Step 1: triage before you convert anything
Read the notes once with a single question in mind: could this be examined? Most of a lecture is scaffolding — the argument, the worked example, the anecdote that made the mechanism click. All useful, none of it card material.
What survives triage is usually: definitions, named mechanisms, classifications, numbers, conditions, dates, causes, and pairings. In a one-hour lecture that's typically twenty to forty items. If you're marking a hundred and fifty, you're carding the narrative.
Step 2: atomise
Each surviving fact becomes one question with one short answer. This is where most decks go wrong, so here it is concretely.
| Raw note | Bad card | Good cards |
|---|---|---|
| Beta blockers — block β-adrenergic receptors, lower heart rate and BP, used in hypertension, angina, post-MI; caution in asthma | “Tell me about beta blockers” | What receptor do beta blockers block? · What effect do they have on heart rate? · Which airway condition is a caution for beta blockers? |
| Treaty signed 1648, ended the Thirty Years' War, established state sovereignty principle | “Describe the significance of Westphalia” | Which treaty ended the Thirty Years' War? · In what year? · Which principle of international relations is it credited with establishing? |
The test is mechanical: if you can imagine two people giving different correct answers, it isn't a card yet.
Step 3: phrasing rules that hold up
- One fact per card. An answer with five parts guarantees you recall four, mark it wrong, and re-drill the four you knew.
- No yes/no questions. A 50% chance of guessing right destroys the signal the scheduler depends on.
- Ask for the thing, not about the thing.“What enzyme catalyses X?” not “Discuss the role of enzymes in X”.
- Keep the answer under about eight words.If it won't fit, the card is really two cards.
- Make it context-free.“What did he conclude?” is meaningless three weeks later when the card appears alone.
Step 4: let something else do the typing
Steps 1–3 are judgement, and judgement is the part worth your time. Typing is not. Paste the notes, import the lecture PDF, or photograph the slides, and let an AI produce the first draft of the deck. Then do the only step that matters:
Delete about a third of what it gives you.Generated decks are reliably over-inclusive — they card the scaffolding along with the facts, because a model can't know what your examiner cares about. Five minutes of pruning turns a mediocre generated deck into a good one, and the pruning itself is a first pass of revision.
Why one deck a day is enough
1generation a day sounds restrictive until you try to review six lectures' worth of cards at once. One lecture, pruned to thirty cards, reviewed the same evening, is a pace you can hold for a semester — and holding it is the entire game. See how the scheduling works for why backlog is what kills decks.
Step 5: review the same day
The forgetting curve is steepest in the first twenty-four hours, which makes the first review disproportionately valuable. Cards made on Monday and first reviewed on Saturday are, for practical purposes, cards you're learning from scratch. Ten minutes the same evening buys more retention than an hour the following weekend.
The four mistakes that kill decks
- Too many cards.Volume is the number one cause of abandonment. Forty cards you review beat four hundred you don't.
- Carding what you already know. It feels productive and it wastes every review it consumes.
- Compound answers. The card you always half-remember and never retire.
- Making cards instead of studying. The most seductive failure mode: a beautifully organised deck, built all week, never reviewed.
What this method is not for
Flashcards handle the recall layer. They will not teach you to solve a problem set, construct an argument, or reason through a case. Use them to make the facts automatic so your attention is free for the parts that need thinking — and do the problems separately.