Anki Cloze Cards for Language Learning
A cloze card hides one word inside a sentence you can read, which is closer to using a language than translating a word pair. This guide covers the Cloze note type, the syntax, the import traps that catch cloze specifically, and how to get a deck that already speaks. For the method rather than the mechanics - what to hide, and when a plain word pair is the better card - start with cloze cards for language learning.
Cloze cards step by step
Know what the Cloze note type actually is
- A Cloze note has two fields, Text and Back Extra, and Anki generates one card per marker found in Text. A note with three markers becomes three cards, each hiding a different word, each with its own schedule.
- The card template renders {{cloze:Text}}, which is what turns the marker into a blank. A Basic template has no such instruction, so the same note shows the raw braces on the card.
- This is why the note type, not the file, decides whether your import works. Text that looks perfect in a spreadsheet renders as literal syntax under the wrong note type.
- A native .apkg carries the note type inside it. A CSV does not, which is the entire source of the trouble below.
Write the syntax
- {{c1::casa}} hides 'casa' and leaves the rest of the sentence as context. The numbering is the card number, not the order of importance.
- Adding {{c2::...}} to the same sentence makes a second card from the same note. Useful deliberately, painful by accident: two markers double the daily reviews for that sentence.
- {{c1::casa::house}} adds a hint shown in place of the blank. It helps on grammar cards and gives the answer away on vocabulary cards.
- Braces cannot nest and the marker cannot be empty. Both produce a note Anki refuses to render as a card.
- NextLang generates exactly one c1 marker per sentence. That keeps one note equal to one card, so the number of cards you asked for is the number you get.
Generate the deck
- Every sentence is written in the language you are learning, with one word hidden. The translation of the whole sentence goes to Back Extra, so the answer side explains the sentence, not just the word.
- Up to 30 sentences per generation, at the CEFR level you set. The level controls the vocabulary and the sentence, not just the hidden word.
- Field selection, the header row and the separator disappear from the form: a Cloze note has fixed fields, and the export is tab-separated with no header for a reason explained below.
- Sentences that come back with a broken or missing marker are dropped before the deck is saved, so a malformed card never reaches your collection.
Download the .apkg
- Double-click the file, or use File then Import. The blanks work immediately because the template came with the notes.
- If you export the text instead, it is tab-separated on purpose, and you must select Cloze as the note type in Anki's import dialog yourself.
- The deck name, the tags and the fields are already set. Nothing about the import dialog needs to be right, because you are not using it.
Let the cards speak
- The recording is of the sentence without the markers, so you hear the sentence as it is spoken, complete, including the word you are being asked to recall.
- Nothing has to be installed in Anki. No AwesomeTTS, no add-on, no network connection while you review.
- The audio sits on the answer side by default. Moving it to the question turns the deck into listening practice: edit the card template once and every note follows.
- The Include audio toggle in the download menu is shared with every other export. Off, the deck is a couple of kilobytes; on, it is a few hundred.
Check the deck before you commit to it
- Open Browse and look at ten cards. The blank should sit on the word being taught, and the rest of the sentence should give you a fair chance of recovering it.
- Run Tools then Empty Cards. It reports notes that produce no card, which for cloze means a note whose Text lost its marker somewhere.
- If a sentence gives the answer away in the surrounding words, edit the note rather than deleting it: the sentence is usually fine with a different word hidden.
Import traps specific to cloze
Cloze decks fail in ways word-pair decks do not, because a cloze card is a sentence and a sentence contains punctuation. Find the symptom, and the cause is next to it.
This is the trap that catches everyone, and it is specific to cloze. Word pairs rarely contain a comma; natural sentences contain them constantly. Import 'Cuando llegué, la casa {{c1::estaba}} vacía' from a comma-separated file and Anki reads it as two fields, so half the sentence lands in Text and half in Back Extra. Use tab-separated text. NextLang forces tabs on every cloze export for exactly this reason.
The notes were imported under a Basic note type. Only a template containing {{cloze:Text}} turns a marker into a blank. Fix it in Browse: select the notes, Notes then Change Note Type, pick Cloze, and map Front to Text and Back to Back Extra. Importing the .apkg instead skips this entirely.
A note whose Text field has no valid marker generates no card. Usually the marker was mangled on the way in: a stray line break, a broken brace, or a spreadsheet that helpfully autocorrected the braces. Tools then Empty Cards lists them; fix the note or delete it.
The note contains c1, c2 and c3. That is cloze working as designed, and it triples the reviews for that sentence. It suits a grammar drill where each slot is worth its own card and works against you on vocabulary, where you end up reviewing one sentence three times a day.
A sentence like 'Fui al {{c1::supermercado}} a comprar comida en el supermercado' answers itself. It is the most common quality problem in generated cloze decks and the reason cards are shown on screen before export - edit the sentence or hide a different word.
Any text-based import treats a newline as the end of a record, so a sentence containing one becomes two broken notes. Generated decks never contain line breaks inside a value; hand-built files often do, especially when the text was pasted out of a PDF.
The same deck in Mochi
Mochi has cloze too, with a different syntax: double braces around the hidden text, no numbering. The same sentence that reads la {{c1::casa}} es grande in Anki reads la {{casa}} es grande in Mochi.
Converting between them by hand is where decks get damaged, so NextLang writes each one natively instead: pick Mochi in the generator and the sentences come back in Mochi's syntax, packed into a .mochi file with the audio inside it and no Mochi Pro subscription required.
The Mochi markdown and cloze guide covers that side in detail, including what Mochi's Markdown understands and where it differs from Anki's templates.
Studying on both is a reason to generate twice rather than convert once. The all-in-one generator keeps your text, languages and card count while you switch platforms.
When cloze is the right card, and when it is not
Cloze is fashionable enough that people build entire collections out of it, including for material where a word pair would have been faster. These are the distinctions that matter in practice.
Cloze wins where the word only means something in context: verbs with prepositions, collocations, cases and aspect, words whose translation is a range rather than a match. A word pair wins for concrete nouns, numbers and closed sets, where the mapping really is one to one and a sentence is just extra reading. A deck of a hundred cloze sentences about kitchen objects is slower than a hundred word pairs and teaches no more.
Multiple markers on one note look economical - one sentence, three cards - but the schedules diverge immediately, and you end up seeing the same sentence on three different days with a different word missing each time. The context stops being new after the first pass, so cards two and three test recall of a sentence you have half memorised. One marker per sentence keeps each card independent and keeps your count honest.
The best sentence for a cloze card is one you have already met. That is why sentence mining exists: subtitles from something you watched, a paragraph from an article you read. Import the subtitle file into your vocabulary, keep the words worth learning, and use those words as the prompt for a cloze deck - the sentences come back in the same register as the source rather than in textbook neutral.
A cloze card sits between the two. You are not translating from your own language, so it is not full production, but you are retrieving a word rather than recognising one, which is more than a word pair asks. Pair it with the audio on the question side and you add listening to the same note without creating a second card. This is the cheapest way to get three skills out of one deck.
Mochi has its own cloze syntax, double braces around the hidden text with no c1 numbering, and NextLang generates it natively - the same sentences, the same audio inside the .mochi package, and no Mochi Pro required. If you study on both, generate once and export twice rather than converting a file between two syntaxes that do not quite match.
Sentences worth hiding a word in usually come from something you watched or read. The guide on turning subtitles into an Anki deck covers that route end to end.
Premium access includes:
Frequently asked questions
No, if you import the .apkg: the package carries the note type with it. Yes, in effect, if you import text - you must select Cloze in the import dialog, and Anki ships that note type by default, so nothing has to be created.
Because cloze cards are sentences, and sentences contain commas. A comma-separated file splits a sentence into two fields at the first comma. Tabs practically never appear inside a sentence, so a tab-separated file survives the trip intact.
Anki supports it with c2, c3 and so on, each producing its own card. NextLang generates exactly one marker per sentence, so the card count stays predictable and no sentence comes back to you three times. Add a second marker by hand if a particular sentence is worth it.
Yes. FSRS schedules each cloze card individually, exactly like any other card in your collection, from your first answer onwards. Nothing special is required at export or import time.
Not by converting the file: a word pair has no sentence to hide anything in. Generate a cloze deck on the same topic instead, or use the words as the prompt, and the sentences are written around them.
Keep reading
Related guides that build on what you just read.