Prompts to Practice: A Pedagogical Framework for Human-Centered AI Engagement

Synthesis: Miles, Haber-Curran, and Arar (2026) respond to the "ghost student" and cognitive debt problem in higher education by shifting the debate away from prohibition and detection toward a constructionist, classroom-level pedagogy. Building on Papert's constructionism, Freire's critical pedagogy and an ethics of care, they introduce the Instructional Model for Human-Centered Generative AI Engagement: five recursive phases running from critical and ethical awareness through prompt literacy, AI-supported learning, reflection and revision, to independent ethical application. The paper's central conceptual move is the distinction between prompt engineering, a technical optimization skill, and prompt literacy, a rhetorical, ethical and reflective process, operationalized through a nested five-step Prompt Literacy Cycle and a sample process rubric. The authors position generative AI as a thinking partner and an object to think with, and argue that metacognitive reflection, authorship and academic integrity are built through transparency and co-constructed norms rather than compliance rules. The framework is offered as a conceptual contribution and is explicitly framed as awaiting empirical evaluation.

Key Findings

1. Human-centered AI engagement is defined against two failed postures. Miles et al. (2026) define human-centered AI engagement as an intentional, reflective approach in which students work with generative AI as a thinking partner rather than a shortcut. It rejects both the reactive posture that treats AI as a threat to be prohibited, which positions students as risks to be managed, and the efficiency posture that prioritizes speed and convenience in ways that quietly displace the productive struggle essential to deep learning. The question the framework asks is not whether students use AI but how they engage with it: critically, transparently and in service of their own intellectual growth.
2. Prompt literacy is the framework's central mechanism, and it is not prompt engineering. Prompt engineering focuses on technical optimization of inputs for performance, whereas prompt literacy emphasizes clarity of purpose, rhetorical awareness, ethical responsibility and critical interpretation of outputs. The authors argue prompt literacy is a foundational literacy for contemporary learners, and that teaching only output optimization leaves the ethical and epistemological dimensions of LLM use untouched.
3. The Instructional Model has five recursive phases. Phase 1, critical and ethical awareness, covers how LLMs are trained and the risks of bias, misinformation, surveillance and academic dishonesty, with expected outcomes of ethical discernment and risk awareness. Phase 2, prompt literacy, runs the Prompt Literacy Cycle and targets rhetorical intentionality and AI literacy. Phase 3, AI-supported learning, uses LLMs for brainstorming, outlining, summarizing and drafting, targeting metacognitive awareness and reflective practice. Phase 4, reflection and revision, targets authorship development and critical judgment. Phase 5, independent ethical application, targets autonomy, academic integrity and transfer of judgment.
4. The Prompt Literacy Cycle gives daily-practice structure to Phase 2. Its five steps are Clarify Purpose (define the intention to explore, produce or understand), Craft the Prompt (design the first prompt with attention to tone, scope, audience and specificity), Engage with Output (critically annotate the LLM text for relevance, bias, completeness and tone), Refine the Prompt (revise and re-engage to see how input changes alter output) and Reflect (examine what the process revealed about how prompts function and what assumptions shaped the response). Instructors can introduce it through demonstrations, scaffolded assignments or workshops.
5. Prompt literacy can be assessed as process rather than product. Table 1 offers a sample rubric with three criteria: Iterative Refinement (from a single unadjusted prompt to systematically refined prompts with articulated reasons), Critical Interpretation of Output (from accepting output at face value to rigorously evaluating accuracy, bias and relevance and explaining the reasoning) and Reflective Revision (from minimal revision to integrating and transforming output while reflecting critically on how AI influenced thinking). The rubric grades engagement with the tool, not the AI-generated artifact.
6. Each phase has concrete classroom activities. Examples include AI Bias Exploration and Ethical Case Studies in Phase 1; a Prompt Remix Workshop, Prompt + Output Annotation, cycle journaling and an explicit Compare Prompt Literacy vs. Prompt Engineering debate or Venn diagram in Phase 2; AI + Human Draft Comparison and a Socratic Seminar with AI Support in Phase 3; a Revision Log, Authorship Workshop and AI-Influence Mapping in Phase 4; and a Personal AI Use Policy, Field-Based AI Case Study and Policy Co-Design in Phase 5.
7. Instructor modeling and shared norms carry the framework's values. The authors hold that faculty mediation determines whether AI functions as a shortcut or as a cognitive partner. Transparent modeling, such as thinking aloud about prompt phrasing, co-constructed classroom norms for LLM use and explicit links to institutional integrity policies, are presented as the practices that build authorship as something students actively construct and protect. A constructionist stance also shifts faculty from content experts to facilitators and co-learners.
8. Constructionism reframes LLMs as objects to think with. Rather than a threat to integrity, generative AI is treated as a cognitive and creative partner that sparks inquiry, deepens engagement and promotes critical thinking. Because LLMs respond dynamically to student input, they fit constructionist environments where knowledge emerges through experimentation, dialogue and reflection; students are invited to annotate or critique LLM contributions and treat outputs as a first draft rather than a finished product.
9. The model is conceptual and explicitly awaiting empirical test. The authors state it was developed by synthesizing established learning theory with instructional experience and ongoing research rather than derived from a single empirical study, that no datasets were generated or analyzed, and that empirical evaluation is currently underway. Future work is proposed on disciplinary differences, how students develop prompt literacy over time and how faculty implement the framework across institutional contexts and evolving integrity policies.

What this means for practice

• Instructors. Run the five phases as a semester-level progression rather than a single AI lesson, teaching ethics, prompting, drafting, reflection and independent application in sequence while returning to earlier phases as the process recurs.
• Instructors. Teach the Prompt Literacy Cycle explicitly — Clarify Purpose, Craft the Prompt, Engage with Output, Refine the Prompt, Reflect — and use the Compare Prompt Literacy vs. Prompt Engineering exercise to make the technical-versus-rhetorical distinction concrete.
• Instructional designers. Grade the process rather than the AI artifact, using criteria for iterative refinement, critical interpretation of output and reflective revision so that judgment and revision carry the credit, and move course and assessment design toward process-oriented, dialogic forms that foreground human presence, reflection and explanation, since output-based assessments are becoming unreliable indicators of learning in AI-mediated environments.
• Faculty developers. Model your own prompt refinement, output critique and ethical deliberation in class, and share that LLM use is iterative rather than a route to perfection or efficiency.
• Administrators. Co-design classroom norms with students and connect them to institutional integrity policies instead of relying on detection and compliance rules, with leadership defining why integration matters and instructors enacting how; resource the relational side of the work, since the authors hold that feedback, encouragement, empathy and cultural responsiveness remain irreplaceable.

Limitations

• The framework is a conceptual contribution synthesized from established learning theory and the authors' instructional experience rather than derived from an empirical study; no datasets were generated or analyzed for it.
• The process rubric presented in the paper is offered as a sample with three criteria, and the paper reports no reliability or validity evidence for it.
• The five phases, the Prompt Literacy Cycle and the classroom activities have not been tested for effects on student learning; the authors state that empirical evaluation is currently underway.

Connected Concepts

• Academic Integrity
• AI Literacy
• Cognitive Offloading
• Constructivism
• Critical Pedagogy
• Generative AI
• Higher Education
• Human AI Collaboration
• Metacognition
• Prompt Engineering

Citation

• Miles, A. J., Haber-Curran, P., & Arar, K. (2026). Prompts to Practice: A Pedagogical Framework for Human-Centered AI Engagement. Open Praxis, 18(3), 481-494.