Prompting techniques that actually work

Agents & loops Advanced Gauntlet Loop A builder–critic agent loop: fresh critics judge each piece of work against a real reference and send the biggest gap back until the output wins. Best for: Long agent runs on one ambitious, inspectable artifact Reasoning Beginner Chain-of-Thought Prompting Get the model to write out intermediate reasoning steps before its final answer. Best for: Multi-step maths, logic and planning on non-reasoning models Structure & format Beginner Few-Shot Prompting Show the model a handful of worked input–output examples so it copies the pattern on a new input. Best for: Locking down format, tone and labelling rules Agents & loops Intermediate ReAct (Reason + Act) An agent loop that alternates a written thought, a tool action and the observed result until it can answer. Best for: Tool-using agents, research and multi-hop questions Reasoning Intermediate Self-Consistency Sample several independent reasoning paths and keep the answer most of them agree on. Best for: Questions with one checkable answer where accuracy beats cost Workflows Intermediate Prompt Chaining Split a task into a sequence of prompts where each step’s output feeds the next. Best for: Multi-stage work: research → outline → draft → edit Structure & format Beginner Role (Persona) Prompting Give the model an identity and audience to set its vocabulary, priorities and tone. Best for: Tone, framing and domain priorities — not accuracy Verification Intermediate Chain-of-Verification (CoVe) Draft, write fact-check questions, answer them independently, then rewrite using what survived. Best for: Cutting hallucinated facts in lists and long answers Reasoning Advanced Tree-of-Thoughts (ToT) Generate several candidate steps, score them, explore the best and backtrack from dead ends. Best for: Puzzles, planning and search problems with dead ends Workflows Beginner Meta-Prompting Use a model to write, critique or improve prompts instead of writing every prompt by hand. Best for: Drafting and improving prompts for repeated tasks