Mapping Generative AI to Tailored Outcomes

Imagine planning a trip to a new city. A quick online search highlights the usual downtown tourist spots, but as you explore more, you uncover unique neighborhoods—a financial district bustling with experts, a hidden restaurant scene, and a college area alive with bookstores and cafés. Yet, none of these appeared in your initial search for “best places to visit.”

Following the common path, you could easily miss these hidden gems, each promising new experiences and chances to meet interesting people. Similarly, you might think of a generative artificial intelligence (AI) knowledge base as a digital city, with layers of foundational knowledge at its "center," surrounded by lesser-known but focused areas of subject-matter expertise in its outer “neighborhoods.” A basic, broad prompt keeps the AI on a main road of common knowledge, while detailed directions venture into the “neighborhoods” that hold the insights, depth, and nuance you truly seek.

This is where prompt engineering—a method to unlock generative AI’s full potential—comes in. What is prompt engineering? How can we begin building confidence in giving detailed directions or prompts? This piece will address these questions and explore the elements that make a prompt efficient. Through practical examples, we’ll see how refining the details in your prompt can significantly influence the quality of output you receive from generative AI.

Importantly, the recommendations provided in this piece are tool-agnostic. They can be applied using any generative AI tools available at your institution, ensuring broad usability.

Prompt engineering is the craft of communicating with AI in a way that’s both clear and strategic (Knoth et al., 2024). By thoughtfully designing prompts, you can guide AI to deliver responses that aren’t just accurate but insightful and tailored to your needs.
 
Think of prompt engineering as creating a detailed roadmap for AI. Just as you wouldn’t expect to find a hidden gem in a city without specific directions, you can’t get meaningful results from AI without guiding it precisely. For instance, asking "What are some ways to improve teaching?" might yield general answers, but providing a specific prompt like "Suggest three active learning strategies for a graduate-level business course" directs AI to focus on actionable and relevant insights. 

For faculty and course developers, prompt engineering can be transformative. Whether you’re brainstorming course topics, drafting learning materials, or creating engaging assessments, well-designed prompts enable AI to support your goals more effectively.

Crafting an effective prompt starts with a clear purpose (Google Cloud, n.d.). Just as you wouldn't plan a trip without first selecting a specific destination, you need a clear objective when seeking AI’s help. For example, if you're developing course materials, clarify what you need—whether it’s a discussion prompt on a specific topic or a rubric for an assignment.

Once the goal is defined, the next critical step is providing precise and detailed instructions. The more specific your directions, the better the AI’s response will be. This is where the RACEF framework—role, action, context, examples, and format—comes into play. It provides clear guidelines for deciding what to include in your prompt and can act as a checklist to ensure all relevant components are addressed. Each element of RACEF plays a unique role in crafting effective prompts:

• Role defines the "who" of the AI. Assigning a role narrows the AI’s perspective or expertise to a specific domain. For example, instruct AI to act as a subject matter expert in curriculum development specializing in creating engaging, competency-based courses, helping to suggest interactive activities to increase student engagement.
• Action specifies what we want AI to do—the task at hand. This is where we direct the AI’s efforts, whether it’s summarizing information, writing content, or identifying patterns. Clarity is important here. Instead of using a vague statement like "analyze course feedback," you might specify "identify key themes and actionable suggestions from student course evaluations."
• Context provides necessary background and any specific rules or constraints the AI should follow. For instance, "Generate a group activity for an online graduate-level business course focusing on peer collaboration." Adding rules, like "use only digital collaboration tools," further refines the response.
• Examples guide AI by providing a model to follow. For example, if you are working on module learning objectives, providing a well-structured example, such as "By the end of this module, students will be able to evaluate different information systems and systems applicability in various business contexts" demonstrates the desired format and focus. Additionally, including examples of poorly written objectives—such as "Understand information systems" or "Learn about their uses"—helps highlight common mistakes and reinforces the desired standards. Clearly labeling these examples as ineffective guides the AI toward producing more precise and actionable objectives.
• Format specifies how you want the response presented (e.g., bullet points, a summary, or a detailed report).

Not every prompt requires all RACEF components, and their order can be adapted to fit the task. The flexibility of RACEF allows you to craft prompts that are both structured and adaptable, ensuring the AI delivers meaningful, relevant responses. The key is to include what’s most relevant to the situation.

Let’s see how it works in practice. Imagine you are teaching students how to provide constructive feedback. Here’s how you might craft a prompt:

• Role: You are an experienced instructional designer who specializes in creating engaging and hands-on learning experiences.
• Context: I am a professor of business analytics, and my students find it challenging to provide constructive feedback.
• Action: Recommend five approaches for them to practice this skill, which I can incrementally introduce in the course.
• Format: Format your response as a bulleted list with a brief explanation and outline the pros and cons for each recommendation.

This prompt is intentionally open-ended, allowing the AI to explore diverse strategies without being constrained by examples. By keeping the field specific to business analytics, the recommendations align closely with the course context.

Now, consider a different scenario where an instructor wants students to use AI to generate content and then critically analyze and evaluate the output. To guide students in thinking about what to include in their own prompts, the instructor decides to provide a sample prompt they can use to generate the initial output. To align with this objective, we’ll modify the previous prompt to fit a different field and instructional goal.

• You are an experienced instructional designer who specializes in creating engaging and hands-on learning experiences. I am a professor of learning science, and my students find it challenging to provide constructive feedback. Recommend five approaches for them to practice this skill, which I can incrementally introduce in the course. Format your response as a bulleted list with a brief explanation, and outline the pros and cons for each recommendation, following the example provided below.
  
  • Provided example: Role-playing exercises
    
    • Explanation: Students pair up and take turns giving and receiving feedback in a controlled, simulated scenario.
    • Pros: This provides a safe environment to practice and helps students understand both perspectives.
    • Cons: The exercise can be time-consuming and may feel artificial to some students.

In this adapted prompt, the inclusion of examples and the shift to learning science significantly changes the AI’s focus and the nature of the output. In the first sample prompt, the goal was to provide actionable recommendations for a business analytics course, with an open-ended structure that enabled the AI to explore diverse strategies without the constraints of specific examples. This approach encouraged creativity and flexibility in generating recommendations aligned with the course context.

In contrast, the second revised prompt adds a layer of specificity by including examples and changing the field to learning science. It structures the output, guiding students to systematically assess the quality, relevance, and effectiveness of AI-generated recommendations. This shift enhances critical thinking by encouraging students to evaluate AI-generated content for accuracy, coherence, and alignment with course concepts.

These examples demonstrate how even small modifications to a prompt—such as adding examples or altering the field—can lead to tailored outputs that align with specific pedagogical goals and enhance the learning experience. The key is to clearly define your objectives and desired outcomes before creating a prompt.

Consider the following examples, each utilizing some elements of the RACEF framework, which you can use as a starting point or adapt as you experiment with generative AI for your own purposes.

• Role: You are an expert in environmental science education specializing in interactive online teaching.
• Context: I am designing an undergraduate course on climate change, and I want students to critically engage with real-world data.
• Action: Suggest three interactive assignments that involve using real-world climate data to analyze trends and propose solutions.

• Role: You are an experienced computer science instructor.
• Context: I am teaching an introductory Python programming course for graduate students with no coding experience.
• Action: Propose three coding challenges that assess students’ ability to apply key programming concepts (e.g., loops, functions, and error handling).

• Action: Develop a study guide outline for a history course focusing on the American Revolution based on the provided excerpt from the textbook.
• Format: Provide a framework with key sections, topics, and sample questions.
• Example:
  
  • Key sections: Major battles, key figures, political impacts
  • Topics: The Declaration of Independence, the role of women in the revolution
  • Sample questions: "What were the economic causes of the American Revolution?" and "Compare and contrast the strategies of the British and American forces."

• Context: I am teaching an MBA course on business ethics and want students to engage in deeper analysis of ethical dilemmas.
• Action: Provide three discussion questions and include one or two examples of poorly written questions, highlighting why they are ineffective.
• Examples of effective questions:
  
  • "What are the ethical implications of prioritizing shareholder value over employee well-being in decision-making?"
    
    • Why this question is effective: This question encourages critical thinking by exploring a nuanced and real-world conflict.
  • "How might cultural differences influence ethical decision-making in global organizations?"
    
    • Why this question is effective: It challenges students to consider diverse perspectives and contextual factors.

• Examples of ineffective questions:
  
  • "Are businesses ethical or unethical?"
    
    • Why this question is ineffective: This question is too broad and binary, lacking nuance and failing to provoke meaningful analysis.
  • "What is the definition of business ethics?"
    
    • Why this question is ineffective: This is factual recall, not critical thinking, and does not align with the objective of fostering deeper engagement.

Experimenting with different combinations of RACEF components will help you understand which elements work best for specific tasks. Here are some general tips:

1. Clarity and simplicity: Use simple, direct language. Avoid ambiguous terms or overly complex instructions.
2. Define the goal: Start with a clear purpose or overview to orient the AI.
3. Be specific but concise: Provide enough detail for clarity, especially for complex tasks, while keeping instructions as straightforward as possible.
4. Avoid assumptions: Don’t assume the AI knows specific topics; spell out any necessary context.
5. Organize instructions: Present the prompt in a structured, logical format.

By refining your approach to prompt engineering, you can unlock the full potential of AI to create transformative learning experiences—one prompt at a time.

Google Cloud. (n.d.). Prompt engineering: Overview and guide.

Knoth, N., Tolzin, A., Janson, A., & Leimeister, J. M. (2024). AI literacy and its implications for prompt engineering strategies. Computers and Education: Artificial Intelligence, 6, Article 100225.