Every USMLE study guide tells you to “make your own Anki cards” — and most of us spend the first week dutifully re-typing First Aid into the editor, get bored, and silently switch to a pre-made deck by week three. The problem isn’t laziness. It’s that the friction of card-making is higher than the friction of just re-reading the chapter. That math has to flip if Anki is going to work for you.

This post walks through the exact workflow we use to turn a chapter PDF into 40-60 high-quality cards in under an hour, with AI generating the first draft and you keeping the editorial pen.

What we’ll need

  • Your source PDF (lecture slides, a First Aid chapter, a Pathoma transcript — anything).
  • Anki, 2.1.50 or later, on Windows / macOS / Linux.
  • Cardivate installed (free trial covers the first batch).
  • About 45 minutes.

Step 1 — pick a tight slice of source material

The single biggest mistake is feeding the AI a 200-page chapter at once. Card quality drops as soon as the source goes beyond ~15–20 pages because the model has to summarize too aggressively. Smaller slices yield denser cards.

For Step 1 prep, a good slice is roughly:

  • One Pathoma chapter, or
  • One major topic in First Aid (e.g. “Renal pharmacology”), or
  • Two to three pages of dense lecture notes on a single concept.

If your source is huge, just open it in any PDF viewer and use Print → Save as PDF with a page range to make a slice file.

Step 2 — load the slice into Cardivate

Open Anki, then Cardivate → 📄 Generate from PDF…. Click the Open PDF… button (it remembers your last folder, so chapter 2 picks up where chapter 1 left off) and load your slice.

Pick the card types you want. For Step 1 work I’d suggest:

  • Multiple Choice for high-yield fact recall (10–15 per chapter)
  • Cloze Deletion for definitions and lab values
  • Image Occlusion for any anatomy or flow-chart figures

Skip summary — for exam prep it tends to over-condense. Skip free-text short answer unless you specifically want to drill recall under time pressure.

Step 3 — review the preview before you commit

Cardivate shows you every card before it lands in your deck. Two minutes spent in the preview pane is the difference between a clean deck and a deck full of corrections you’ll have to make six months from now during a review session.

What to look for:

  1. Generic questions — “What is the function of the liver?” is useless. Reject or rewrite.
  2. Questions with the answer in the question — common AI failure mode. Reject.
  3. Cloze cards where the gap is too short to be testable — “The {{c1::heart}} pumps blood” tests nothing. Reject.
  4. Image occlusion boxes that landed on the wrong word — happens occasionally with very dense pages. Click the card in preview and re-position the box, or just reject and let AI try again.

Step 4 — use the Refine button instead of editing one card at a time

The preview has a ✨ Refine with AI button. Use it. If a batch of MCQs is too easy, hit Make harder. If a set of cloze cards has multi-line answers, Shorten. One refine pass costs 1 credit and revises every selected card.

This is the single biggest workflow improvement over the old “edit cards manually after import” approach.

Step 5 — let the autotag do the chapter labelling

Toggle Auto-tag on before generating. Cardivate reads the source page heading and adds cardivate::topic::<chapter-slug> to every card. Later, when you want to grind only renal pharmacology, you can filter by that tag in the Browse window — or feed it straight into a Mock Exam.

Step 6 — when exam week comes, drill with Mock Exam

This is where the prep workflow pays off. Cardivate → 📝 Mock Exam… lets you run a timed, randomly-sampled exam over any deck — with filters for topic, difficulty, and (importantly) “focus on cards I’ve lapsed on”. The lapsed-card filter is the killer feature: it surfaces exactly the material you’ve been quietly forgetting.

If you want to simulate the real test, turn on negative marking (matches USMLE-style scoring) and confidence-weighted scoring (trains calibration — being right but unsure is worth less than being right and sure).

A realistic time budget

For a single 15-page chapter, our typical run is:

  • Slice the PDF: 2 min
  • Cardivate generates: 3 min
  • Preview + refine: 20 min
  • Mark wrong-answer review: 10 min (after first day of practice)

That’s ~35 minutes for a chapter’s worth of high-quality drilling material. Compare to the “type each card by hand” workflow which is typically ~2 hours per chapter.

One last thing

Cards are a tool, not the goal. If you find yourself spending more time making cards than reviewing them, the system is broken. The right ratio is roughly 20% creation, 80% review. AI helps the 20% become 5%, which gives you 95% on the part of studying that actually moves the needle.

Good luck with Step 1.