AI Prompting Study Guide
Your quick reference guide to all the key concepts, techniques, and frameworks for mastering AI prompts. Keep this handy whilst practising.
🧠 The Core Mindset
You're Programming, Not Chatting
- 1. Prompts are programs, not questions. Each word is a line of code guiding the AI's output.
- 2. LLMs are prediction engines. They calculate the most probable next words based on your input.
- 3. Your goal is to hack probability. Craft prompts that narrow the field of possible responses.
- 4. Adopt the "skill issue" mindset. A bad output means the prompt needs improvement, not that the AI is broken.
💡 Key Insight
"A prompt is a call to action to the large language model. It's not just a question, it's a program." — Dr. Jules White, Vanderbilt University
📋 The Four Pillars
These foundational techniques improve results by approximately 80%.
Pillar 1: Persona
Give your AI an identity
Assign a specific role, expertise, or perspective. This forces the AI to draw from a focused knowledge base rather than responding as a generic "nobody".
❌ Without Persona
"Write an apology email about a service outage."
✅ With Persona
"You are a senior site reliability engineer. Write an apology email about a service outage."
Pillar 2: Context
The cure for hallucination
Provide all relevant facts, figures, and constraints. LLMs will fill gaps with invented information if you don't provide sufficient context.
⚡ The #1 Hallucination Fix
Add this to your prompts: "If you cannot find the answer in the provided context, say 'I don't know'."
Key principle:
More Context = Fewer Hallucinations
Pillar 3: Output Requirements
Define your desired result
Specify exactly how you want the response formatted. Control format, length, style, tone, and structure.
Format Options
- • Bullet points
- • Numbered list
- • Table format
- • Prose paragraphs
- • JSON/structured data
Style Controls
- • Length (under 200 words)
- • Tone (professional, casual)
- • Headers and sections
- • Technical level
- • No corporate fluff
Pillar 4: Few-Shot Examples
Show, don't just tell
Provide concrete examples of the output you want. This reduces guesswork and dramatically improves quality.
Example Structure:
# Instruction:
Write a product description in this style:
# Example 1:
[Your first example of good output]
# Example 2:
[Your second example of good output]
# Now write for:
[Your actual product]
🚀 Advanced Techniques
🔗 Chain of Thought (CoT)
Force the AI to "think step by step" before providing a final answer. This increases accuracy and makes reasoning transparent.
Prompt addition:
"Think through this step by step before providing your final answer. Show your reasoning process."
🌳 Tree of Thoughts (ToT)
Explore multiple reasoning paths simultaneously, like branches on a tree. Useful when the first idea isn't always best.
Prompt structure:
- Brainstorm 3 different approaches to this problem
- Evaluate the pros and cons of each approach
- Identify the best elements from each
- Synthesise into a final "golden path" solution
⚔️ Adversarial Validation (Battle of the Bots)
Use competing AI personas to critique and refine outputs. AI is often better at editing than original writing.
Three-Round Process:
Draft Round
Two personas (e.g., Engineer + PR Manager) each write their version
Critique Round
A critic persona (e.g., Angry Customer) brutally critiques both drafts
Collaboration Round
Original personas collaborate to address the critique in a final version
📝 Prompt Templates
Basic Prompt Template
# PERSONA
You are a [specific role] with [years] years of experience in [domain].
# CONTEXT
[Provide all relevant background, facts, and constraints]
# TASK
[Clear description of what you want done]
# OUTPUT REQUIREMENTS
- Format: [bullets/table/prose]
- Length: [word count or section count]
- Tone: [professional/casual/technical]
- Structure: [headers/sections needed]
# IMPORTANT
If you cannot find the answer in the provided context, say "I don't know".
Chain of Thought Template
# PERSONA
You are a [role] who thinks systematically through complex problems.
# TASK
[Describe the problem or question]
# APPROACH
Think through this step by step:
1. First, identify [key factors]
2. Then, analyse [relationships]
3. Consider [alternatives or risks]
4. Finally, provide your conclusion
# OUTPUT
Show your reasoning process, then provide a clear final answer.
Quick Reference Card
✅ Always Do
- • Define a specific persona
- • Provide complete context
- • Specify output format
- • Include examples when possible
- • Add "say I don't know" instruction
- • Think before prompting
- • Save successful prompts
❌ Never Do
- • Use vague, one-line prompts
- • Assume the AI knows your context
- • Blame the AI for bad outputs
- • Skip the persona definition
- • Leave output format undefined
- • Expect perfection on first try
- • Forget to iterate and refine