There is no shortage of internal medicine content online. What is missing is curation.

A student on their IM rotation is handed a list of objectives and pointed at everything — 290 episodes of Intern at Work, our internal medicine podcast, with no way of knowing which three to listen to tonight. The IM Playbook makes that decision for you: one topic a day, matched to the day you are having. It is free, with nothing to sign up for.

Built around a real rotation

This is designed first for the medical student on their core internal medicine rotation. We looked at how long that rotation actually runs at Canadian medical schools and sized the curriculum to fit inside it — a curriculum you cannot finish is one you abandon in week two. The order is a suggestion, not a rule: if you are already three weeks in, go straight to the topic in front of you.

It is not a reference. UpToDate and its equivalents answer a question you already have; they cannot tell you what you do not yet know to ask. This is a curriculum — it decides what is worth learning tonight, curates across podcasts, guidelines and landmark trials rather than replacing them, and links you to the primary source. Use both.

Residents — the second layer of every topic is yours: guideline rationale, dosing nuance, and the trial evidence you want when someone asks you to justify the plan on rounds, with landmark trials linked to our Trial Files. Attendings — teach from it: assign the topic your team is about to admit, or use a reflection prompt as a ready-made discussion opener. Outside Canada — welcome; every guideline is linked so you can compare against local practice.

How a topic works

Three layers, separated by when you would actually read them.

Why each topic ends with a reflection

Seeing a case does not automatically teach you anything; being made to stop and articulate why you did what you did is what converts it into something you keep. Structured, deliberate reflection has one of the strongest evidence bases for improving diagnostic accuracy in learners. Each prompt is built around a real clinical decision, with a sample answer so you can calibrate rather than wonder whether you were close.

Why we ask questions at all

Re-reading feels like learning and largely is not: retrieving an answer yourself produces substantially better long-term retention than studying the same material again. That is why questions appear twice — after the topic, and again days later in the week review. They are learning events, not exams: nothing is scored against you, which is exactly why every option gets an explanation, not just the correct one.

Where the content comes from

Every topic starts from a guideline, not from anyone’s memory. The process is deliberately boring:

1. Scope the topic against core clerkship and Medical Council of Canada objectives, then against what actually walks onto a GIM ward.
2. Pull the current source documents — national guidelines where they exist (Canadian Cardiovascular Society, Hypertension Canada), international ones where they do not, plus the relevant episodes and landmark trials.
3. Draft from those documents, with the guideline text in front of the model rather than behind it.
4. Check every number back to its source — thresholds, doses, targets, trial names. Anything untraceable gets cut, not softened.
5. Physician review and edit before anything publishes.

Why that matters — retrieval, not recall

Ask a large language model a clinical question cold and it answers from what it absorbed in training: knowledge frozen at some point in the past, blending good sources with bad, unable to tell you where any claim came from. That is where models produce something that sounds exactly right and is quietly wrong — a plausible dose, a threshold that moved two guidelines ago.

We use retrieval-augmented generation instead: the current guideline is placed in front of the model and the draft written from that document. Claims stay traceable to a source we chose, content moves when the guideline moves rather than when the model is retrained, and a reviewer can check the draft against the same document. It is the difference between asking someone what they remember and asking them to read the guideline out loud and explain it.

How the questions are written

Every question is a clinical vignette testing whether you can apply knowledge to a patient, not recall a fact. We wrote to the published literature on medical MCQ design — the NBME item-writing guide, Haladyna, Steele, and Downing on what item flaws do to students: they make a test harder for the people who actually know the material.

• The cover test. You should be able to answer with the options covered. If not, we rewrite the stem.
• No filler distractors. Every wrong option is a mistake someone has genuinely made on a ward.
• Every option gets explained, with a link back to the exact section to reread.
• No pattern to exploit — answers spread evenly across positions, no negative stems, no “all of the above”, no answer clued by being longest.
• Difficulty is hidden until you answer, so you never talk yourself out of one badged advanced.
• Canadian context: SI units, Canadian practice and thresholds.

Who writes it, and who checks it

We use an artificial intelligence tool — a large language model — to draft and curate. Specifically Claude, made by Anthropic. A large language model is software trained on enormous amounts of text that can read documents and write prose; it is not a doctor, it does not reason clinically, and it has no way of knowing whether what it produces is true.

What it is good at is the work that is enormous and mechanical: reading across hundreds of episodes and full guideline documents, mapping them onto clerkship objectives, and drafting from the sources we hand it. That curation — deciding what belongs on day three of an IM rotation and what does not — is what makes this different from a search box.

Every topic and every question is then reviewed by a physician — an internal medicine attending and a subspecialty fellow read the clinical content, the questions and the explanations before anything appears. Content is reviewed at least annually against current guidelines, and each topic carries a last-updated date. If you spot something stale or simply wrong, tell us.

What we collect

The Playbook is also a medical education research project — we study how trainees actually use and learn from self-directed clinical content. When you answer a question or write a reflection we record your answer, whether it was correct, how long it took, whether it is a repeat attempt, and your screen width. There is no name, no email and no account — just a random identifier for your browser, so we can tell a first attempt from a third. Reflections are free text, so please do not include anything that could identify a patient.