Anki for Language Learning: What It's Great At, and the One Place It Breaks
Anki is still the best spaced-repetition tool for language learners, and almost nobody quits because of the algorithm. They quit because of where the cards come from. Here's what Anki does well, the card types that actually work for languages, and how to fix the part that breaks.
Anki is the most powerful free tool a language learner can install, and the reason most people abandon it has nothing to do with the algorithm, the settings, or their discipline. It's the supply line: cards have to come from somewhere, and typing them by hand is a chore that quietly outlasts everyone's motivation. This guide covers what Anki genuinely does better than any alternative, which card types work for languages (and which waste your time), the failure modes that end most Anki streaks, and how the capture problem gets solved in 2026.
If you want the algorithm itself explained first, read FSRS explained — this guide is about using it for a language.
What Anki actually is (and what it costs)
Anki is a flashcard program built around spaced repetition: instead of reviewing everything at a fixed pace, it schedules each card individually so it comes back just before you'd forget it. That's the whole product. Everything else — decks, note types, add-ons — is scaffolding around that one idea.
The cost structure trips people up, so plainly:
- Anki on Windows, macOS, and Linux — free.
- AnkiDroid on Android — free.
- AnkiWeb (browser review + sync between your devices) — free.
- AnkiMobile on iPhone/iPad — a one-time $24.99 purchase. It's the only paid piece, and it funds development of everything else. There's no subscription anywhere in Anki.
Modern Anki also ships FSRS, the machine-learned scheduler that replaced the 1987-era SM-2 formula. You enable it in deck options; it fits its parameters to your own review history. Per Anki's documentation, the practical payoff is the same retention with roughly 20–30% fewer reviews. Its desired-retention setting defaults to 90%, which is the right number for almost everyone — the single most common self-inflicted wound is cranking that to 97% and then drowning in reviews.
The card types that work for languages
Most bad Anki experiences trace back to one decision: what goes on the front of the card.
Isolated word → translation. perro → dog. Fast to make, and the weakest card you can build. You learn to recognize the word in one shape, in no context, with one English equivalent glued to it. Fine for concrete nouns; misleading for verbs, prepositions, and anything idiomatic.
Word in its sentence. The word you're learning, shown inside the actual sentence you met it in, with a translation available. This is the workhorse. Context does two jobs: it disambiguates the meaning (the difference between quedar meaning "to remain" and quedarse meaning "to stay" is visible in a sentence and invisible in a word list), and it gives your memory a hook that isn't just a string of letters.
Cloze (fill in the blank). The sentence with the target word deleted. Excellent for grammar patterns and collocations — which preposition, which verb form — because it forces production of one specific piece rather than recognition of the whole.
Listening cards. Audio on the front, meaning on the back. Underused and disproportionately valuable, because the skill that actually fails in conversation is hearing the word at native speed, not reading it. If your cards are text-only, you're training a skill you'll rarely need.
The practical rule: one card per thing you're actually trying to learn, and prefer sentences over words. A deck of 400 sentence cards you personally met is worth more than 4,000 words someone else chose.
Which brings us to shared decks
AnkiWeb hosts thousands of shared decks, and the top-2000-words deck for your language is genuinely a reasonable way to start from zero. Use one for the first few hundred words, where the vocabulary is so common that any list converges on the same items.
After that, shared decks stop working, for a reason that isn't obvious: the value of a card is mostly in the fact that you met the word. A word you hit in a video you chose to watch is a word that recurs in the content you consume, tied to a moment you remember. The same word in position 3,412 of a frequency deck is a stranger. Learners who plateau on shared decks are usually reviewing thousands of items they have no relationship with — which is also why the reviews feel like a chore instead of recognition.
The four ways Anki streaks die
1. The card supply dries up. This is the big one, and it's mechanical, not motivational. To make a good sentence card by hand you switch to another window, type the sentence, type the translation, look up and paste the dictionary form, maybe rip the audio, then get back to what you were doing. That's 60–90 seconds of clerical work per card, while you're in the middle of watching something. Nobody sustains it. The deck stops growing, then the reviews run out, then the habit is gone.
2. Leeches accumulate. Some cards never stick. Anki eventually flags these as leeches and tags or suspends them, and that mechanism is correct — but only if you act on it. A pile of half-remembered leeches is the single most demoralizing thing in a deck. The fix is unsentimental: rewrite the card (usually it was a bad card — too much on the front, or no context), or delete it and let the word come back to you from real content.
3. Settings anxiety. There's an entire genre of internet content about optimal Anki settings, and it is almost all noise for language learners. Turn FSRS on, leave desired retention at 90%, pick a new-card limit you can sustain on a bad day (10–20 is realistic; 50 is a promise you'll break), re-run the optimizer occasionally once you have a few hundred reviews of history. Done. Time spent tuning is time not spent reviewing.
4. Reviews and input compete. Anki is not the point — comprehension is. If review time starts eating the hour you'd spend watching or reading in your target language, the trade is bad. Reviews should be a five-minute side dish to comprehensible input, not the meal. Cap new cards until the daily load fits in the gaps of your day.
Sentence mining, without the toolchain
Sentence mining is the practice of pulling sentences you almost understand out of real content and turning them into cards — the "i+1" idea, where a sentence has exactly one unknown piece. It came out of the Japanese immersion community (AJATT, then MIA, then Refold) and it works. It's the correct answer to "which cards should I make."
Historically it also meant an evening of setup: a browser extension for subtitles, a pop-up dictionary, a media player fork, AnkiConnect, a note type wired to accept audio and screenshots, and a config file you'd re-fight after every update. That toolchain is why sentence mining stayed a hardcore-only practice while everyone else typed word lists by hand.
It doesn't need to be a toolchain anymore, which is the entire reason LingoKeep exists. You watch a YouTube video with dual subtitles, click a word or drag across a phrase to look it up, and press one key. The card is created with the word, its dictionary form and reading, the full sentence, the translation, and a timestamp back to the exact second of the video — no window switching, no typing, no config. Then it comes back on an FSRS schedule for free, in the extension or on your phone, and cards that keep failing are one click to find and fix (there's a filter for exactly the "cards I keep forgetting" pile).
And when you want them in Anki, they go to Anki: CSV export with the fields already mapped, or one-click AnkiConnect push straight into a deck you pick, de-duplicated so re-pushing doesn't create twins. We're not trying to replace Anki for people who love Anki — we're trying to fix the supply line that kills everyone else's deck.
Full disclosure: the extension isn't on the Chrome Web Store yet — we're in the last pre-launch stretch. If the supply-line problem is exactly your problem, join the waitlist and you'll get one email when it ships.
The honest summary
Anki, in 2026, is still the right foundation: free, private, cross-platform, with the best scheduler in the business built in. Its weakness was never the algorithm — it's that a spaced-repetition system is only as good as the cards you feed it, and it has never had an opinion about where cards come from.
So use it, keep the settings boring, prefer sentences with context and audio, kill your leeches, and above all solve capture — either with a pipeline that makes cards in one keystroke, or by accepting that you will only ever make cards for the words that truly matter. What doesn't work is planning to hand-type a thousand cards. That's the plan that has failed everyone.
FAQ
Q: Is Anki good for language learning? A: Yes — it's the standard for a reason: individually scheduled reviews mean you spend your time on the words you're about to forget rather than the ones you already know. Its weakness isn't the algorithm, it's that creating good cards by hand is slow, which is why most abandoned decks died of card starvation rather than bad scheduling.
Q: Is Anki free? A: On desktop (Windows, macOS, Linux), on Android via AnkiDroid, and for syncing through AnkiWeb — yes, completely free with no subscription. The one exception is iPhone/iPad, where AnkiMobile is a one-time $24.99 purchase that funds development of the rest.
Q: Should I use shared Anki decks or make my own cards? A: Use a shared frequency deck for your first few hundred words, where every list contains roughly the same items anyway. After that, your own cards win decisively, because a card's value comes largely from having met the word in content you chose — a word from a video you watched recurs in what you consume; word 3,412 of a stranger's deck does not.
Q: How many new Anki cards per day should a language learner add? A: Pick a number you can sustain on your worst day — 10 to 20 new cards daily is realistic for most people alongside real input, and it compounds to thousands per year. Every new card is a commitment to future reviews, so a 50-a-day sprint usually ends in a backlog you abandon.
Q: Should I turn on FSRS in Anki? A: Yes. It ships built into modern Anki, learns from your own review history, and Anki's documentation puts it at the same retention as the old SM-2 scheduler with roughly 20–30% fewer reviews. Leave desired retention at the default 90% and re-run the optimizer occasionally — raising retention toward 97% is the most common way people accidentally triple their workload.
Q: What is sentence mining? A: It's taking sentences out of real content that you almost fully understand — ideally with exactly one unknown word — and turning those into flashcards, so what you study is always one notch beyond your current level and always in context. It originated in the Japanese immersion community and it's the best answer to "which cards should I make," though it historically required a fair amount of tool setup to do smoothly.