Prompt Engineering Playbook
A structured reference covering 20 prompt engineering techniques — each documented with original before-and-after examples, real Google Gemini API outputs, personal verdicts, and common failure modes.
Browse Foundational Techniques Browse Advanced Techniques
What Is This?
The Prompt Engineering Playbook is a practitioner-first reference that treats prompt engineering as a craft — something learned through empirical testing, not memorising technique names.
Every page follows a consistent seven-section template:
| Section | Content |
|---|---|
| Definition | Plain-English explanation of the technique |
| Use Cases | Three practical scenarios where the technique applies |
| Weak Prompt | An example prompt that underperforms, and why |
| Improved Prompt | The same task rewritten using the technique |
| LLM Outputs | Actual Gemini API responses — side by side |
| Verdict | Honest two-sentence assessment based on testing |
| Pitfalls | Common mistakes and how to avoid them |
Foundational Techniques
The ten foundational techniques every AI practitioner should know.
| # | Technique | Best For |
|---|---|---|
| 1 | Zero-Shot Prompting | Quick tasks, general queries |
| 2 | Few-Shot Prompting | Pattern-matching, formatting tasks |
| 3 | Chain-of-Thought | Reasoning, maths, multi-step problems |
| 4 | Role-Play / Persona | Expert tone, domain framing |
| 5 | Structured Output | JSON, tables, parseable responses |
| 6 | Negative Constraints | Scoping, avoiding unwanted content |
| 7 | Context Injection | Grounding LLM in specific facts |
| 8 | Audience Specification | Calibrating reading level & depth |
| 9 | Tone Control | Formal, casual, persuasive writing |
| 10 | Instruction Clarity | Reducing ambiguity in any task |
Advanced Techniques
Ten advanced techniques for complex, multi-step, and high-stakes tasks.
| # | Technique | Best For |
|---|---|---|
| 11 | Self-Consistency | Improving accuracy on reasoning tasks |
| 12 | ReAct Prompting | Agentic tasks, tool-use framing |
| 13 | Step-Back Prompting | Abstract reasoning, first principles |
| 14 | Prompt Chaining | Long, multi-stage workflows |
| 15 | Least-to-Most Decomposition | Complex problems, sub-task breakdown |
| 16 | Generated Knowledge | Factual accuracy, context priming |
| 17 | Calibration Prompts | Uncertainty, confidence estimation |
| 18 | Directional Stimulus | Steering creative or analytical tone |
| 19 | Meta-Prompting | Prompt generation, self-improvement |
| 20 | Evaluation Prompts | LLM-as-judge, quality assessment |
Findings Summary
After testing all 20 techniques across four task types (summarisation, factual Q&A, creative writing, and reasoning), key patterns emerged.
Tech Stack
- Site: GitHub Pages with Just the Docs Jekyll theme
- LLM Testing: Google Gemini API (
gemini-1.5-flash) - Scripts: Python 3.11+ with
google-generativeaiSDK - Prompts: 40+ original weak/improved pairs across 4 task types
Note: All LLM outputs shown in this playbook were generated during hands-on testing. Results may vary with different models, temperatures, or dates. Treat verdicts as empirically-grounded opinions, not universal truths.