My Patient Brought a ChatGPT Diagnosis: Now What?

Main ideas / lessons conveyed

• Patients increasingly present with diagnoses or suggestions generated by AI tools (e.g., ChatGPT/Gemini).
• Clinicians should respond safely using a consistent structure that:
  • respects the patient’s effort,
  • verifies accuracy against the real clinical context,
  • rules out dangerous/serious causes,
  • and provides a clear plan for next steps.
• A key safety concern: AI output depends on the prompt, which may omit critical context. Also, AI cannot perform exams, order/interpret objective tests, or fully access the verified medical record.
• The clinician should use AI to start the conversation, but not to end the diagnostic responsibility.

Methodology: the “6R framework” (step-by-step, as presented)

1. Recognize

   • Acknowledge the patient’s search and that they used AI.
   • Do not dismiss or invalidate them; simply acknowledge their input.

2. Retrieve

   • Ask for the exact prompt the patient entered.
   • Review what the AI got right vs. what it missed or overstated.
   • Pay attention to whether the patient omitted critical details (e.g., age, medications, surgeries, medical history).

3. Reconstruct

   • Re-do the clinical reasoning independently:
     • take a full history,
     • ask the required questions,
     • perform an exam,
     • obtain vitals,
     • order appropriate tests (e.g., EKG, labs) if indicated.

4. Identify red flags

   • Look for “sensitive risks” and warning signs.
   • Emphasize that a “plausible” explanation is not the same as a safe diagnosis.
   • Example issue highlighted: reflux could be plausible, but exertional symptoms (pain with walking improving with rest) are more concerning for cardiac ischemia/ACS.

5. (Profile evaluation / risk assessment)

   • Conduct evaluation focused on risks and safety based on the verified clinical picture (as opposed to AI-generated possibilities).

6. Roadmap (action plan + monitoring + escalation)

   • Provide a clear, patient-facing plan for:
     • what to do today,
     • what symptoms to monitor,
     • when and how to escalate care,
     • and when to refer or send the patient to urgent care/emergency care or cardiology.
   • The goal is a structured path grounded in verified clinical data.

Case example used in the video

• Scenario: Patient reports burning chest pain and says ChatGPT suggested it could be GERD/heartburn.
• Crucial missing context: The patient’s pain also occurs with walking and improves with rest—a pattern more characteristic of anginal/ischemic pain.
• Takeaway: AI may suggest reflux based on limited prompt information, but clinicians must verify and rule out ACS.

Limitations of AI emphasized

• AI can:
  • organize symptoms into questions,
  • prompt seeking professional care,
  • provide general health education.
• AI cannot:
  • examine the patient,
  • obtain/interpret objective testing,
  • access the complete verified medical record reliably (and uploading personal medical data into ChatGPT is likely not advisable as stated).
• Accountability point: AI cannot accept clinical responsibility for a misdiagnosis—clinicians remain responsible to ensure safety.

Closing / recommendations

• Share the video with colleagues because these cases are becoming more common.
• Use the structured framework to maintain a strong patient relationship and reinforce evidence-based practice.
• Non-sponsored educational intent.

Speakers / sources featured

• Dr. Ash (speaker; hospitalist in Florida, internal medicine core faculty)