Generative artificial intelligence (AI) is already part of how our students learn, study, and communicate. The question is no longer whether students will use it. The question is whether we, as educators, will guide its use in a way that strengthens clinical reasoning, protects patient safety, and preserves academic integrity.¹
I have seen the best and the worst of AI in higher education. I teach health sciences courses where students are preparing for real clinical decision making, and I serve on my college’s AI advisory board. When AI is treated as a shortcut, it can quietly erode the exact skills we are trying to build: judgment, synthesis, and accountability. When AI is treated as a tool with clear expectations, guardrails, and reflection, it can support learning in ways that were difficult to scale before.²
This article is written for a broad audience of health professionals who teach, precept, or assess learners. It is also written for those who are trying to decide how to respond when a student submits work that feels too perfect, or when a learner admits they used AI, and is not sure whether the student crossed a line. We need shared language, clear expectations, and a practical approach that does not rely on guessing or policing.
Why this matters in health professions education
Health professions education is different from many other disciplines because the end point is not only learning the content, but also applying it to patient care. That reality changes the stakes. If a learner uses AI to produce a polished assignment but cannot explain the clinical reasoning behind it, the risk is not theoretical. It shows up later as missed red flags, incomplete histories, unsafe prescribing, or overconfidence in incomplete information.
At the same time, our learners are overwhelmed. They balance work, clinical hours, family responsibilities, and complex content. AI tools can assist with studying, tutoring, writing support and early organization of ideas. Used responsibly, these tools can reduce friction and free time for higher level learning.¹ The goal is not to ban a tool that students will encounter throughout their careers. The goal is to teach responsible use that strengthens – not replaces – critical thinking.
Start with shared expectations, not enforcement
One of the most helpful shifts I have seen is moving from fear-driven reactions to clearly labeling the expected level of AI use for a course or assignment. The University of Kentucky’s Student AI Use Scale offers a practical framework that describes a continuum from Level 0, where the student is the sole author, to higher levels where AI may support brainstorming, outlining, drafting, and synthesis, with the student maintaining responsibility for oversight.³
This kind of scale does two important things. First, it reduces ambiguity. Students do not have to guess what is permitted. Second, it reframes AI as a set of choices that must match learning objectives. In clinical education, that alignment matters. If the objective is to assess clinical reasoning, then the work must reveal the student’s reasoning process. If the objective is to practice patient education writing, then a tool that generates an entire handout defeats the purpose.
At my college, we communicate expectations at the start of the course and revisit them as assignments become more complex. Our approach is generally supportive of AI, with specific exceptions based on course outcomes.
Why an in-house tool can change the conversation
Many schools are trying to manage AI use while students independently use public tools that vary in quality, privacy protections, and cost. One response is to provide an in-house option that is available across programs. At my institution, we developed Scout, an in-house generative AI assistant available throughout the college. Students use Scout for tutoring and content creation, and faculty can incorporate Scout into assignments while emphasizing guardrails, verification, and professional behavior.⁴
The advantage of a college-supported tool is not that it eliminates misuse. The advantage is that it lets faculty teach AI skills transparently using a shared platform, with more consistent access across students.
How I use AI in the classroom without losing the point of the course
I teach in a way that tries to make clinical thinking visible. AI tools can support that goal if we use them intentionally.
One approach that works well is using AI for case study creation, and then grading the student’s clinical skepticism. I ask students to use Scout to generate a realistic patient scenario for a specific disease topic. Then the assignment becomes an exercise in verification. Students must identify what is accurate, what is incomplete, and what is wrong. They correct the case using course resources and evidence-based guidance. They also explain why each correction matters for patient safety.
This turns AI into a simulation partner rather than an answer key. It mirrors real practice where patients present with messy, incomplete stories, and clinicians must reconcile what they are hearing with what is plausible and safe.
A concrete example: PrEP decision making in pathophysiology
In my pathophysiology course, I have used AI to generate case studies focused on HIV prevention, specifically scenarios that require selecting the most appropriate pre-exposure prophylaxis (PrEP) approach. Students may see a scenario where oral PrEP might fit the patient’s preferences and clinical context, or where long-acting injectable PrEP may be a better match. AI can draft the case and even propose a plan, but students must verify every recommendation against current guidance and explain the rationale in their own words. ⁵
This approach is valuable because it forces students to demonstrate the thinking we want them to carry into clinical practice: patient-centered decision making, attention to contraindications and adherence factors, and guideline literacy. AI can accelerate scenario building, but it cannot be the final authority.
Where AI helps learners do more real learning
AI can support health professions education when it is intentionally matched to learning goals. The most meaningful benefits I see include the following:
Tutoring and immediate feedback. Students can ask for explanations of complex concepts, request alternative explanations and practice with self-testing. For students who hesitate to ask questions in front of peers, an AI tutor can reduce shame and increase repetition.¹
More efficient preparation. AI can help students outline reading, generate study questions, and organize notes. It can also help faculty create practice cases, discussion prompts, and scaffolding materials that support learning.²
Support for writing and communication. Many students struggle to translate clinical knowledge into clear writing. AI can assist with readability and structure when students maintain ownership of ideas, verify claims, and cite primary sources.
A bridge to modern clinical decision support thinking. Whether we call it AI or decision support, clinicians have long interacted with tools that suggest orders, flag interactions, and summarize data. Learning to question outputs, check sources, and understand limitations is part of modern clinical competence.²
The pitfalls we must teach directly
The risks are real, and they show up in predictable patterns.
Confident misinformation. AI can produce inaccurate information in a convincing tone. In health care, that is a safety issue. Therefore, I repeatedly tell students: that the tool can support your learning, but it cannot replace your accountability.
Fabricated citations and false authority. AI can generate references that look real but do not exist. That is especially harmful in evidence-based education because it teaches the wrong habit: citing for appearance instead of citing for truth. Students must learn that AI may help with keywords and organization, but it cannot be trusted to generate final references.
Erosion of critical thinking through overreliance. If students use AI to produce answers before they attempt the problem, they skip the struggle that builds reasoning. Over time, that can reduce confidence in their own judgment.
Privacy and professionalism. Students may paste sensitive information into tools without understanding data risks. Even without protected health information, professionalism matters. We need to teach learners to treat prompts as part of their professional footprint.
Assessment of ambiguity and academic integrity. Educators often ask, “How do I know if this is AI?” In reality, AI detection is not a reliable strategy. It can miss AI generated text and it can wrongly flag authentic writing. An editorial in BMJ Open Sport and Exercise Medicine demonstrated how generated text could evade detection tools and highlighted the broader risks of false confidence in detection.⁷
A better approach than detection: design assessments that reveal thinking
If detection is unreliable, assessment design becomes the most powerful lever we have. The best strategies do not rely on “gotcha” enforcement; they require evidence of reasoning.
Process-based grading. Require intermediate products such as outlines, rationale statements, annotated references, and reflection memos. AI can help produce text, but it cannot replace a coherent explanation of why decisions were made.
Oral defense and brief check-ins. Ask students to explain their decisions, justify a plan, or respond to follow up questions. This can be done in small groups or short recorded responses.
In class application tasks. Use short, supervised case analysis where students must apply concepts in real time.
Clinical reasoning rubrics. Grade reasoning, not just the final answer. Reward identification of uncertainty, safe decision making, and proper use of evidence.
These strategies are not about punishment. They are about aligning assessment with what we truly care about in clinical education: safe thinking.
The ethical standard: AI is a tool, not a clinician
I take a firm stance that AI is a tool. It does not replace the human aspects of care or the critical thinking that protects patients. At the same time, a well-used tool can help us notice what we might miss, generate alternative considerations, and support patient safety when it is paired with human oversight.
To make responsible AI use practical for students, I emphasize three expectations whenever AI use is permitted:
- Transparency. Students disclose how they used AI and for what purpose.
- Verification. Students cite primary sources and demonstrate cross-checking.
- Reflection. Students explain what they accepted, what they rejected, and why.
This aligns with the direction of scholarly publishing guidance that emphasizes disclosure of AI use and clear human responsibility for accuracy and integrity.⁶
Conclusion
AI is not going away. If we respond with fear alone, students will still use it, just without guidance. If we respond with structure, honest expectations, and clear boundaries, we can turn AI into a teaching partner that strengthens clinical reasoning instead of weakening it.
For health professions education, the best outcome is not an arms race between student tools and faculty detection. The best outcome is a learning culture where AI use is discussed openly, guided intentionally, and assessed through demonstration of reasoning. That is how we prepare the next generation of clinicians to use emerging tools without compromising safety or trust.
Justin Hooks, DNP, FNP-BC, AAHIVS, is an HIV primary care provider who brings over two decades of diverse and impactful health care experience to the table. From his roots as a critical care paramedic to his roles as a director of quality improvement, registered nurse, and board-certified nurse practitioner, Justin’s journey has been defined by a commitment to excellence in patient care. His dynamic career also extends to the realm of education, where he serves as a college educator, sharing his wealth of knowledge and shaping the next generation of health care professionals.