Think Like a Doctor: Work a Diagnostic Case
Maps to: Physician · Nurse, Physician Assistant, Emergency Medicine, Medical Researcher, Public Health
You're going to work a realistic but completely made-up patient case end to end: take a history from an AI playing the patient, build a ranked list of what it could be, pick the one test that would settle it, and make the call. The skill is diagnostic reasoning under uncertainty: figuring out what's most likely, what would change your mind, and what to do right now, when you can't be sure. That's what a doctor's mind actually does, and doing one tells you fast whether reasoning toward a call without certainty is your kind of work. Important: these are fictional practice cases for learning to reason, never medical advice and never for diagnosing a real person.
The plan
0/4 doneYou're 20% in just for starting, the hardest part. Mark your first step done to keep the momentum.
Pick a common-but-ambiguous complaint: chest pain, a bad headache, stomach pain, shortness of breath. Have the AI generate a FICTIONAL patient case and play that patient for you. Then do the first thing doctors actually do: write a 'problem representation' (one tight sentence summarizing who this patient is and what's going on) plus your first-guess list of what it could be. (These cases are made up, for reasoning practice, never real advice.)
Objective: A synthetic case in hand, a one-sentence problem representation, and a first-guess differential.
- 1
Ask the AI to generate a realistic but FICTIONAL case for reasoning practice and to role-play the patient. Use a prompt that locks in 'fictional' and 'not medical advice.'
Tool: Claude / ChatGPT / Gemini
Generate a realistic but completely FICTIONAL patient case for diagnostic-reasoning practice (this is a learning exercise, NOT medical advice and not about any real person). Give me only the opening line: age, sex, and the main complaint (e.g. chest pain). Then role-play the patient: answer my history questions in character, but do NOT tell me the diagnosis or hint at it until I ask you to reveal it.
- 2
From the opening line, write your problem representation: one sentence in plain clinical terms (who + key features). Then jot your first 3 guesses.
Your call
Pick the presentation type and write your problem representation (one sentence) and first three guesses, yourself, from the opening line.
Your one-sentence problem representation + your first three guesses.
What good looks like: You've got a synthetic case in hand and a one-sentence problem representation plus three first guesses, written from the opening line before the AI revealed anything.
- A sharp one-sentence summary is half the reasoning. 'Older smoker with sudden crushing chest pain' already points somewhere.
- Don't peek. The AI knows the 'answer'; your job is to reason toward it, not get it told to you.
- 1
The bar to look back against
At least one synthetic case worked end to end: a problem representation, a ranked differential with your reasoning for each, the one test that would change your mind, and a defensible triage call, all committed BEFORE the AI revealed anything, plus a published write-up. The reasoning is the work: not 'I guessed the right disease,' but 'my reasoning holds up even where I turned out to be wrong.'
Finish the final step, then submit what you built. Your progress is saved.
Tools you'll use
Step 1 · Get a case and frame the problem
Step 2–3 · Commit your workup, before the reveal
How this shows up on a resume or college app
I worked through clinical cases the way medical training teaches diagnostic reasoning, building a problem representation, ranking a differential diagnosis with my reasoning, choosing the test that would change my mind, and making a triage call under uncertainty, then published my written workups (on fictional practice cases). I learned that good clinical reasoning is judged by the quality of the thinking, not just whether the final label turns out right.