RACE — Role, Action, Context, Expectation — is a four-part framework that adds explicit context and a clear success bar to the prompting basics. Full breakdown with an example for each element.
By The DDH Team at Digital Dashboard Hub·Updated
RACE stands for Role, Action, Context, and Expectation — a four-part prompt structure that tells the model who to be, what to do, the background it needs, and what a successful answer looks like. Its distinguishing move is the Expectation slot: you state the success criteria up front, so the model writes toward a defined bar instead of guessing what 'good' means.
RACE is a community-popularized framework; the techniques it bundles — role assignment, clear actions, supplying context, and specifying success criteria — are standard practice documented in the DAIR.ai Prompt Engineering Guide and Learn Prompting. To assemble a RACE prompt quickly, start with the ChatGPT Prompt Generator.
Digital Dashboard Hub
Writing good prompts for ONE AI is hard. Writing them for GPT-5, Claude, Gemini, Perplexity, Midjourney and 6 more is a full-time job. DDH's AI Prompt Builder writes once, runs everywhere — locked to your niche, voice, and brand tone.
Free 14 days, no card. →
| Feature | Plain prompt | RACE prompt |
|---|---|---|
| Assigns a specific role | ||
| States one clear action | Often vague | Yes |
| Supplies real context | ||
| States explicit success criteria | ||
| Output matches a definition of done | Rarely | Usually |
| Easy to review against a bar | No | Yes |
| Best for | Quick questions | Work deliverables |
Comparison reflects established prompting practice per the [DAIR.ai guide](https://www.promptingguide.ai/) and [Learn Prompting](https://learnprompting.org/). Current as of June 2026.
A complete breakdown of the RACE framework, built to skim and copy. Sections, in order:
1. What RACE is and how it compares · 2. R — Role · 3. A — Action · 4. C — Context · 5. E — Expectation · 6. A full assembled RACE prompt · 7. RACE vs plain prompting (table) · 8. Common mistakes · 9. FAQs · 10. Sources & further reading.
Each element has a copy-paste line, and the assembled example shows all four working together.
RACE sits between the bare-bones RTF (Role-Task-Format) and the voice-heavy CO-STAR. It keeps things lean — four slots — but swaps RTF's 'Format' for an 'Expectation' that's broader than format alone: it covers what the output must achieve and how you'll judge it, not just its shape.
The two slots that earn RACE its keep are Context and Expectation. Context grounds the model in your real situation (the thing plain prompts always omit), and Expectation sets the success bar so the model self-targets quality. That makes RACE a strong default for work deliverables — briefs, analyses, reports, structured drafts — where 'good' has concrete criteria.
Like every framework here, RACE is the same core idea — supply what the model lacks — packaged as a memorable checklist. Use the slots that apply and drop the rest.
Role assigns the model a specific identity and expertise level, which sets its vocabulary, depth, and priorities. Be concrete — name the specialty and, where relevant, the seniority.
``` Role: You are a senior product marketing manager who writes launch briefs for engineering-led teams. ```
**Why it works:** role prompting is a reliable, low-cost lever documented by DAIR.ai. A specific role pulls the model out of generic-assistant mode and toward the conventions of that field. 'A marketer' is weak; 'a senior product marketing manager for engineering-led teams' is actionable.
Action is the concrete thing you want done — the verb and the deliverable. Keep it to a single clear action; if you're asking for several things, that's a sign to split or chain the work.
``` Action: Write a one-page launch brief for our new analytics dashboard. ```
**Why it works:** a precise action gives the model an unambiguous target. The failure mode is fuzziness ('help with the launch') — name the artifact and the verb. If your Action contains an 'and ... and ...,' consider the prompt chaining approach instead of overloading one prompt.
Context is the background the model needs: your situation, constraints, audience, and any facts it must treat as ground truth. This is RACE's highest-leverage slot — most generic output traces back to missing context.
``` Context: B2B SaaS, mid-market buyers. The dashboard turns raw event data into plain-English weekly summaries. Differentiator: no SQL required. Main competitor positions on 'powerful' (read: complex). Launch is in 3 weeks. Do not invent metrics or customer names. ```
**Why it works:** grounding the model in your reality stops it from defaulting to safe generalities and lets it write around your actual constraints. The 'do not invent' clause is a guardrail worth keeping whenever facts are involved — the everyday form of the negative-constraint pattern from our 12 prompt patterns guide.
Expectation states what a successful answer looks like: the output shape, length, and the criteria you'll judge it by. This is where RACE goes beyond a plain format spec — you're handing the model your acceptance test.
``` Expectation: One page, three sections — Positioning, Key messages (3 bullets), Launch checklist. Lead with the 'no SQL' differentiator. Plain language, no hype words. Success = a PM could run the launch from this brief alone. ```
**Why it works:** stating the success bar lets the model optimize toward it and lets you check the draft against the same criteria. The 'success =' line is the most useful habit in RACE — it converts a vague hope into a testable target. For machine-readable output, pair the shape spec with native structured-output modes (OpenAI, Claude).
All four elements stacked into one prompt you can paste and adapt:
``` Role: You are a senior product marketing manager who writes launch briefs for engineering-led teams. Action: Write a one-page launch brief for our new analytics dashboard. Context: B2B SaaS, mid-market buyers. The dashboard turns raw event data into plain-English weekly summaries; differentiator is no SQL required. Main competitor positions on 'powerful' (read: complex). Launch in 3 weeks. Do not invent metrics or customer names. Expectation: One page, three sections — Positioning, Key messages (3 bullets), Launch checklist. Lead with the 'no SQL' differentiator. Plain language, no hype. Success = a PM could run the launch from this brief alone. ```
Stacked this way, each element shuts down a different failure mode, and the Expectation line gives you and the model the same definition of done. That shared success bar is what makes RACE output feel finished rather than first-draft.
---
RACE compresses well, too. For a quick task: `As a recruiter (R), draft a 4-line outreach DM (A) for a backend engineer at a fintech who already has a job (C); success = it earns a reply without sounding like a mass blast (E).` The order — Role, Action, Context, Expectation — keeps it memorable.
Use RACE when: you're producing a work deliverable — brief, analysis, report, structured draft — where success has concrete criteria you can state up front, and context matters more than fine-grained tone control.
Use RTF / CO-STAR instead when: the task is a trivial single-shot (RTF), or voice is the hard part and you need separate style/tone/audience dials (CO-STAR). RACE's strength is the success-bar, not nuanced voice.
The clearest way to see RACE's value is to watch the same request transform. Suppose you need a recruiting outreach message.
Before (plain): `Write a recruiting message for a backend engineer.`
That returns a generic, often over-eager template — wrong tone for a passive candidate, no sense of your company, and no bar for what 'working' even means. You'll rewrite it three times.
After (RACE):
``` Role: You are a senior technical recruiter who places backend engineers at fast-growing fintechs and respects candidates' time. Action: Write a first outreach message (LinkedIn DM length). Context: Target is a backend engineer with 6+ years in payments who is currently employed and not actively looking. We are Series B, remote-first, and the differentiator is owning a payments rail end to end. Do not invent salary numbers or claim we're 'the best.' Expectation: 4-5 sentences, no buzzwords, opens with something specific to their work (not about us), one clear low-pressure ask. Success = a busy, happily-employed engineer would actually reply. ```
The RACE version produces a message calibrated to a passive candidate, grounded in your real pitch, and measurable against a stated bar. The 'success =' line is doing the heavy lifting — it tells the model the message is judged by reply-worthiness, not word count. Productize this with the Business Email Generator.
---
Notice that Context carries the guardrail ('do not invent salary numbers') and Expectation carries the quality definition. That division of labor — facts and constraints in Context, the success bar in Expectation — is the habit that makes RACE prompts consistent across runs.
Skipping the success bar. The whole point of Expectation is the 'success =' criterion. Stating only a format ('one page, three sections') without saying what makes it good wastes the slot.
Thin Context. A one-line context produces a generic brief. The richer the context — constraints, audience, competitor, facts — the less boilerplate you get back.
Overloaded Action. Multiple deliverables in one Action line dilute all of them. One artifact per prompt; chain the rest (see prompt chaining).
No factual guardrail. When the task touches facts, add a 'do not invent' clause to Context, or the model will fill gaps with plausible fabrications. Always verify any number or name in model output against a real source.
RACE is a community framework; the techniques it bundles are documented in established sources (current as of June 2026):
DAIR.ai Prompt Engineering Guide: https://www.promptingguide.ai/
Learn Prompting: https://learnprompting.org/
OpenAI prompting guide: https://platform.openai.com/docs/guides/prompt-engineering
Claude prompt-engineering overview: https://docs.claude.com/en/docs/build-with-claude/prompt-engineering/overview
Google Gemini prompting strategies: https://ai.google.dev/gemini-api/docs/prompting-strategies
Related on this site: Complete Guide to Prompt Engineering, 12 Prompt Patterns That Convert, and the related RTF, CO-STAR, and CRISPE frameworks.
RACE stands for Role, Action, Context, and Expectation. You tell the model who to be (Role), what to do (Action), the background it needs (Context), and what a successful answer looks like (Expectation). The Expectation slot — stating success criteria up front — is what distinguishes it from simpler frameworks.
RACE keeps RTF's Role and a task-style Action, but adds an explicit Context slot and replaces 'Format' with 'Expectation.' Expectation is broader than format: it covers what the output must achieve and how you'll judge it, not just its shape. RACE is the better fit when 'good' has concrete success criteria; RTF is leaner for trivial tasks.
Context and Expectation. Context grounds the model in your real situation, which is what plain prompts almost always lack. Expectation states the success bar so the model writes toward a defined target — the 'success =' line is the single most useful habit in RACE, because it doubles as your acceptance test when reviewing the draft.
Use RACE for work deliverables — briefs, analyses, reports, structured drafts — where success has concrete criteria but fine-grained voice control isn't the hard part. Use CO-STAR when tone is the challenge and you need to control style, tone, and audience as separate dials.
No — RACE is a community-popularized acronym, not a single published method with a documented inventor. The underlying techniques (role prompting, clear actions, supplying context, specifying success criteria) are standard practice documented in the DAIR.ai guide and Learn Prompting. Treat the acronym as a memory aid.
The free ChatGPT Prompt Generator turns a rough idea into a Role-Action-Context-Expectation prompt ready to paste — no signup, part of 40+ free prompt tools.
Browse all prompt tools →