSV | Data for other tools | "Return as valid JSON with keys: name, score, reason" |
| Markdown | Documentation, reports | "Use markdown with H2 headings and code blocks" |
| Specific length | Summaries, social posts | "Keep to exactly 280 characters" or "Write 100-150 words" |
Common Mistakes to Avoid
- Vague instructions: "Write something about marketing" → Be specific about topic, audience, length, and tone.
- No role assignment: Skipping the role means the AI defaults to generic responses. Always set expertise level.
- Overloading a single prompt: Asking for 10 things at once leads to shallow outputs. Break complex tasks into steps.
- Ignoring format: If you don't specify format, you'll waste time reformatting. Tell the AI exactly what structure you need.
- One-and-done mindset: Treating AI like a search engine instead of a collaborator. The best results come from iteration.
- Forgetting context: The AI doesn't know your industry, audience, or constraints unless you tell it.
- Not reviewing outputs: AI can hallucinate facts. Always verify claims, data, and citations.
Remember: Prompt engineering is a skill, not a talent. The difference between a mediocre prompt and a great one is specificity, structure, and iteration. Practice the RCTF framework daily and you'll see dramatic improvement within a week.