AI Prompt Generator

Created by the Data Science & AI team at GovTech Singapore, CO-STAR is an industry-standard framework designed for enterprise use. It is rigorous about context and audience, ensuring that the AI output is not just accurate, but also tonally appropriate for its intended readers.

R-T-F strips away complexity to focus on the three essentials: who the AI is, what it needs to do, and how the answer should look. Perfect for single-turn requests where the output shape (JSON, table, email, etc.) matters most.

PACE defines who is acting, what they are doing, the context they are working within, and the expected format. It excels at producing structured, professional outputs for business use cases.

RACE treats the prompt as delegation to a skilled worker: "Here is your role, here is what you need to do, here is the situation, and here is exactly what I expect back." It minimises ambiguity by making success criteria explicit.

RISE is excellent for tasks that require processing data or following a specific logic chain. The "Input" and "Steps" components encourage Chain-of-Thought prompting, asking the AI to show its work or follow a defined procedure before producing the final output.

APE focuses on three things that matter most: what the AI should do (Action), why it matters (Purpose), and what the output should look like (Expectation). The Purpose field helps the model make better judgement calls when the task is ambiguous.

TAG separates the mechanical steps from the desired outcome. By explicitly stating the Goal, it helps the AI understand the underlying purpose of the request, leading to more insightful and aligned responses.

GRADE is designed for complex, multi-faceted requests. By separating Goal from Request and explicitly requiring Detail, it forces specificity about constraints. The Examples field gives the model a concrete target to match.

CRISPE is the most comprehensive framework available. It adds Instruction and Purpose layers on top of the usual Role and Context, making it effective for complex reasoning tasks where the model needs to understand not just what to do, but why.

COAST is useful for scenario planning and strategic analysis. The Scenario component lets you introduce hypothetical conditions ("What if budget is cut?", "What if a competitor launches first?"), making it powerful for contingency planning.

TRACE separates the broad Task from the specific Request, and mandates an Example. This makes it one of the best frameworks for ensuring the AI output matches a specific style, tone, or format that you provide.

ROLE is the framework of choice when you need predictable, high-quality output. The Limits section acts as a guardrail against common errors, and Evaluation provides a rubric so the AI understands exactly what "done well" looks like.