Prompt Engineering Technique Selection Guide

Last updated: July 02, 2026

ai generative-ai llm prompt-engineering decision-making

Technique Reference

Technique	Definition	When to Use
Zero-shot	Clear, direct instructions with no examples	Simple, well-defined tasks where the model has strong baseline knowledge
Few-shot	Examples included in the prompt to guide format, tone, or style	Tasks requiring specific formatting or evaluation criteria not obvious from instructions alone
Chain-of-Thought (CoT)	Model articulates its reasoning step-by-step along a single path	Math problems, logical reasoning, multi-step analysis, debugging
Tree-of-Thought (ToT)	Model explores multiple reasoning paths in parallel, evaluates each, and can backtrack from dead ends	Planning problems, creative tasks with competing constraints, comparing alternatives
Persona-based	Assigns a role to shape vocabulary, depth, and assumed audience knowledge	Domain expertise, audience-specific explanations
Task decomposition	Breaks a complex task into smaller, ordered subtasks	Any task difficult to accomplish well in a single prompt
Self-consistency	Samples multiple outputs or reasoning paths and selects the most common answer	High-stakes decisions, complex reasoning where a single response might be wrong
Prompt scaffolding	Wraps user input in a structured template with explicit constraints	Defending against adversarial or off-topic input

Task Type to Technique

Task Type	Recommended Technique
Simple, well-defined	Zero-shot with clear instructions
Specific format or style	Few-shot with examples
Complex reasoning	Chain-of-thought
Multiple valid approaches	Tree-of-thought
Domain expertise	Persona-based + RAG
Multi-step workflow	Task decomposition
High-stakes decisions	Self-consistency

Common Pitfalls

Pitfall	Solution
Vague instructions	Be explicit about what you want
Too many constraints	Prioritize the most important requirements
No examples for complex formats	Add 2-3 diverse examples
Assuming model knowledge	Provide necessary context
Single-shot for complex tasks	Break into steps or use chain-of-thought