How do I write a prompt that gets usable work?

What it is

A simple four-part structure, taught by Dan Martell among others, that turns a one-line request into a brief.

How it works

1. Role. Who to act as, for example “a B2B conversion copywriter”. The role narrows style and vocabulary.
2. Context. The real material: the brief, past examples, the audience, the constraints. Martell’s point is that most people give the model a text message when it can take in books’ worth of background (view: Martell).
3. Command. Exactly what to do: the steps, the length, what must be included and what must be avoided. Make the implicit explicit.
4. Format. A table, bullets, CSV or headed sections, chosen for where the output goes next.

Benefits

• Repeatable. The same structure can later become a system prompt.
• Faster review. A fixed format is quicker to check and to paste into the next tool.

Limits and risks

• More is not always better. Irrelevant or conflicting context lowers quality (see context rot). Include what matters, and lead with it. Martell makes the same point: too much information confuses the model (view: Martell). Research on long contexts found models use information at the start and end of a long input better than the middle (Liu et al.).
• A teaching model. This is a useful checklist, not a tested method. For deeper work, Anthropic and OpenAI publish their own prompt-engineering guides (checked 2026-10-17); Anthropic’s begins with defining success criteria and tests.