Prompt Patterns|

A Pattern Language for Knowledge Engineering with Large Language Models

Core Prompting Principles

After reviewing dozens of papers, I’ve distilled prompting down to six core principles. Following these principles makes your prompts more reliable by reducing ambiguity, preventing missing-context failures, improving reasoning quality, and producing outputs that are easier to verify and use.

OFP01

Define the outcome, scope, constraints, audience, and success signals so the model does not guess what “good” means.

ISP02

Provide the facts, constraints, and definitions that shape the outcome, and omit details that do not affect the task.

MPP03

Specify a method when the work has trade-offs or multiple steps so the model follows a deliberate process rather than jumping to a generic conclusion.

QVP04

Frame the output as a set of claims, surface assumptions and uncertainty, and validate against correctness, constraints, and completeness before relying on it.

ORP05

Declare the required format and fields up front so the response is a stable artifact you can review, share, and reuse.

IAP06

Iterate toward alignment by using each response to adjust the objective, inputs, method, checks, or format until it meets acceptance criteria.


These principles are general rules (similar in spirit to SOLID in software design), they guide judgment without prescribing exact steps. The patterns in this book are the concrete, reusable prompt designs that put the principles into practice, and there are many patterns you can choose from depending on the task.