{"id":106,"date":"2026-07-30T19:31:28","date_gmt":"2026-07-30T19:31:28","guid":{"rendered":"https:\/\/pressbooks.ulib.csuohio.edu\/usingaiinacademics\/?post_type=part&p=106"},"modified":"2026-08-06T20:16:04","modified_gmt":"2026-08-06T20:16:04","slug":"2-prompting-as-a-pedagogical-skill","status":"publish","type":"part","link":"https:\/\/pressbooks.ulib.csuohio.edu\/usingaiinacademics\/part\/2-prompting-as-a-pedagogical-skill\/","title":{"rendered":"2. Prompting as a Pedagogical Skill"},"content":{"raw":"Zhiqiang Gao, Fengxia Zhu, Graham Stead, Marnie Rodriguez\r\n

INTRODUCTION<\/h2>\r\nWhen generative AI tools became broadly accessible in late 2022, they entered classrooms before most institutions had time to consider whether they should. Three years on, the question for instructors is no longer whether students will use these systems; it is whether students (and the instructors who teach them) will use them with discipline or by imitation. Prompt quality determines the quality of the interaction and because so much learning now passes through that interaction, it has become an unexpected determinant of learning quality.\r\n\r\nThis chapter treats prompting as a pedagogical skill: a craft we model for our students by scaffolding their practice, assessing their development, and using teaching frameworks we trust. The University of Michigan\u2019s GenAI portal names this craft prompt literacy<\/em> and treats it as essential for everyone on campus (University of Michigan, n.d.). We frame it here as a skill analogous to information literacy or quantitative literacy. We adopt that framing for two reasons. First, it locates prompting where it belongs in the curriculum; not as a productivity hack tucked into a one-off workshop, but as a transferable skill that students will exercise across courses, careers, and civic life. Second, \u00a0as prompt teachers, we model our design choices, guide students through their own design choices, and assess whether they can transfer the practice to new domains. Doing either role well means treating prompting as a craft to be studied, not as a tool to be downloaded (Mollick & Mollick, 2023).\r\n\r\nThis chapter has three operational aims. After reading it you should be able to (1) explain pedagogical prompting, (2) design a prompt aligned to a specific learning outcome, and (3) scaffold a sequence of prompting instructions for students that follow the model\u2013guided practice\u2013independent practice\u2013reflection arc familiar in any well-designed skills curriculum.\r\n\r\nWe turn first to definitions and the core concepts that the rest of the chapter will lean on.\r\n

FOUNDATIONS<\/h2>\r\n

Definitions<\/h3>\r\nA prompt<\/em> is the input, namely text, examples, instructions, or context that shapes a generative AI system\u2019s response. Prompts can be a single sentence (\u201cExplain this concept to a first-year undergraduate\u201d) or several paragraphs of role-setting, constraints, and exemplars. Modern systems also accept images, code, and documents. What matters is not the form but the function. The prompt is the interface where human intent meets model behavior, and where the quality of the interaction is largely decided.\r\n\r\nA pedagogical prompt<\/em> (Xiao et al., 2025) is a prompt guided by a learning outcome, rather than a deliverable that governs the design. The distinction is easier to see by example. \u201cWrite a 500-word essay summarizing this week\u2019s reading\u201d<\/em> is a prompt that focuses on the deliverable, where the goal is the essay. \u201cGenerate three increasingly challenging short-answer questions a student can use to self-test their understanding of this week\u2019s reading, and provide a worked rationale for each correct answer\u201d<\/em> is a pedagogical prompt. The goal is the student\u2019s understanding. Productivity prompts can be useful in education, but only when the deliverable itself serves a learning purpose. From the start, pedagogical prompts are designed for learning.\r\n

Core Concepts<\/h3>\r\nA small set of design elements are important in pedagogical prompting. We name them here and return to them in the HOW-TO section.\r\n
    \r\n \t
  1. Role or persona.<\/strong> The position the model is asked to occupy: patient tutor, skeptical examiner, standardized patient, devil\u2019s advocate. Personas constrain the response in ways that are pedagogically useful when chosen deliberately.<\/li>\r\n \t
  2. Context.<\/strong> What the model needs to know to be useful: discipline, student level, prior topics covered, any source material from which the student is working.<\/li>\r\n \t
  3. Task framing.<\/strong> The verb that defines the request: explain, generate, critique, simulate. Sloppy framing produces sloppy output.<\/li>\r\n \t
  4. Constraints.<\/strong> What the response must or must not include: length, vocabulary, format, what information not to reveal.<\/li>\r\n \t
  5. Format.<\/strong> One or two demonstrations of what good output looks like. These often do more work than long instructions.<\/li>\r\n \t
  6. Chain of reasoning.<\/strong> A request that the model shows its working (its thought out processes) rather than a leap to a conclusion. For students, the visible reasoning is often more valuable than the answer.<\/li>\r\n \t
  7. Evaluation criteria.<\/strong> Stated inside the prompt: what counts as a good response; what should be flagged as uncertain.<\/li>\r\n \t
  8. Iteration.<\/strong> Prompts are drafts. The first response is data, not a result.<\/li>\r\n<\/ol>\r\nMIT Sloan\u2019s instructional guidance frames the essentials as three working strategies: (1) provide context, (2) be specific, and (3) build on the conversation (MIT Sloan EdTech, 2023). This guidance is a serviceable starting framework for any instructor whose students are new to prompting. The longer list above is what those three strategies expand into when the goal is learning rather than output.\r\n

    Connections to established frameworks<\/h3>\r\nThis section argues that prompting is not a new pedagogical category. Rather, it maps onto frameworks instructors already use.\r\n\r\nBloom\u2019s revised taxonomy.<\/strong> A prompt can be engineered to target any cognitive level. A prompt asking the model to define a key term from this week\u2019s reading<\/em> targets the understand<\/em> level. A prompt may ask it to generate a worked example illustrating when that term is applicable and explain why a small change in conditions would alter the outcome<\/em> targets of application<\/em> and evaluation<\/em>. The principle for the instructor is the same as for any assignment, namely choose the cognitive level the learning outcome calls for, and engineer the prompt to that level rather than defaulting to the lowest one (Anderson & Krathwohl, 2001).\r\n\r\nConstructive alignment.<\/strong> Biggs\u2019s (1996) principle of constructive alignment states that learning outcomes, teaching activities, and assessment must be aligned. This principle applies to AI-supported learning as much as it does to traditional instruction. In this context, a pedagogical prompt serves as the teaching activity. The intended learning outcome should be defined before the prompt is created, and the assessment should be designed to measure whether the outcome has been achieved.\r\n\r\nScaffolding.<\/strong> A well-designed prompt is itself a scaffold and a temporary support that lets a student attempt a task they could not yet be able to do on their own. When we teach students to prompt, we are also teaching them to scaffold their own learning process. The ability to breakdown tasks, seek appropriate support, and work toward independence is a valuable skill that will remain useful regardless of how AI technologies evolve.\r\n\r\nFormative assessment.<\/strong> Model output is a low-stakes mirror, i.e., that which does not provide severe consequences. A student who reads a model-generated worked example sees an external attempt at the reasoning they are trying to internalize; their job is not to copy it but to critique it. That critique is formative assessment in both directions. The student tests their understanding, and the instructor reads the critique to test their teaching.\r\n

    HOW-TO<\/h2>\r\nThis section translates the concepts of FOUNDATIONS into three working tools: a prompt design procedure, a decision tree, and a teaching scaffold.<\/em> They are designed to be used together. Adopt the procedure to draft a prompt; use the decision tree to confirm you have chosen the right kind of prompt for the learning task; and follow the scaffold when you teach students to do this work themselves.\r\n

    A seven-step prompt-design procedure<\/h3>\r\nBegin every prompt design with a learning outcome and end with evaluated output. The procedure is iterative; expect to loop back at least once.\r\n
      \r\n \t
    1. Goal.<\/strong> Clearly identify the learning outcome the prompt is designed to support. Which Bloom's Taxonomies of Learning (University of Arkansas, 2022) level is expected? Focus on the skills or understanding students should develop, not on the product or output the tool will create. Before finalizing a prompt, complete the sentence: \u201cAfter this activity, students should be able to\u2026\u201d<\/em> If you can\u2019t clearly articulate the intended learning outcome, pause the prompt-design process, define the outcome and then develop the prompt to support it.<\/li>\r\n \t
    2. AI's Role and Task.<\/strong> Specify who the AI should act as and state exactly what you want. For example, \u201cact like an experienced marketing professor (AI\u2019s role<\/em>) and explain segmentation, targeting and positioning (the task<\/em>)\u201d.<\/li>\r\n \t
    3. The Audience and Context.<\/strong> Clearly define the intended audience for the activity. Specify the discipline, students\u2019 academic or skill levels, prior topics covered, and any common misconceptions students may bring to the task. The model cannot reliably infer this information on its own. Without context, it defaults to a generic audience, and the responses may be too simple, too advanced, or focused on the wrong concepts.<\/li>\r\n<\/ol>\r\nIn addition, provide relevant background information about the learning context. This may include course materials students are using, where the activity fits within the broader course sequence, and any vocabulary or terminology that should be emphasized or avoided. The more relevant context you provide, the better the model can tailor its response to support the specific learning objectives and needs of your students.\r\n
        \r\n \t
      1. Format.<\/strong> Indicate the desired output, namely bullet point list, table format, flow chart, detailed explanation, word count, etc.<\/li>\r\n \t
      2. Constraints.<\/strong> Define the limits and requirements for the model\u2019s response. Specify details such as format, length, vocabulary level, tone, content to include, and content to avoid. Constraints help ensure that the output supports the learning goal. For example, instead of asking the model to \u201cgive me a sample answer,\u201d<\/em> you might ask for \u201ca sample answer that includes a common student misconception on this topic,\u201d<\/em> thus creating an opportunity for students to analyze and discuss the error.<\/li>\r\n \t
      3. Exemplars.<\/strong> Where possible, provide one or two brief examples of high-quality work. Examples often shape model output more effectively than lengthy instructions because they show exactly what is expected. Creating exemplars also helps you, as the prompt designer, to clarify and define what success looks like for the task.<\/li>\r\n \t
      4. Evaluate and iterate.<\/strong> Before using a prompt, decide what would make the response successful and what would make it ineffective. After running the prompt, compare the output against those criteria and identify areas for improvement. Then revise the prompt and try again. Treat the first response as feedback that helps refine the prompt, not as the final product.<\/li>\r\n<\/ol>\r\nColumbia Teachers College\u2019s (n.d.) Tips and Tricks for Prompt Writing<\/em> turns similar ideas into a practical checklist. It encourages prompt writers to be specific, use a persona, specify the output format, name what to do and not do, give examples, set the audience and tone, correct mistakes, and (when needed) ask the model to help draft a prompt. That checklist can be given directly to students as a guide. The six-step procedure above is intended for instructors as they design the prompt with which the students will work.\r\n

        Decision tree: what kind of prompt does this task need?<\/h3>\r\nA common error is to default to generation prompts when the learning task calls for something else. We find it helpful to make the type explicit. While not meant to be an exhaustive list, most pedagogical prompts fall into one of four kinds:\r\n
          \r\n \t
        1. Explanation prompts.<\/strong> Used when the students need a concept clarified or made more understandable. These prompts often include multiple representations of the idea, worked examples, analogies, or alternative ways of framing the same concept. For example: \u201cExplain the difference between X and Y using two different analogies and indicate to which audience each is best suited.\u201d<\/li>\r\n \t
        2. Generation prompts.<\/strong> Used when the student needs raw material to work with, such as drafts, problem sets, scenarios, or sample data. These prompts are defined by clear specifications for quantity, variation, and difficulty level. For example: \u201cGenerate five short cases at increasing levels of difficulty, each illustrating a different decision the student must make.\u201d<\/li>\r\n \t
        3. Critique prompts.<\/strong> Used when the student needs feedback on their own work or on the work of an imagined peer. These prompts typically include a rubric, adopt a Socratic or questioning stance, and explicitly avoid rewriting the work. For example: \u201cCritique the attached draft against this rubric. Ask three clarifying questions about the student\u2019s reasoning; do not rewrite any sentence.\u201d<\/li>\r\n \t
        4. Assessment prompts.<\/strong> Used when the prompt itself becomes the object of evaluation because the student has written it. In these cases, students are assessed on their ability to design, run, and reflect on prompts they create. The focus is on the quality of their prompting as a skill, rather than only on the final output. student\u2019s craft of prompting becomes the object of feedback. Hallmarks: assignment requires students to design, run, and reflect on prompts of their own.<\/li>\r\n<\/ol>\r\n[caption id=\"attachment_95\" align=\"aligncenter\" width=\"1024\"]\"\" Figure 1. Choosing the kind of prompt for which the learning task is called.[\/caption]\r\n\r\nThe four types are not mutually exclusive across a course; a single week can include all four. They are mutually exclusive within a single prompt: try to do more than one at a time and the output becomes muddy on all of them. After a prompt is drafted, review it against the seven steps outlined above.\r\n

          A scaffold for teaching students to prompt<\/h3>\r\nThe same four-stage scaffold (Fisher and Frey, 2021; Pearson and Gallagher, 1983) used in most skill-based instruction can also be applied to teaching prompt design.\r\n
            \r\n \t
          1. Model.<\/strong> The instructor demonstrates prompt design in real time, explaining each decision as it is made. This includes showing the initial draft prompt, the model\u2019s response, identifying what is weak or unsatisfactory, revising the prompt, and then comparing the improved response with the initial draft prompt response. The explanation of thinking is more important than the final output, since students need to understand the reasoning behind each choice.<\/li>\r\n \t
          2. Guided practice.<\/strong> Give students a draft prompt and a description of the desired output. Students are asked to propose several revisions and predict which revision will work best before testing them. During this stage, the instructor supports learning by asking questions that challenge students\u2019 reasoning rather than providing direct answers.<\/li>\r\n \t
          3. Independent practice.<\/strong> Students design their own prompts for assigned or self-chosen tasks aligned with course outcomes. They document the revision process and submit both the final prompt and the interaction history that led to the result.<\/li>\r\n \t
          4. Reflection.<\/strong> Students explain what worked, what did not, and what they would change in future iterations. This stage should be assessed, because without accountability, students tend to treat prompting as simple trial and error rather than a skill to be developed.<\/li>\r\n<\/ol>\r\nThree notes on the scaffold. First, it can be used at any education\/course level as what changes is the complexity of the underlying learning task, not the structure of the learning process. Second, modeling is the most frequently skipped stage, and when it is missing, students often fail to move beyond superficial prompting skills. Third, reflection is often treated as optional, but when it is not assessed, students learn to treat metacognitive work as unimportant, which undermines the entire approach.\r\n

            EXAMPLES<\/h2>\r\nNext, we examine examples that apply the design principles and scaffolding approach to specific learning tasks.\r\n

            Example 1. Scaffolded prompting activity for junior or senior marketing students.<\/h3>\r\nPedagogical goal.<\/strong> \u00a0By their junior or senior year, marketing students should be able to design, test, and refine AI prompts that generate useful marketing insights by applying effective prompting principles.\r\n\r\nStep 1: Instructor Demonstration. <\/strong>Instructor chooses a marketing scenario (e.g., a company plans to launch a new line of athletic shoes targeted at college students) and write an initial prompt (e.g., identify potential customer segments for athletic shoes) and show the response generated by the initial prompt. Next, the instructor refines the prompt following the seven-step prompt design process and adds role, task, context, and constraints to the prompt:\r\n\r\nRole: You are a marketing research analyst.\r\n\r\nContext: A company is launching a sustainable athletic shoe targeted at U.S. college students aged 18\u201324.\r\n\r\nTask: Identify three customer segments most likely to purchase this product.\r\n\r\nFor each segment:\r\n