Role, Action, Context, Example — A fast, four-part structure for when you need a good prompt in under a minute.
RACE strips ORACLE down to the essentials. Reach for it when the task is simple enough that you don't need to spell out layout and length separately — quick drafts, short-form content, one-off questions, and situations where you'd rather move fast than be exhaustive.
Who the AI should be while it answers — this shapes vocabulary and judgment more than any other single instruction.
The exact thing you want done, phrased as a direct instruction rather than a vague ask.
Just enough background for the answer to be usable — audience, purpose, or constraints.
A short reference point so the AI can match your intended tone or format.
Bad prompt:
“Give me some tweet ideas.”
Structured with RACE:
Role: You are a social media manager for a bootstrapped SaaS product. Action: Write five tweet ideas announcing a new feature. Context: The feature is a dark mode toggle. Our audience is mostly developers who dislike marketing-speak. Example: Tone should be closer to a changelog entry than an ad — plain, specific, a little dry.
Five short tweet drafts, each announcing the feature from a slightly different angle (the feature itself, why it exists, a specific use case, a light joke, a direct CTA), written in the plain, low-hype tone the example called for.
RACE works well as a universal format, and is especially effective with ChatGPT and Claude for short-form, high-frequency content where speed matters more than exhaustive detail.