Seven prompt patterns that work in production

Reliable prompting is closer to interface design than to writing. You are defining a contract: here is the input shape, here is the output shape, here is what to do when something is missing.

Pattern one is role framing — state who the model is and what standard the answer must meet. Pattern two is structured output: ask for JSON or a fixed template so the result can be parsed.

Few-shot grounding is the highest-leverage pattern of all. Two or three well-chosen examples typically outperform several paragraphs of instructions.

Finally, always specify the failure path. Telling a model what to do when it does not know is the difference between a confident wrong answer and a useful one.