Prompting Basics
Theory
The biggest upgrade in your AI use is not a new model. It is a clearer prompt.
A weak prompt forces the model to guess your job, audience, and standards. A strong prompt removes those guesses.
Weak
"Help me with this email."
The AI invents: who you are, who reads it, what you want changed, length, tone, and what facts to keep.
Strong
"Rewrite this email to my landlord. Ask for a one-week rent extension because my paycheck is late. Respectful, not apologetic. Under 100 words. Don't promise a specific new date."
Three patterns worth knowing by name:
- System prompt. The hidden setup line that tells the model its role and rules for the whole chat. In ChatGPT this lives in "custom instructions."
- Few-shot prompting. Give two or three examples of input and the answer you want. The model copies the pattern. Great for tagging, formatting, or matching a style.
- Chain-of-thought. For multi-step problems, "think step by step" often catches mistakes a one-shot answer would miss.
When output is bad, treat the prompt as the first suspect. Most "the AI is dumb" moments are really "the brief was thin."