From Mapping to Refining: Prompting Techniques for Generative AI

This piece builds on a previous piece where we explored the importance of keeping your desired outcome in mind when working with generative AI (genAI). We also highlighted the value of including specific data and contextual details in your prompt to align with your goals. If you haven’t read that piece, we recommend starting there:

• Mapping Generative AI to Tailored Outcomes

In this piece, we will explore how to refine your approach to generate accurate and relevant responses from genAI. In short, we will shift our attention from what to say to genAI to how to structure your interaction with it. While the quality of your initial prompt matters, the most effective results emerge through refinement, follow-up, and clarification. GenAI is built for interaction. It rarely produces high-quality responses from a single request, especially for complex tasks.

Prompting techniques offer a way to intentionally structure the interaction process. They allow us to move beyond isolated requests toward a more collaborative, iterative engagement that shapes AI output over time. In the sections that follow, we outline three techniques—progressing from beginner-friendly to advanced—aligned with different levels of prompting experience, task complexity, and instructional needs.

The iterative refinement or prompt series technique is an easy and effective strategy for gradually exploring genAI's capabilities. Even if you don't yet have experience with using genAI, you can readily apply this technique to enhance the quality of the output you receive.

This technique involves engaging with genAI through a sequence of prompts that progressively refine or expand its output (Google Cloud, 2025). Many users engage with this technique instinctively—posing an initial question or a problem, reviewing the answer, and then issuing follow-up prompts to clarify, extend, or better align the output with their goals.

1. Create an initial prompt. Start with a clear and concise question or statement that sets the stage for interaction. Keep your goal in mind. Frameworks like RACEF (role, action, context, example, format), CIDI (context, instructions, details, input), or ACTS (audience, content, task, strategy) can help you structure your prompt effectively (Hardman, 2023).
   
   • Example: "You are a machine learning engineer. Explain machine learning to me."
2. Evaluate the AI’s response. There are two paths you might take here:
   
   • If the output is unclear, incomplete, or too technical, revisit and revise the initial prompt to guide the AI more precisely.
     
     • Refinement example: "You are a machine learning engineer. Explain machine learning to me without using any technical terms."
   • If the output is aligned with your topic area but not quite what you need, continue the conversation with a follow-up prompt to clarify, deepen, or redirect the response.
3. Create follow-up prompts. These are essential for shaping the AI’s response and aligning it with your evolving needs. Follow-ups can do the following:
   
   • Point out mistakes. Example: “That’s incorrect—machine learning isn’t a type of programming language. Can you revise that explanation?”
   • Ask for justification. Example: “Why did you choose that example to explain machine learning?”
   • Request more examples. Example: “Can you explain machine learning using a healthcare scenario?”
   • Add a rule or constraint, if your focus shifts. Example: “Now, explain machine learning in the context of financial analysis”. 
   • Ask for a next step, if your inquiry is part of a larger task. Example: “What’s the next step after understanding the basics of machine learning?”

Let’s say you're developing an undergraduate course on leadership, and you want to design an engaging first module that introduces key leadership concepts while encouraging active student participation.

You're aiming to:

• identify the foundational topics students should learn,
• reinforce these topics with interactive learning experiences, and
• assess students’ understanding in a way that supports long-term retention.

Using the iterative refinement technique, your interaction with genAI might unfold like this:

• Prompt 1: “You are an experienced university professor specializing in leadership development. List key topics that should be covered in an introductory module on leadership skills for undergraduate students.”
• Prompt 2: “For each topic listed, provide a succinct description (2–4 sentences) explaining its importance in the context of effective leadership.”
• Prompt 3: “Propose a creative, interactive activity for online settings tailored to undergraduate engagement, with a short note on how it reinforces the topic's learning objectives.”
• Prompt 4: “Create one short assessment question (e.g., a multiple-choice, short answer, or scenario question) to test understanding of the topic.”

This structured sequence allows you to build content collaboratively with AI while maintaining control over instructional goals, content quality, and overall alignment. You are not simply generating responses; you are progressively shaping the output toward your desired outcomes.

The logical pathway technique, also known as the chain of thought method, is well-suited for users who already have some experience working with genAI and are ready to adopt a more structured prompting approach. It’s particularly effective for tasks that require step-by-step reasoning, such as solving complex problems or addressing multi-part requests.

This technique enhances the accuracy and coherence of AI-generated responses by guiding the model to reason through a task in defined steps (DAIR.AI, 2025). It introduces intermediate reasoning stages, enabling the AI to build its response gradually, which often results in more thoughtful and relevant output (Wei et al., 2022).

This method is especially valuable when your task requires the AI to follow a clear process, apply a known framework, or simulate expert thinking. Rather than simply requesting a final output, you prompt the AI to break down the problem, move through individual components, and articulate each decision or insight along the way.

Consider how you would walk a student through analyzing a complex case study. Rather than asking for a final answer, you prompt them to break down the problem, evaluate options, and justify each decision. Similarly, the logical pathway technique leads AI through structured reasoning, encouraging step-by-step development just as we support learners in building thoughtful, well-founded conclusions.

1. Define your objective or a problem. Clearly articulate the task or problem you want the AI to address, including the intended outcome.
2. Break the task into actionable steps. Identify logical components or stages that the AI should follow to arrive at a comprehensive result. 
3. Add rules or conditions. Indicate any frameworks, conditions, or standards the AI should follow to align with your instructional or domain-specific needs.
4. Refine and iterate. Assess whether the AI’s reasoning is coherent and aligned with your expectations. Revise your initial prompt or use follow-up prompts as needed to improve clarity, accuracy, or alignment.

Suppose you're designing a reusable prompt to help generate practice quizzes that align closely with a final exam. You want to establish a consistent structure for quiz creation, while allowing flexibility across topics or audiences.

To accomplish this, you can use the logical pathway technique:

“You are an expert assessment strategist specializing in the design of innovative and insightful practice quizzes. You will need to ask me for the following information:

• What the quiz is intended to test
• Who the target audience is
• The final exam questions

Wait for my response. Then, using the provided information, create a set of multiple-choice practice questions. Ensure these new questions target the same underlying concepts and learning objectives as the final exam. Vary the context, wording, values, or examples to encourage deeper understanding and transfer of knowledge without duplicating the original questions. Maintain alignment with the difficulty level and style appropriate for the target audience.”

This approach supports both consistency and adaptability. The underlying reasoning process stays the same, while inputs can vary by topic, audience, or instructional goals—making it a scalable solution for assessment design.

The branching pathway, or tree-of-thoughts technique, is best suited for users who are already comfortable working with genAI and want to explore more complex interactions. The branching pathway technique is designed for tasks that involve open-ended problems, competing priorities, or multiple valid approaches. It allows you to compare and contrast alternatives, surface potential trade-offs or limitations, approach the problem from different angles, and assess each scenario against relevant criteria. This results in a more comprehensive analysis and supports critical decision-making (Gadesha & Kavlakoglu, 2024).

This approach is similar to how decision-makers might plan a major initiative. Rather than committing to one path immediately, they explore various possibilities in parallel—much like drafting several contingency plans before selecting the best course of action. In prompting terms, this means guiding the AI to generate multiple lines of reasoning or potential solutions so you can evaluate them before converging on the most appropriate one.

It is especially effective when used to simulate the type of multi-perspective thinking often required in program design, curriculum development, or strategic planning.

1. Define the central question or problem. Start by clearly framing the goal or decision you need to explore.
2. Identify key decision points or variables. Determine where different directions, options, or approaches could emerge from the central task. Prompt the AI to simulate or represent diverse viewpoints if needed.
3. Ask genAI to generate multiple paths. Ask for multiple alternatives, approaches, or pathways that address each key decision point. This step encourages divergent thinking before narrowing in.
4. Repeat the process for each new decision point. Once options have been explored at one level, prompt the AI to branch again—extending the exploration by introducing new decision layers, consequences, or adjustments.
5. Synthesize or compare outcomes. Conclude by asking the AI to evaluate the explored paths, weigh trade-offs, or recommend a solution based on defined criteria such as feasibility, alignment, or impact.

Imagine you are designing a graduate course on "Data-Driven Decision Making." You want to explore different ways to structure the course before selecting the most effective approach.

To accomplish this, you can use a prompt that uses the branching pathway technique:

“You are a curriculum designer for a graduate course on 'Data-Driven Decision Making.' First, propose three distinct ways to structure the course:

• One based on types of analytics (descriptive, predictive, prescriptive)
• One based on business functions (marketing analytics, operations analytics, finance analytics)
• One based on learner progression (beginner concepts to advanced applications)

For each structure, outline five modules. Then, evaluate the pros and cons of each structure in terms of learner engagement, skill development, and real-world application.”

This approach encourages a deeper analysis of multiple course design strategies, allowing for a more deliberate and effective final decision aligned with instructional goals.

Each of the three techniques—iterative refinement, logical pathway, and branching pathway—offers a practical way to structure your interactions with genAI, depending on the complexity of your task and the depth of reasoning required. As you become more comfortable with prompting, you can move between these techniques or even combine them to tackle multidimensional challenges.

Effective prompting is not just about asking better questions—it's about designing an interaction that mirrors how thoughtful professionals reason, adapt, and create. Through this approach, you position genAI not simply as a tool for content generation but as a structured thinking partner in the learning design process.

DAIR.AI. (2025, April 5). Chain-of-thought prompting. Prompt Engineering Guide.

Gadesha, V., & Kavlakoglu, E. (2024, August 15). What is tree of thoughts prompting? IBM AI Tech.

Google Cloud. (2025, April 17). Prompt iteration strategies. Generative AI on Vertex AI.

Hardman, P. (2023, November 30). Structured prompting for educators. Substack.

Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., Chi, E. H., Le., Q. V., & Zhou, D. (2022). Chain-of-thought prompting elicits reasoning in large language models. In S. Koyejo, S. Mohamed, A. Agarwal, D. Belgrave, K. Cho, & A. Oh (Eds.), NIPS '22: Proceedings of the 36th international conference on neural information processing systems (pp. 24824–24837).