Flashcards for language learning

Migaku sells a flashcard tool. Its own guide to making Anki cards budgets 30 seconds a card, then does the multiplication out loud: 1,500 cards is "750 minutes or about 12.5 hours."

That figure comes from a company with every reason to make card-building sound quick.

Gabriel Wyner, who wrote the book on building your own cards, is blunter about the steady state. On his own blog he reports making 110 cards in 1.5 hours, covering about 40 words, and says a 25-new-cards-a-day habit costs "2–3 hours a week for card creation." He also notes the first thousand cards go roughly twice as fast as the ones after them, so the rate gets worse, not better.

Nobody quits flashcards because the algorithm was wrong. They quit because Tuesday night arrived and the deck needed feeding before it would give anything back.

What flashcards cost you in time

Run the arithmetic on a normal week and the shape of the problem shows up fast.

Anki's FAQ says: "By default, Anki will show you a maximum of 20 new cards each day." Take Migaku's 30 seconds a card, which is the optimistic end, and 20 new cards is 10 minutes of typing before you have reviewed anything at all. At Wyner's realistic rate it is closer to 25 minutes.

Then the reviews arrive on top. The Anki manual puts the ratio in writing: "If you are consistently learning 20 new cards a day, you can expect your daily reviews to be roughly about 200 cards/day." So a 20-a-day habit is 10 to 25 minutes of construction plus a 200-card review pile, every day, forever.

The build cost, from the people selling the method

0

hours to hand-build 1,500 cards at 30 seconds each, per Migaku's own Anki guide (12.5 hours exactly)

0

hours a week Wyner budgets for card creation at 25 new cards a day, and he says the rate slows after the first thousand

This is the number that decides whether you are still doing this in March, and it is missing from every page ranking for this query. Busuu spends 2,100 words teaching a 7-step card-building process and never mentions what those steps cost you.

So that becomes the question to take shopping. Every flashcard app schedules reviews competently now. They differ on who does the typing.

What they cost in money

I checked all of these on 21 September 2026. Where a price is missing from the table it is because the company does not publish one, and I would rather leave a hole than guess at it.

Three of the six will not tell you what they charge on the page asking for your attention. Noji goes furthest: it lists exactly which features sit behind Pro, raises pricing in its own FAQ, and still shows no number, with the pricing URL dead.

On Quizlet I want to be precise about what I do not know. Those three figures are real App Store in-app purchase prices, and Apple's listing attaches no billing period to any of them, so I cannot tell you whether $44.99 is a year or something else. Quizlet's own pages return a 403 to me. What is documented, from its help centre, is the gating: Learn and Test modes are paid for individuals, with one free practice test per set.

Babbel, Memrise and Clozemaster are absent for the same reason. I could not read a first-party price for them today, and a wrong price would discredit the five I did verify.

Why they work anyway

Given that cost, the method has to earn its place. It does, and the evidence is old enough to be boring.

Karpicke and Roediger taught students 40 Swahili-English word pairs and tested them a week later. Pairs that kept being tested after the first correct answer came back at about 80%. Pairs studied again but dropped from testing came back at 36%. Same students, same words, same time in the lab. Students predicted about 50% for themselves in every condition, so they could not feel the difference while it happened.

That is the retrieval half. The spacing half comes from Cepeda and colleagues, who pooled 317 experiments and found the same hour of study buys more memory chopped across days than spent in a block.

A flashcard app is those two results wired to a calendar. The longer version of both is in the testing effect, and the scheduler machinery in spaced repetition without the jargon.

One correction worth carrying with you, since the apps rarely make it. Pictures on vocabulary cards are not the free win they are sold as. Carpenter and Olson found that pictures did not initially beat native-language translations for foreign vocabulary, and that learners were more overconfident recalling from a picture than from a word. The image only helped once that overconfidence was corrected. Add the picture, then still grade yourself on whether the word came out of your mouth.

Where the cards come from

Here is the question the method-teaching pages walk past. You have three sources, and they trade off against each other honestly.

Hand-built. Best retention, highest quit rate. Slamecka and Graf named the generation effect in 1978: material you generate is remembered better than the same material read, so writing the card is already a rehearsal. Anki's manual agrees, calling your own deck "the most effective way to learn a complex subject" and shared decks "a supplement to external material, not as a replacement for it." The cost is the 12.5 hours per 1,500 cards from the top of this page.

Premade decks. Instant, and they teach somebody else's words. A 5,000-card Spanish deck downloads in seconds carrying a stranger's example sentences, translations and idea of which words matter. For arbitrary material that is fine and I recommend it: numbers, days of the week, colors and the first few hundred concrete nouns all have one right answer that does not shift with context. For anything with a boundary against a neighbour word, it is thin. A downloaded card reading saber / "to know" can be passed for a year without ever telling you which of Spanish's two verbs for knowing to say at dinner.

Generated from your own material. The words you already met, turned into cards without you typing them. This is where the time cost goes away, and it is what I built, so treat the next section with the suspicion it deserves.

The split I would actually give a friend: download one frequency-ordered core deck, cap it at the first 1,000 words, and write or capture everything else from things you met.

What a good card has on it

Regardless of where it came from, these four things travel with you to any app, including a paper box.

Make it produce, not recognize. A blank in a sentence beats a multiple-choice lineup, because generating the word is the half that shows up when you are speaking. Flipping a card over to see whether it looks familiar is not a retrieval.

Keep the sentence you met it in. Register, collocation and the boundary against a neighbour word all live in context, and a bare word pair throws all of it away.

Put sound on it. You are training for speech, and a word you have only ever read is a word you will not catch when someone says it at speed.

Hold the deck size down. The Anki manual warns that "More than one Anki user has excitedly studied hundreds of new cards over their first few days of using the program, and then has become overwhelmed by the reviews required." Five to ten new cards a day is a rate a bad week survives, and I ran the full arithmetic in how many flashcards a day.

If you want a cadence to follow by hand, the Leitner box is the honest paper version: five stacks, reviewed daily, every other day, every 3 to 4 days, weekly, then fortnightly, moving a card up when you get it right and back to stack one when you miss. Rosetta Stone publishes exactly that schedule, alongside a sensible recommendation to work with 30 to 50 cards at a time. Software's only real advantage is knowing which 12 of your 600 cards are due today.

Who should skip this entirely

Three kinds of people are better served elsewhere, and I would rather say so than sell past them.

If you are in week one, you have nothing to put on a card. A learner who has met eleven words cannot build a deck, and should not try. Take lessons, or open a course, and come back when you have material.

If you enjoy making cards, install Anki and stop reading. It is free on desktop and Android, US$24.99 once on iPhone, and your collection lives in a file you can export. Turn FSRS on at the bottom of the deck options page, since it is opt-in and Anki still ships the 1987 SM-2 algorithm by default. Nothing I sell competes with that on cost or control. For a certain kind of person Anki is the correct answer to this entire page, and I have written the honest comparison rather than pretending otherwise.

If your problem is freezing mid-sentence, a deck will not fix it. Flashcards build recall of words you have already met. They do not teach you to parse fast speech with no replay button, and a learner with 3,000 well-drilled words and no listening hours behind them will still stall in a real conversation. Run reading and listening alongside the deck from day one. That argument is in input vs flashcards.

Your first week

Day 1. Pick the source of your cards before you pick the app. If you take lessons, your cards come from lessons. If you do not, download one frequency deck and cap it at 1,000 words.

Day 2. Set new cards to 5 a day. You will want 20. The 200-review pile is six weeks away and it is real.

Day 3. Grade honestly on the first card that tempts you. If the word arrived after the flip, that is a miss, no matter how much you "basically had it." The scheduler sees your memory through one channel, and the grading post has the rubric with the edge cases.

Day 7. Time yourself for a week, building and reviewing separately. If building is eating more than a third of your minutes, change something: fewer new cards, a premade deck for the arbitrary words, or a tool that does the typing.

Where Verbamor fits

I built Verbamor because the card-building hour was the thing that kept killing my own deck. You record or import a lesson, or paste a transcript, and it mines the vocabulary you actually used, capped at 60 words and 30 phrases per recording, then builds fill-in-the-blank cards with generated audio and a picture and schedules them with FSRS. Spanish, French, Italian, Portuguese and Russian. It is $19.99 a month or $129.99 a year, with the 7-day trial on the annual plan, iPhone and iPad only, and there is no free tier.

The honest limits: the card audio is synthesised text to speech rather than a native speaker in a booth, and Verbamor only holds words you have already met, so it will not teach a beginner their first thousand. If you want those first thousand, that is a premade deck or a course, and I said so two sections ago.

Questions people actually ask

---

Sources

Every load-bearing claim Verbamor makes is traced to its paper on the research page.