Clinical Learning with AI

Best Practices for UCSF GME Learners & Faculty in the Clinical Learning Environment

USE AI TO DEEPEN CLINICAL THINKING — NOT TO REPLACE IT.

PURPOSE

Generative AI can support curiosity, individualized learning, clinical reasoning, and discussion with faculty. Used passively or uncritically, it can also reduce opportunities to retrieve knowledge, work through uncertainty, develop independent reasoning, and engage with teachers and colleagues. These best practices are intended to help residents and faculty use AI in ways that strengthen learning and human interaction. They supplement — and do not replace — UCSF GME, departmental, program, and clinical-site policies.

A PRACTICAL LEARNING CYCLE

Five steps: 1. THINK: Reason first; 2. ASK: Extend thinking; 3. CHECK: Verify claims; 4. DISCUSS: Bring it to people; 5. APPLY: Return to patient.
  1. THINK | Reason first

    Do the cognitive work first. Before asking AI, generate your own differential, explain the physiology, predict findings, or commit to an assessment and plan.

  2. ASK | Extend thinking 

    Use AI to challenge or elaborate your reasoning: “What am I missing?” “What argues against my leading diagnosis?” “Compare these mechanisms.” “Quiz me on this case.”

  3. CHECK | Verify claims

    Treat output as a hypothesis, not an authority. Verify consequential claims with guidelines, primary literature, trusted clinical resources, the medical record, and/or faculty expertise.

  4. DISCUSS | Bring it to people

    Bring an uncertainty or disagreement back to the team: “I thought X. AI suggested Y. I’m not convinced because… What do you think?”

  5. APPLY | Return to patient

    Return to the actual patient. Decide whether generalized information fits this patient’s history, exam, trajectory, preferences, context, and epidemiology.

PROTECT LEARNING: WHEN TO DELAY AI

Preserve moments of “productive struggle.” Work independently first when developing an initial differential, synthesizing an assessment/plan, explaining pathophysiology, practicing retrieval, preparing to defend a clinical decision, or completing an assessment intended to represent your own knowledge or skills.

SELF-CHECK: After using AI, can I explain the reasoning myself, defend it to my team, identify its uncertainties, and apply it specifically to my patient?

FACULTY: TURN AI USE INTO TEACHING

Reason before AI 
“Before we look it up or ask AI, what do you think?”

Critique the output 
Ask what is correct, incomplete, biased, or misleading.

Compare with evidence 
Why does — or doesn’t — this recommendation apply to this patient?

Challenge the team 
After learners commit, ask AI for competing diagnoses or counterarguments.

Model verification 
Show how an experienced clinician checks an authoritative-sounding response.

Normalize transparency 
Ask how AI was used and what the learner accepted, rejected, or learned.

DEFT-AI: An approach to promote critical thinking among learners when using AI. Take it to the next level and use Discussion-Evidence-Feedback-Teaching to improve learners’ use of AI. See https://meded.ucsf.edu/resource/diagnosis-evidence-feedback-and-teaching-deft-ai and https://doi.org/10.1056/NEJMra2503232 to learn more.

RESPONSIBLE USE & PATIENT SAFETY

Residents and faculty remain responsible for the accuracy and appropriateness of AI-assisted work. UCSF GME policy requires users to understand limitations and bias, verify generated content, protect confidential information, follow attribution expectations, and comply with the policies of the applicable program, department, campus, and clinical site.

CLINICAL CARE

AI use in active clinical care — including EHR documentation — is limited to tools approved by the individual health system.

DATA & PRIVACY

Do not put sensitive data into unapproved commercial AI tools. UCSF-approved tools and permissions do not automatically extend to affiliated institutions.

SITE-SPECIFIC RULES

UCSF Health, ZSFG, SFVAMC, and other affiliates may have different policies. Follow the policy of the site whose data or clinical environment is involved.

ASSESSMENT

AI is prohibited when learners are being evaluated as representing their own knowledge or skills unless faculty explicitly permit its use.

GOOD USES FOR LEARNING

  • Generate questions that test understanding rather than simply reveal an answer.
  • Ask for alternative hypotheses, mechanisms, counterarguments, or missing considerations after committing to your own reasoning.
  • Practice explaining a concept at different levels — then check the explanation against a trusted source.
  • Use AI to identify knowledge gaps or search terms for subsequent evidence review.
  • Create a teaching question or uncertainty to discuss with a senior resident, fellow, pharmacist, consultant, or attending.

USES THAT CAN UNDERMINE LEARNING

  • Generating a differential or assessment/plan before independently working through the case.
  • Copying an explanation, plan, or documentation that you cannot independently explain and defend.
  • Accepting references, recommendations, calculations, or factual claims without verification when they matter clinically.
  • Using AI during team teaching in a way that disengages from the patient, discussion, or supervising clinician.
  • Allowing a generalized AI response to override patient-specific context or professional judgment.

KEY REFERENCES & UCSF RESOURCES

  1. UCSF Office of Graduate Medical Education. Artificial Intelligence (AI) Usage Policy. Approved July 20, 2026. https://meded.ucsf.edu/policies-residents-fellows#artificial-intelligence-ai-usage-policy
  2. AI@UCSF. Considerations for Using Generative AI. https://ai.ucsf.edu/considerations-using-generative-ai
  3. UCSF Medical Education. Use ChatGPT Effectively. https://meded.ucsf.edu/use-chatgpt-effectively
  4. AAMC. Principles for the Responsible Use of Artificial Intelligence in and for Medical Education. Version 2.0, 2025. https://www.aamc.org/about-us/mission-areas/medical-education/principles-ai-use
  5. Triola MM, Rodman A. Integrating Generative Artificial Intelligence Into Medical Education: Curriculum, Policy, and Governance Strategies. Academic Medicine. 2025;100(4):413–418. https://doi.org/10.1097/ACM.0000000000005963
  6. Masters K, Herrmann-Werner A, Festl-Wietek T, Taylor D. Preparing for Artificial General Intelligence (AGI) in Health Professions Education: AMEE Guide No. 172. Medical Teacher. 2024;46(10):1258–1271. https://doi.org/10.1080/0142159X.2024.2387802
  7. Abdulnour RE, Gin B, Boscardin CK. Educational Strategies for Clinical Supervision of Artificial Intelligence Use. N Engl J Med. 2025 Aug 21;393(8):786-797. https://doi.org/10.1056/NEJMra2503232

AI tools and policies evolve quickly. Review current UCSF GME and clinical-site guidance before using AI with clinical or sensitive information.

Developed by Duncan Henry MD for the UCSF CFE 2026 Licensed under CC BY-NC-SA 4.0: https://creativecommons.org/licenses/by-nc-sa/4.0/

Creative Commons license icon: BY-NC-SA 4.0