Prompt Engineering 101

Forget tricks. There are seven core moves. Master these and you'll outperform most "prompt engineers."

The seven moves

1. Be specific about the output format. Don't say "summarize this" — say "summarize this in 3 bullet points, each under 15 words."
2. Give examples (few-shot). Two or three input→output examples beats any amount of explanation.
3. State the role. "You are a senior financial analyst." gives the model a persona to perform.
4. Decompose. Break complex tasks into steps. Ask the model to "think step by step" or use explicit numbered subtasks.
5. Show, don't tell. Paste an example of "good" — the model will reverse-engineer it.
6. Constrain. "Reply in valid JSON matching this schema: {...}" cuts hallucinations dramatically.
7. Iterate. Your first prompt is a draft. Refine based on the actual output.

A complete template

ROLE: You are an experienced [persona].
TASK: [single sentence].
INPUT: [the data].
OUTPUT FORMAT:
- [constraint 1]
- [constraint 2]
EXAMPLES:
Input: ...
Output: ...

Try this template on any task in the Playground. The improvement over a one-liner is usually dramatic.