Sign in with Google

How to Use Spaced Repetition Apps (Anki and Others)

What flashcards are genuinely for, how to write cards that work, and how to avoid the review avalanche.

beginner10 min read
learning-metaspaced-repetitionankiflashcardsstudy-skills
The usual story goes like this. You install Anki, download a deck with four thousand cards in it, and do two hundred reviews on day one feeling extremely capable. On day nine the app says 640 cards are due, most of them phrased in a stranger's words about things you do not care about, and you never open it again. Nothing was wrong with the method. The material was wrong, the cards were wrong, and the daily limits were left at values that quietly built a debt.
iWhat you need
A computer for authoring cards (the desktop app is free on Windows, macOS, and Linux) and about forty minutes. A phone is optional but useful for reviews — AnkiDroid on Android is free, while AnkiMobile for iPhone and iPad is a paid one-time purchase, which funds Anki's development. Syncing between them through a free AnkiWeb account is included.
Every time you successfully pull a fact out of memory, that memory gets more durable — retrieval practice, the effect Henry Roediger and Jeffrey Karpicke demonstrated when learners tested on material remembered far more of it a week later than learners who reread it for the same time. Separately, Hermann Ebbinghaus showed in the 1880s that spacing repetitions out beats packing them together. An app combines the two: it shows you a card just before you would have forgotten it, and each success pushes the next showing further out — a day, then three, then a week, then a month. Total time collapses compared with rereading, because the app stops spending your attention on what you already know.
This distinction is the whole guide, and getting it wrong is why people conclude flashcards do not work for them. Spaced repetition is a recall tool: it moves discrete, checkable facts into memory and keeps them there. It does not build understanding or skill.

Right material versus wrong material

  1. 1Good fit: vocabulary, anatomy, chemical properties, formulas, keyboard shortcuts, command flags, function signatures, dates, definitions, syntax you keep looking up.
  2. 2Bad fit: why recursion works, how to debug, how to design a schema, how to write an essay. These are models and skills — you need to do them, not recall them.
  3. 3The test: if the goal is to answer correctly, make a card. If the goal is to do something in a new situation, that is practice, and a card will not get you there.
  4. 4Use the grey zone well: facts a skill depends on go in cards, the skill goes in practice. Memorize what a Git command does; learn merging by merging.
Download the desktop app from apps.ankiweb.net. The interface is deliberately plain: a deck list, an Add button for making cards, and a Browse window for editing them. Everything else sits behind the gear icon beside each deck, which opens deck options.

First twenty minutes

  1. 1Create one deck. One. Resist subdecks until you have a real reason — a tidy tree of empty decks is procrastination wearing a hat.
  2. 2Click Add. Leave the note type as Basic. Front is the question, Back is the answer.
  3. 3Write ten cards from material you learned this week and actually want.
  4. 4Open deck options via the gear icon and set New cards per day between five and fifteen. This one number decides whether you are still here in a month.
  5. 5Study the deck: Anki shows a card, you answer before revealing, then grade yourself. Grade honestly — a generous grade only postpones the reckoning.
  6. 6Create a free AnkiWeb account and sync. Images and audio in your notes sync too.
✓Which scheduler you are on
Anki's legacy algorithm descends from SuperMemo 2, written SM-2. Newer versions also ship FSRS, which fits intervals to your own review history and generally spaces cards more accurately; it lives in the FSRS section at the bottom of deck options. Turning it on is a reasonable default — but do not spend week one tuning it, because card quality matters far more than scheduler choice.
Most failed decks are a card-writing problem, not a discipline problem. Piotr Woźniak, who built SuperMemo, wrote a set of twenty rules for formulating knowledge, and a handful of them carry almost all the weight. The central one is the minimum information principle: a card should test one thing, because a card testing four things fails whenever any one of them slips, and then the three you knew get punished alongside it.

The rules that matter most

  1. 1One fact per card. If the answer has an "and" in it, split the card.
  2. 2Ask an actual question. A front reading "Photosynthesis" is a topic, not a prompt — there is nothing specific to retrieve.
  3. 3Never put a list on a card. Lists recall in order, badly, and one gap fails the whole thing. Split them or use a cloze.
  4. 4Write it in your own words. Rephrasing forces you to process the idea — the generation effect — and someone else's phrasing is something you will misread for months.
  5. 5Add enough context to be unambiguous but not guessable. "What is the limit?" is unanswerable; "Anki's default leech threshold, in lapses?" is answerable.
markdown
BAD  (four facts, a list, no real question)
  Front: Python list copying
  Back:  a = b is a reference, a[:] and list(a) are shallow copies,
         copy.deepcopy handles nesting, [[0]*3]*3 shares rows

GOOD (one fact each, phrased as questions, in my own words)
  Front: After  b = a  on a Python list, how many list objects exist?
  Back:  One. Two names, one object -- mutating either changes both.

  Front: What does  a[:]  give you, and how deep does it copy?
  Back:  A new list, one level deep. Nested objects stay shared;
         use copy.deepcopy for those.

  Front: Why does  [[0]*3]*3  break on  grid[0][0] = 1 ?
  Back:  The outer *3 repeats the SAME inner list, so all three rows
         are one object. Build rows in a comprehension instead.

One bad card, rewritten as three that work

Shared decks are the most common way this fails. It looks like a shortcut, but the value of a flashcard is mostly created at the moment you write it. Writing a card is when you decide what the question really is, notice you cannot state the answer cleanly, and go reread the part you skimmed. A stranger's deck skips exactly the part that was doing the learning, and leaves you grading yourself on phrasings you have to decode first.
!The one defensible exception
Large shared decks are reasonable for genuinely standardized material where the facts have no local flavour — a language's core vocabulary, or a well-known professional exam list. Even then, delete aggressively on first pass and rewrite anything you do not immediately understand. A card you do not understand is not a card; it is a future leech.
Reviews are meant to be small, boring, and daily — ten to twenty minutes for a mature personal deck, attached to a routine you already have. The trap is that today's new cards become tomorrow's reviews, and next week's, and next month's. Thirty new cards a day feels fine for about eleven days and then produces a load you will not sustain.

Keeping the load flat

  1. 1Set new cards per day low and leave it. Ten suits a deck you intend to keep for years.
  2. 2Judge the load by your steady-state review count, not today's. New cards compound into reviews for months.
  3. 3Never skip days and then binge. Review daily, or set new cards to zero and let the backlog drain.
  4. 4Sitting on a backlog of thousands? Set new cards to zero, clear the old reviews over a week or two, then resume. Do not delete the deck.
  5. 5Use cloze deletions on prose: switch to the Cloze note type, select the words to hide, click the bracketed button. Anki makes one card per deletion, keeping each single-fact.
  6. 6Watch for leeches. By default, once a card lapses eight times Anki tags the note as a leech and suspends it. That is the app telling you the card is broken.
Take the leech tag seriously rather than fighting it. A card failed eight times is almost never a memory problem — it is ambiguous, it tests two things, or you do not actually care about it. Rewrite it as two clearer cards, or delete it. Deleting cards is normal and healthy; a deck that only ever grows is a deck heading for abandonment.

Honest comparison

  1. 1Anki: free on desktop and Android, paid on iOS, plain-looking, endlessly configurable, deepest scheduler. The right default if you expect to do this for years.
  2. 2RemNote: notes and flashcards in one app, so cards get made inline as you write. Its free tier includes unlimited notes and flashcards, with caps on PDF annotation and image occlusion; paid plans run about eight dollars a month.
  3. 3Quizlet: easiest onboarding, strong for language vocabulary and school material, but less configurable scheduling and several study modes behind a subscription. Fine for a semester, weaker as a decade-long system.
  4. 4A paper Leitner box: dated dividers, index cards, no notifications. Genuinely works, and worth knowing about if software is the thing that keeps derailing you.

Symptom, cause, fix

  1. 1Reviews ballooned to hundreds a day in week three. New cards per day was too high. Set it to zero, drain the backlog, resume at ten.
  2. 2You forget cards despite reviewing them. They test multiple facts, or use phrasings you never internalized. Split them and rewrite in your own words.
  3. 3You know it in the app but cannot use it in real work. That material was a skill, not a fact. Move it into practice.
  4. 4One card keeps failing forever. It is a leech and it is broken. Rewrite it as two cards or delete it — do not grind it.
  5. 5The deck you downloaded feels meaningless. It is someone else's phrasing of someone else's priorities. Keep the twenty cards you care about, write your own.
  6. 6You skip days and dread opening the app. The habit has no anchor, and the load is too big. Attach reviews to a routine event and cut new cards.
  7. 7You grade generously to finish faster. That inflates intervals and guarantees failures later. More than a few seconds of thinking was not a pass.
Cards handle recall, not understanding, and the two need different homes. How to Take Notes While Learning to Code covers where explanations and worked examples live — and pairs directly with this, since a good concept note is usually where your best cards come from. To fit the daily review into a week with other demands, How to Make a Study Schedule covers anchoring it to a routine, and How to Use Obsidian for Note-Taking keeps the notes side navigable as it grows. The Learning Mastery roadmap sequences the whole thing: what to memorize, what to practice, when to review.