How to Write Flashcards That Actually Stick
Why most flashcards fail: the minimum-information principle, one-idea-per-card rewrites, front/back direction, and what never belongs on a card.
The minimum-information principle
The single most-cited rule in flashcard design comes from Piotr Wozniak’s formulation guidelines for SuperMemo: items should be as simple as possible — the “minimum information principle.” A card should test exactly one retrievable unit. Not because simple is lazy, but because of how grading works: a card gets one grade, so a card testing three facts forces you to average “knew two, blanked one” into a single button press. The scheduler can’t tell which third you lost, so it schedules all three wrong — too soon on the two you know, too late on the one you don’t.
This is why atomic cards feel almost trivially small when you write them, and why that’s correct.
Bad card → good cards
Bad: Front: “Tell me about the French Revolution.” Back: a paragraph. — tests nothing retrievable; you’d grade vibes.
Good: split into retrievable units —
- “In what year did the French Revolution begin?” → 1789
- “What event on 14 July 1789 is commemorated by Bastille Day?” → the storming of the Bastille
- “Which king was executed during the French Revolution?” → Louis XVI
Bad: “Spanish verb endings” → a six-row conjugation table. — six facts hiding in one card; you lapse the card, re-read the table, and still don’t own the weak ending.
Good: one card per conjugation pair, or better, per pattern with an example — “hablar (to speak): yo form, present tense → hablo”.
Bad: “Photosynthesis” → “the process by which plants convert light energy into chemical energy, using chlorophyll in chloroplasts to fix CO₂ into glucose while releasing oxygen.” — the back is four answers stapled together.
Good: “What gas does photosynthesis release?” → oxygen, “Where in the cell does photosynthesis occur?” → the chloroplast, etc.
Rules that survive contact with real decks
- Understand first, card second. A card schedules retrieval; it can’t create comprehension. If you can’t explain the answer in your own words, the card is premature — read the source, then card the residue.
- One fact per card. If the back contains “and”, commas of enumeration, or “also”, suspect two cards.
- Make the front a real prompt. “The Krebs cycle” isn’t a question; “what does the Krebs cycle produce directly?” is. A prompt tells your brain what shape of answer to retrieve.
- Keep the back minimal. The answer, optionally plus one clarifier. Every extra clause is one more place a fuzzy memory can quibble with itself during grading.
- Prefer concrete cues over abstract ones. “Capital of Australia?” works; “Discuss Australia” doesn’t. Numbers, names and single-fact definitions card beautifully; essays don’t.
- Both directions are separate cards. EN→ES and ES→EN are different skills; add both only if you need both.
- Use context when items look alike. If “ser” and “estar” keep colliding, a cue like “to be (permanent trait)” disambiguates at retrieval time — that’s why the sample deck phrases them that way.
- Delete freely. A card you grade Easy ten times is tax, not study. FlashDeck’s scheduler stretches easy cards into the far future anyway, but pruning dead weight keeps sessions honest.
When flashcards are the wrong tool
Retrieval practice is for retrievable material: vocabulary, capitals, formulas, anatomy terms, conjugations, API signatures, dates, theorem statements. It is weak for skills that live in performance — writing proofs, debugging, designing, arguing. Card the vocabulary of a skill, not the skill itself; then practice the skill.
A quick self-test for any card you’re writing
Read only the front. Could you answer it in two seconds, with exactly one defensible answer, on a cold morning three weeks from now? If yes — ship it. If you hesitated, the card is probably two cards wearing a trenchcoat.
Frequently asked questions
How long should one card be?
Short enough that the answer is recallable in a couple of seconds. A good rule: if the back needs more than a line or two, it's two cards. The front can carry context ('In photosynthesis, …') but the tested unit stays atomic.
Should I make cards in both directions?
For vocabulary and pairs you genuinely need in both directions, yes — two separate cards (Spain→Madrid and Madrid→Spain are different retrievals). Recognition (capital → country) is much easier than production (country → capital); don't assume one covers the other.
Is it better to have few big cards or many small ones?
Many small ones, almost always. Twenty atomic cards survive lapses individually — you re-learn the two facts you actually forgot instead of re-reading a wall of text where the forgotten bit hides.
Can I put images or audio on cards here?
FlashDeck's card format is text-only (with a small markdown subset: **bold**, *italic*, `code`) so CSV export stays lossless. If a subject truly needs images — anatomy plates, flags, maps — a text cue can still work ('describe the parts of…'), or Anki-family apps handle media decks.
What do I do with a card that keeps failing?
Don't just re-drill it — rewrite it. Three-plus lapses usually means the card tests several facts at once, tests something you never actually learned, or lacks a retrieval hook. Split it, add a concrete cue, or delete it and learn the idea properly first.