RACE
Role · Action · Context · Expectation — RTF plus the two things RTF is missing: background, and a definition of done.
Use it for
The default upgrade when a three-slot prompt keeps producing near-misses. Good for professional writing, analysis, and advice where the situation matters.
Avoid it for
Tasks with a strict output contract. RACE has no format slot, so structure lives inside Expectation and often gets lost there.
The slots
Each letter is a question the prompt has to answer.Role
Whose expertise?
Name the discipline and its seniority. Skip superlatives — "world-class" adds nothing measurable.
Example: A hiring manager who has run 200+ structured interviews for engineering roles.
Action
What must be done?
The verb and the deliverable.
Example: Write eight interview questions that separate candidates who have shipped production systems from those who have only studied them.
Context
What background changes the answer?
Constraints, audience, history, what you already tried. Include what must not happen.
Example: Backend role, Python and Postgres, 45-minute remote interview, candidates are mid-level, and our last set of questions rewarded memorised trivia.
Expectation
What counts as done?
Criteria plus shape. Say what a weak answer would look like so the model can steer around it.
Example: Each question with a one-line note on what a strong versus weak answer sounds like. No brain-teasers, no questions answerable from a blog post.
Strengths
- Four slots is about the limit of what people will fill in reliably.
- Context and Expectation together remove the two most common failure modes: invention and drift.
Weaknesses
- No format slot and no tone slot.
- Expectation gets overloaded, absorbing criteria, format and length all at once.
Sources and grounding
- 1Prompt design strategiesGoogle (Gemini API docs)Google documents context and constraint-setting as core prompt components.
- 2Prompting best practicesAnthropic (Claude Platform docs)Covers roles, context and explicit criteria as separate techniques.