Anki has a reputation problem. It’s simultaneously the most effective flashcard app ever built and the one most people abandon within two weeks. The reason isn’t the tool — it’s the setup. Out of the box, Anki’s defaults are mediocre, its interface is intimidating, and the settings that matter most are buried three menus deep.

This guide fixes that. Every recommendation here is grounded in spaced repetition research — particularly the FSRS algorithm developed by Ye (2024), Ebbinghaus’s foundational work on memory decay, Bjork’s desirable difficulties framework, and decades of empirical work on optimal review scheduling. By the end, you’ll have an Anki installation that is configured to maximize retention while minimizing the time you spend reviewing.

Let’s build this properly.

Why Anki and Not Other Apps

The short answer: the algorithm.

Most flashcard apps — Quizlet, Memrise, even Duolingo’s internal review system — use either fixed intervals or simplistic scheduling. They show you a card, you get it right, they show it again in a few days. The scheduling logic is basic and identical for every user.

Anki, since version 23.10, ships with FSRS (Free Spaced Repetition Scheduler) — a machine-learning-based algorithm developed by Jarrett Ye and the open-spaced-repetition team. FSRS models three components of memory for every card you study:

The practical result: FSRS achieves the same retention rate as Anki’s legacy SM-2 algorithm with 20–30% fewer reviews (Ye, 2024). That’s not a marginal improvement. If you’re doing 100 reviews per day, switching to FSRS saves you 20–30 reviews daily — for the same learning outcome. Over a year, that’s roughly 80–100 hours of your life.

No other consumer flashcard app offers anything close to this level of scheduling sophistication. That’s why Anki.

The other reasons matter too: Anki is free and open source on desktop and Android, has an enormous ecosystem of add-ons, supports audio, images, and cloze deletions natively, and syncs across devices. But the algorithm is the reason serious language learners converge on Anki. Everything else is convenience. The scheduling is the science.

Installation and Initial Setup

Desktop (Required)

Download Anki from apps.ankiweb.net. Install the latest stable release (24.x or newer). The desktop version is essential — it’s where you configure decks, create card templates, install add-ons, and enable FSRS. You can review on mobile, but configure on desktop.

Mobile

AnkiWeb Sync

Create a free account at ankiweb.net. In the desktop app, go to Sync (top-right button) and sign in. This syncs your cards, review history, and scheduling data across all devices. Sync before and after every session — make it a habit.

One critical rule: never review on two devices without syncing between sessions. Anki handles sync conflicts poorly. Review on your phone during the commute, sync, then review on desktop at home. Never the other way around.

Enabling FSRS: What It Does and How to Turn It On

FSRS replaces Anki’s legacy SM-2 scheduler with a modern, adaptive algorithm. Here’s exactly how to enable it:

  1. Open Anki desktop.
  2. Go to Tools → Preferences → Scheduling.
  3. Check “FSRS” to enable it.
  4. Click “Optimize” to let FSRS analyze your existing review history and compute personalized parameters. If you’re starting fresh with no review history, FSRS will use sensible defaults and optimize as you accumulate data.
  5. Set your Desired Retention to 0.90 (90%).

That last setting deserves explanation. The desired retention tells FSRS what probability of recall you want when a card comes up for review. At 0.90, FSRS schedules reviews so that you have a 90% chance of remembering each card when you see it.

Why 90% and not higher? Because the relationship between retention and workload is exponential, not linear. Going from 90% to 95% retention roughly doubles your review workload. Going to 97% roughly triples it. The research consensus — supported by Wozniak’s work on SM-2 and Ye’s FSRS analysis — is that 90% represents the optimal trade-off between retention and time investment for most learners. You forget some cards, but the cost of preventing that forgetting is not worth the extra hours.

If you’re studying for a high-stakes exam and need near-perfect retention, you might push to 92–93%. For casual maintenance of a language you already speak, 85% is fine. For most active learners: leave it at 90%.

What FSRS Does Differently

Under the old SM-2 algorithm, every card followed the same scheduling formula regardless of your personal forgetting patterns. SM-2 used a single “Ease Factor” per card that could spiral downward (the notorious “ease hell”), creating cards that appeared with punishingly short intervals no matter how many times you got them right.

FSRS eliminates ease hell entirely. It models your memory mathematically, tracks how your actual recall probability changes over time, and schedules each review at the precise moment when your retrievability drops to your target threshold. It learns from your data: the more you review, the better it predicts your forgetting curves.

The practical difference: reviews feel better-timed. Cards you know well disappear for months. Cards you struggle with appear more frequently — but not so frequently that you’re grinding through the same card every day forever.

Deck Configuration: Optimal Settings Based on Research

Go to your deck, click the gear icon, and select Options. Here are the settings that matter:

New Cards

Reviews

Lapses

FSRS-Specific

Leave everything else at defaults.

Card Types: Which Ones to Create for Language Vocabulary

Not all flashcards are equal. The type of card you create determines what cognitive process you’re practicing. For language learning, three card types cover virtually everything you need.

1. Sentence Cards (The Default Choice)

A sentence card presents a full sentence in the target language, with one unknown word highlighted or underlined:

Front: 彼は毎朝コーヒーを飲む。 Back: He drinks coffee every morning. (飲む / nomu = to drink)

Why sentences and not isolated words? Because isolated word pairs (飲む → to drink) strip away context, collocations, and grammar. You learn that 飲む means “to drink” but not that it takes the particle を, not how it sounds in a natural sentence, not what words typically surround it. Sentence cards encode all of this simultaneously. As we discussed in the sentence mining article, this approach aligns with what research tells us about contextual encoding and elaborative processing.

The rule: one unknown element per card. If a sentence has three words you don’t know, it’s not a good sentence card — it’s three cards pretending to be one. You should understand 90–95% of the sentence effortlessly. The target word is the one thing you’re retrieving.

2. Cloze Deletions

A cloze card hides part of a sentence and asks you to fill it in:

Front: Elle est {{c1::partie}} sans dire au revoir. Displayed as: Elle est […] sans dire au revoir. Back: Elle est partie sans dire au revoir. (She left without saying goodbye.)

Cloze deletions are excellent for grammar patterns, verb conjugations, and collocations. They force productive recall — you must generate the missing word, not just recognize it. Research on the generation effect (Slamecka & Graf, 1978) shows that producing an answer, even when wrong, leads to stronger encoding than passive recognition.

To create a cloze card in Anki: select the “Cloze” note type, then highlight the word you want to hide and press Ctrl+Shift+C (or Cmd+Shift+C on Mac).

3. Audio Cards (For Listening Comprehension)

Front: [Audio clip of a native speaker saying the sentence] Back: Written sentence + translation

Audio cards train your ear. They’re particularly valuable for tonal languages (Mandarin, Vietnamese, Thai) and for languages whose phonology differs sharply from your L1. You hear the sentence, try to understand it, then check.

For creating audio cards: use Forvo (forvo.com) for individual word pronunciation, or extract audio clips from podcasts and shows using tools like Audacity. The Anki add-on “AwesomeTTS” can generate text-to-speech audio directly on your cards — useful for bulk creation, though native speaker recordings are always preferable.

What About Image Cards?

Adding an image to your sentence cards is a bonus, not a separate card type. Paivio’s Dual Coding Theory (1971) shows that encoding information through both verbal and visual channels creates stronger, more retrievable memory traces. If you can add a relevant image to a sentence card without spending more than 10 seconds, do it. Don’t let image sourcing become a bottleneck that slows down card creation.

Best Pre-Made Decks for Common Languages

Creating your own cards is always superior — the act of creating a card is itself an encoding event. But pre-made decks serve as a launchpad, especially for the first 1,000–2,000 most frequent words. Here are the best options:

Japanese: - Core 2K/6K Optimized — Sentence-based, with audio and example sentences. The gold standard for Japanese beginners. - Tango N5/N4 — Based on the Tango book series. High-quality sentence cards organized by JLPT level.

Spanish: - 5000 Most Common Spanish Words — Frequency-ordered with example sentences. - Spanish Sentences (Tatoeba) — Community-sourced sentences with translations.

French: - French 5000 — Frequency list with audio. - Assimil French — Companion deck to the Assimil course. Excellent sentence-level content.

Mandarin Chinese: - HSK Vocabulary (by levels) — Organized by proficiency exam levels, with pinyin and audio. - Spoonfed Chinese — Graded sentence cards from simple to complex. One of the best-designed language decks available.

German: - German 5000 Frequency List — With example sentences and audio. - Nicos Weg (Deutsche Welle companion) — Pairs with the free DW online course.

Korean: - TOPIK Vocabulary by Level — Organized by proficiency test levels. - Korean Sentences — Sentence-based with audio.

To find these: go to ankiweb.net/shared/decks and search by language. Sort by rating and download count. Always preview a deck before committing — check that it has audio, that the sentences are natural, and that the formatting is clean.

Important: treat pre-made decks as a starting point, not an endpoint. After the first few weeks, you should be creating your own cards from content you’re actually consuming — books, podcasts, TV shows, conversations. Cards mined from your own immersion are more memorable because they carry personal context and emotional salience.

How to Create Sentence Cards Efficiently

The fastest workflow for creating sentence cards:

The Mining Loop

  1. Consume content in your target language (read a book, watch a show, listen to a podcast).
  2. Notice an unknown word in a sentence you otherwise understand.
  3. Copy the sentence to Anki (or a staging area like a text file or the Anki “Add” dialog).
  4. Add the translation of the target word and, optionally, the full sentence translation.
  5. Add audio if available (from Forvo, AwesomeTTS, or the source material).
  6. Move on. Don’t spend more than 30–60 seconds per card.

Tools That Speed This Up

The cardinal rule of card creation: don’t let perfection slow you down. A sentence card with just the sentence and a translation is already a good card. Audio and images are bonuses. If adding audio takes you from 30 seconds per card to 5 minutes per card, skip the audio and make more cards.

New Cards Per Day: What Number to Choose and Why

This is the single most consequential setting in Anki, and the one most people get wrong. The math is unforgiving.

The Review Accumulation Problem

Every new card you learn today becomes a review card tomorrow. And the next day. And the next week. And the next month. New cards are not a one-time cost — they’re a recurring subscription to future reviews.

Here’s the math. Assume FSRS with 90% retention and typical intervals:

These are steady-state numbers. The review load builds gradually over weeks, which is why people set 30 new cards/day in the first week (when reviews are light), feel fine for a month, and then face an avalanche of 250+ daily reviews by month two. They burn out and quit.

The Recommendation

The principle comes from Pimsleur’s graduated-interval recall research (1967) and from what Bjork calls desirable difficulty: the optimal challenge level is the one you can sustain indefinitely. A habit of 10 cards/day for a year (3,650 cards) beats 30 cards/day for six weeks followed by quitting (1,260 cards). Consistency dominates intensity.

If you’re ever unsure, err on the side of fewer new cards. You can always increase later. You can never un-burn-out.

Common Configuration Mistakes and How to Avoid Them

Mistake 1: Capping Maximum Reviews

Setting maximum reviews to 100 or 200 “to save time” is the most destructive mistake you can make. When you cap reviews, due cards pile up behind the cap. Tomorrow you have even more due cards. The backlog grows exponentially. Within weeks, you’re seeing cards so late that you’ve forgotten them, which triggers more failures, which generates more reviews, which hit the cap again.

Fix: Set maximum reviews to 9999. Control your workload by adjusting new cards per day instead.

Mistake 2: Not Enabling FSRS

If you’re still running the default SM-2 scheduler, you’re doing 20–30% more reviews than necessary for the same retention. There’s no reason not to switch. FSRS is strictly superior.

Fix: Enable FSRS in Preferences → Scheduling. Click Optimize. Done.

Mistake 3: Too Many Learning Steps

Some guides recommend learning steps like 1m 10m 1h 1d 3d. This micromanages the short-term learning process and interferes with FSRS’s long-term scheduling. FSRS is designed to handle the transition from short-term to long-term memory. Let it.

Fix: Use 1m 10m for learning steps and 10m for relearning steps. Keep it simple.

Mistake 4: Isolated Word Cards

A card that says “laufen” on the front and “to run” on the back teaches you a dictionary entry, not a word. You won’t know how to use it, what prepositions it takes, or what it sounds like in a sentence.

Fix: Always use sentence cards. One target word per sentence, 90–95% comprehension of the surrounding context.

Mistake 5: Ignoring Leeches

A leech is a card you’ve failed 8+ times. It’s telling you something: the card is badly designed, the information isn’t sticking in this format, or you’re not ready for this word yet.

Fix: When Anki tags a card as a leech, don’t just unsuspend it and keep grinding. Redesign the card: add context, change the sentence, add an image, add a mnemonic, or split it into simpler cards. If the word is obscure and you’ve never encountered it in real content, consider deleting it entirely.

Mistake 6: Studying Anki Instead of the Language

Anki is a retention tool, not a learning tool. It helps you remember what you’ve already encountered. If you’re spending 60 minutes on Anki and 0 minutes reading, listening, or speaking, your priorities are inverted.

Fix: Anki should take 15–20 minutes of your daily study time. The rest — the majority — should be input (reading, listening) and output (speaking, writing). Anki serves the immersion, not the other way around.

Mistake 7: Rating Cards Too Generously

When you press “Good” on a card you actually struggled with, you’re feeding bad data to FSRS. The algorithm schedules the next review based on your rating. A dishonest “Good” pushes the interval too far out, and you’ll forget the card.

Fix: Be brutally honest. If you hesitated for more than 5–6 seconds, press “Hard.” If you couldn’t recall the answer, press “Again.” FSRS works better when it has accurate data about your actual memory performance.

Putting It All Together

Here’s your complete setup checklist:

  1. Install Anki desktop + mobile app
  2. Create AnkiWeb account and sync
  3. Enable FSRS (Preferences → Scheduling → FSRS → Optimize)
  4. Set desired retention to 0.90
  5. Set new cards/day to 10
  6. Set learning steps to 1m 10m
  7. Set relearning steps to 10m
  8. Set max reviews to 9999
  9. Set leech threshold to 8
  10. Download a pre-made frequency deck for your target language
  11. Start creating your own sentence cards from day one alongside the pre-made deck
  12. Review every day. No exceptions. No “I’ll catch up tomorrow.”

That last point isn’t a configuration setting — it’s the setting that matters most. Ebbinghaus showed us in the 1880s that memory decays exponentially without review. Pimsleur formalized graduated-interval recall in 1967. Wozniak operationalized it with SM-2 in 1987. Ye perfected it with FSRS in 2024. But none of it works if you don’t show up.

Anki is a daily practice. Fifteen minutes, every day, for as long as you’re learning. The algorithm handles the complexity. You just have to press the button and answer honestly.

This article is part of the series “The Science of Language Learning” — where we break down what research actually says about how adults acquire languages, and how to use that science to learn faster.

Previous in the series: Sentence Mining: The Most Underrated Vocabulary Method

Next in the series: FSRS vs. SM-2: Why You Should Switch Your Anki Algorithm Today

References: