AI Prompt Frameworks

Resources on writing Generative AI prompts for Medical Education

AI prompt frameworks are structured approaches to crafting effective prompts for AI models like Microsoft 365 Copilot, ChatGPT, etc., helping users guide the AI and achieve desired outputs. These frameworks provide a structured way to define the AI's role, task, context, and expected format, leading to more precise and relevant responses.

Popular AI Prompt frameworks

General Purpose

RISEN (Role, Input, Steps, Expectation, Narrowing)

Focuses on defining the AI's role, providing specific input, outlining clear steps, setting expectations for the output, and narrowing the scope of the task.

PGTC (Persona, Goal, Task, Context)

4-sentence framework helps structure prompts by defining the AI's persona, its goal, the specific task, and relevant context. 

RTF (Request-Task-Format)

Breaks down prompts into a request, the specific task, and the desired format of the response. 

RACE (Role, Action, Context, Execute)

Focuses on defining the AI's role, specifying the action, providing context, and setting clear expectations for execution. 

TAG (Task, Action, Goal)

Suitable for setting clear expectations about the process and outcome of a task. 

STAR (Situation, Task, Action, Result)

While commonly used in interviews, the STAR framework can also be applied to structuring responses and tasks, providing a clear and logical narrative flow. 

CLEAR (Concise, Logical, Explicit, Adaptive, and Reflective)

Emphasizes clarity, conciseness, and adaptability in prompt creation. 

SMART (Specific, Measurable, Achievable, Relevant, Time-bound)

While primarily used for goal-setting, the SMART framework can also be applied to prompt engineering to ensure clear and well-defined goals. 

RIDE (Role, Input, Directive, Examples)

Best for general-purpose prompting.

Example Prompt: 

You are a medical tutor. Based on the following patient case, help the student identify a differential diagnosis and explain their reasoning. Include 3 possible diagnoses and justify each briefly.

PROMPT (Purpose, Role, Output, Mechanics, Parameters, Testing)

Good comprehensive and adaptable prompt for different tasks.

TAP (Task, Audience, Purpose)

Designed for educational or instructional use cases.

AI Reasoning & Chain-of Thought

Chain-of-Thought Prompting

Enhances reasoning and logic output using phrases like “Let’s think step by step” or “First, list all the relevant symptoms.  Then, narrow down the diagnoses.”.  This type of prompting encourages multi-step reasoning and is especially useful in medical, legal, math, and programming domains.

5C Framework (Clarity, Contextualization, Command, Chaining, and Continuous Refinement)

Emphasizes clarity in the prompt, providing context, using commands, chaining thoughts, and continuously refining the prompt for better results.

Tree-of-Thought Prompting

Uses AI to break the problem into branches of possible paths or hypotheses and is used in decision-making and diagnostic scenarios using prompts such as “List multiple approaches and evaluate each one.”

Zero-Shot/Few-Shot Prompting

Used to teach the model through examples (or none at all). Zero-shot provides no examples relying on clear instruction. Few-shot includes 1-3 examples to guide the format or tone. Few-shot Chain-of-Thought combines example(s) with step-by-step reasoning prompts.

Example Prompt:

Here are two examples of a well-written patient note. Now write one for the following case...

AI-Aided Clinical Reasoning Prompt Framework

A custom hybrid used in medical simulations and AI-assisted learning.  It includes the Clinical Input (case info, symptoms, vitals, etc.), AI Role (“Act as a clinical mentors” or “standardized patient”), Output Directive (“Guide the student to a diagnosis with reasoning”), and Reflection Option (“Ask the learner to justify their choices”).

Example Prompt

You are a virtual standardized patient presenting with chest pain. Answer questions based on typical symptoms of myocardial infarction, and ask the learner to explain their diagnosis at the end.

GOLD Framework

Used for structured- problem-solving tasks. The components are Goal (What should the AI achieve?), Obstacles (What constraints should it account for?), Logic (What process or steps should be used?), and Data (What inputs should it consider?).

Specialized Prompts

PEACE Framework (Present the situation, Explore thoughts/feelings, Analyze alternatives, Choose a path, Evaluate the outcome)

Used for reflection or analysis and with educational, coaching or reflective tools. 

Reflective Prompting (e.g. Gibbs or DIEP)

Used to prompt reflection in learners using AI tools.  For DIEP, you would Describe (“Describe the situation”), Interpret (“What did you learn?”), Evaluate (“What was significant about it?”), Plan (“What will you do differently next time?”).

Example Prompts:

“Describe what happened during the patient interaction.”

“What went well? What would you do differently?”

“How did this experience shape your understanding of communication?”

Visual Prompt Framework (for DALL-E or Stable Diffusion/image generation)

Best when creating images or visuals and includes Subject (Who or what is in the image?), Setting (Where is it taking place?), Style (Realistic, Cartoon, Watercolor, etc.), Lighting (warm, soft, dramatic, etc.), and Perspective (close-up, aerial, wide angle, etc.).

Framework for code generation

This includes the Language (specific the coding language such as Python, JavaScript, etc.), Task (clearly state what the code should do), Input/Output (include sample inputs and expected outputs), Constraints (runtime, complexity, no libraries, etc.), Edge Cases (ask to handle common or tricky conditions).

These frameworks provide a starting point for crafting effective AI prompts and can be adapted based on specific needs and the type of AI model being used.

Citation: Microsoft Copilot, Accessed 2025-07-16.  Starting Prompt: “Develop a list of frameworks that can be used for developing AI prompts”)