10 Prompt Engineering Patterns That Actually Work

Prompt engineering isn't magic words — it's a set of repeatable patterns. Learn these ten and you'll get consistently better results from any model.

1. Be specific about the role and goal

Tell the model who to be and exactly what success looks like. Specificity removes guesswork.

2. Few-shot examples

Show two or three examples of the input-output pattern you want. Models are excellent at imitation.

Convert to a polite tone.
 
In: "Send me the file now."
Out: "Could you send me the file when you have a moment?"
 
In: "This is wrong, fix it."
Out: "I think there may be an error here — could you take another look?"
 
In: "Call me back."
Out:

3. Chain-of-thought

For reasoning tasks, ask the model to think step by step before answering. It dramatically improves accuracy on multi-step problems.

4. Output formatting

State the exact format you want — JSON, a table, bullet points — and the model will comply.

5. Constraints and guardrails

Tell it what not to do. "Don't invent sources" is as useful as any positive instruction.

6. Self-critique

Ask the model to draft, critique its own draft, then improve it. A second pass catches a surprising amount.

1. Write the answer.
2. List three weaknesses in it.
3. Rewrite addressing those weaknesses.

7. Decomposition

Break a big request into smaller prompts. Smaller tasks are easier to get right and easier to debug.

8. Grounding

Provide the source material and instruct the model to use only that. Essential for factual accuracy.

9. Persona priming

Set a consistent voice up front so a long conversation stays on tone.

10. Iterative refinement

Don't regenerate — revise. Point at the specific part that's off and ask for a targeted fix.

Patterns compose. The best prompts combine a clear role, a few examples, an output format, and a self-critique pass.

Putting it together

You don't need all ten at once. Add one pattern at a time to a prompt and watch the quality climb.