Most learners blame the app when their flashcards stop working. Usually the app is fine. The problem sits on the card itself: it asks five questions at once, hides the answer inside a paragraph, or asks something you could guess from the layout. The highest-leverage thing you can do before reviewing a single card is to fix what is on it.
This is a general method. It works whether you write cards by hand, generate them, or download a shared deck — the card is the unit that decides whether a review is worth your time.
One card, one fact
The core rule is atomicity: each card should test exactly one thing. A card that asks for the meaning, the pronunciation, the gender and three translations is really four cards wearing a trench coat. When you fail it, the scheduler cannot tell which part you failed. When you pass it, you cannot tell which part you actually know.
Split it:
- “What does la llave mean?” — meaning
- “What is the gender of llave?” — gender
- “Which article pairs with llave?” — usage
Each card now passes or fails on its own, and FSRS can schedule each one at the interval it deserves. Hard facts come back sooner; easy ones disappear for months.
What belongs on a card
A good vocabulary card carries a small, fixed set of fields. Everything else is decoration.
| Field | Why it earns its place | |
|---|---|---|
| The word or phrase | The prompt | |
| A short definition | The answer, in plain language | |
| An example sentence | Shows the word in the grammar it lives in | |
| A translation of the sentence | Removes ambiguity | |
| Pronunciation | IPA, pinyin with tone marks, romaji, or the equivalent | |
| Part of speech | Stops you learning a noun as a verb | |
| Common collocations | Teaches the pairings real speakers use |
For most words that is enough. A card with these fields takes seconds to build and does the job.
What to leave off
- Paragraphs. If the answer is longer than a sentence, you are testing reading comprehension, not memory.
- Lists of synonyms. “big, large, huge, enormous, vast” teaches none of them, because you never have to choose one.
- Everything you know. A copy of the dictionary entry feels thorough and reviews terribly.
- Facts you can guess. If the answer is visible in the prompt, the card tests nothing.
Make the answer harder than recognition
Recognition — “do I recognise this word?” — is the weakest test. You can recognise a word for years and still never produce it. Two habits fix that:
- Ask in both directions. Word to meaning, and meaning to word. Production is the skill you actually need in conversation.
- Use cloze. Hide the target word inside a real sentence and let context do the work: “Ella ___ la puerta cada mañana.” (abre / opens).
Fields to add, by language
The fields worth adding depend on what makes a language hard to hold on sight.
| Language type | Add onto the card |
|---|---|
| Tonal (Chinese, Vietnamese, Thai) | The character, its romanisation with tone marks, and an audio clip |
| Non-Latin script (Arabic, Cyrillic, Hangul, kana) | The written form alongside a transliteration, so you can read it before you can sight-read it |
| Gendered nouns (Spanish, French, German) | Gender on every noun card, learned with its article |
| Heavily conjugated (Spanish, Japanese) | One card per key form, not one card for the whole table |
| Irregular spelling (English) | Pronunciation on the front, not buried in the back |
The pattern is the same everywhere: add the field your memory keeps dropping, and nothing more.
Build in batches, but review the output
If hand-building cards is the step you keep skipping, generating them is a fair shortcut — the point is to get from a word list to a usable deck without spending the whole evening on it. But generated is not finished. Read the cards, split the ones that clump several facts together, and rewrite any example sentence that sounds like a textbook. Those ten minutes of editing are what make the next hundred reviews worth doing.
Who this method is not for
If you are studying a fixed list for an exam next week, atomic cards and long intervals work against you — you are better off cramming the exact list you will be tested on. Atomic cards pay off over months, not days.
For everyone else the sequence is simple: build cards that test one thing, add only the fields you need, and let the scheduler decide the rest. To see how the review side works, start with our guide to spaced repetition for vocabulary, then look at vocabulary flashcards for the review loop itself. If making the cards is the part that keeps stopping you, see how AI card generation turns a word list into a finished deck.
Download EnjoyCards for free and try the method on your next ten words.