Memory

How many new words should you learn per day?

About 10 new words a day is a sensible start for most learners; after that, let your review time set the number. Every new word brings reviews for months. In our model, where you recall 90% of reviews, each new word a day adds about 3.7 reviews a day after one month, 6.2 after three months and 8.6 after a year. So 10 new words a day means about 86 reviews a day once you have kept it up for a year.

By the Readlang Mobile team · Updated 31 October 2026

How many new words a day should you learn?

Start with about 10 new words a day and change the number only after a few weeks, when you can see what your reviews cost. The limit on new words is not how many you can take in at once; it is the reviews they bring with them. A word you add today comes back the next day, a few days later, a week or two later, then at longer and longer gaps for months. Adding new words is quick. The reviews that follow keep growing for a long time after you start.

Three questions decide whether 10 is right for you:

• How much time do you have each day? In our model, 10 new words a day settle at about 86 reviews a day after a year (table below). If that does not fit, use 5; if it takes you a few minutes, try 15 or 20.
• Is a word one card or two? A word studied in both directions, foreign-to-native and native-to-foreign, is two cards. Ten words then cost what twenty single cards cost.
• Can you review every day? Reviews you skip do not go away. Miss a day and the next day brings two days' worth.

This page works out the numbers behind that advice. Our hub on spaced repetition covers why spaced reviews work and how far apart they should be.

How many reviews will 5, 10, 20 or 30 new words a day create?

Daily reviews grow quickly in the first weeks, then more slowly, and are still creeping up after a year. The table gives our estimate for someone who adds the same number of new words every day, one card per word, and recalls 90% of the cards they review. The last column is the rule of thumb from Anki's manual, 20 new cards for about 200 reviews, scaled to each pace.

Every figure in our columns is the number of new words a day multiplied by one factor for each day: about 3.7 on day 30, 6.2 on day 90 and 8.6 on day 365. If you recalled every card every time, the factors would be 3, 5 and 6, so forgetting adds about a quarter to the load after a month and about 40% after a year.

How we calculated the review load

The estimate is our own. It uses a fixed review schedule and a fixed recall rate, so you can repeat it in a spreadsheet and change the assumptions.

1. New words. You add N new words every day, one card per word, starting on day 1. The first study of a card does not count as a review.
2. Schedule. A card is first reviewed one day after you learn it. Each later gap is 2.5 times the one before, rounded to whole days. 2.5 is the starting ease in Anki's original scheduler (Anki manual), and we assume you press Good every time you recall a card.
3. Recall. You recall 90% of the cards you review, the default target of Anki's FSRS scheduler (Anki manual).
4. Forgetting. A card you fail comes back the next day and climbs the schedule again from the first gap, as Anki does by default when it resets a forgotten card's interval and applies a one-day minimum (Anki manual). The extra repetition a few minutes after a failure, while you relearn the card, is not counted.
5. On time. You do every review on the day it is due.

If you never forget, the count is simple. On day T, the words you added on earlier days are between 1 and T − 1 days old, and a batch of N words is due whenever its age matches a day in the right-hand column. On day 30 the batches aged 1, 4 and 12 days are due: 3 × N reviews. On day 90 the batches aged 32 and 82 days join them: 5 × N. On day 365 the batch aged 207 days joins too: 6 × N. The next step comes on day 521, when the first batch is 520 days old.

With forgetting, each review is recalled with probability 0.9 and failed with probability 0.1, and a failure starts a new schedule from that day. Let F(x) be the expected number of reviews a card receives in its first x days, with d₁ … d₇ = 1, 4, 12, 32, 82, 207, 520 and F(x) = 0 for x below 1:

F(x) = sum over every d_k up to x of 0.9^(k-1) × [1 + 0.1 × F(x - d_k)]

reviews on day T = N × F(T - 1)

The first term counts the review itself, which happens only if the card has survived the earlier reviews (probability 0.9 for each one). The second adds the reviews of the restarted schedule when this review fails. We worked the recursion through day by day up to day 89 and with a coarser step beyond: F(29) ≈ 3.69, F(89) ≈ 6.19, F(364) ≈ 8.57. Treat the one-year figure as accurate to about ±0.1.

A way to see where the load comes from: SuperMemo's documentation defines a collection's burden, the average number of reviews needed per day, as the sum of 1/I over all items, where I is each item's current interval (SuperMemo). A card on a 1-day interval costs one review a day; a card on a 50-day interval costs 0.02. New and recently failed cards sit on short intervals, which is why they make up most of the load.

Why does Anki say 20 new cards make about 200 reviews?

Anki's manual says that if you consistently learn 20 new cards a day, you can expect roughly 200 reviews a day, and it warns that users who study hundreds of new cards in their first few days have been overwhelmed by the reviews that follow (Anki manual). That is about ten reviews a day for each daily new card. Our model gives 8.6 after a year, so the two agree on the size of the load. Ours comes out lower for four reasons, all of which push real review counts up:

• Anki's original scheduler lowers a card's ease by 20 percentage points each time you forget it, so its later gaps grow more slowly than in our model (Anki FAQ).
• Hard multiplies the current gap by 1.2 by default instead of the 2.5 of Good, and lowers the ease by 15 points (Anki FAQ); our model assumes Good every time.
• We do not count the extra repetition while you relearn a failed card.
• The load keeps creeping up after a year: in our schedule the seventh review step arrives at day 520 and adds another N reviews a day.

Use the factor of ten from Anki's manual as a long-run budget and our factors as the path towards it: about four reviews per daily new word after the first month, about six after three months, eight to ten after a year.

How long will the reviews take?

Multiply the number of reviews by the seconds a review takes you and divide by 60. Time yourself over a day's session, because review speed varies a lot with the kind of card and with how well you know the words. As an example, 86 reviews at 10 seconds each, the one-year load for 10 new words a day, take about 14 minutes.

The same arithmetic run backwards turns a daily time budget into a number of new words. Using the one-year factor of 8.6:

new words a day = (minutes × 60 ÷ seconds per review) ÷ 8.6

If you are slower, say 15 seconds a review, each row supports two thirds as many new words.

Is there research on the right number of new words a day?

Not directly. Studies of vocabulary learning compare ways of spacing and testing a fixed set of words. They do not compare daily quotas kept up over months, and none recommends a number. What they show is that the size of a study set matters less than how far apart its repetitions are:

• Set size. Nakata and Webb (2016) compared learning 20 English words as one block with learning them in sets of 4 or 10. With the spacing between repetitions held equal, the split made little difference; changing the spacing had a larger effect.
• Bigger stacks feel worse but work better. In Kornell's (2009) flashcard experiments, one large stack, in which each card came back only after many others, beat several small stacks for 90% of participants, yet 72% believed the small stacks had worked better.
• Native speakers. Goulden, Nation and Read (1990) estimated that well-educated adult native speakers of English know around 17,000 base words, which works out at two to three new words a day. Nation and Waring (1997) put native speakers' growth at roughly 1,000 word families a year. Those words come mostly from reading and listening, not flashcards, so the figure shows that a modest daily rate adds up rather than setting a target.

How many words can you learn in a year at each pace?

New words a day times 365 gives the number of cards you add in a year. With a 90% recall target, you should still recall about 9 in 10 of them whenever they come up for review.

The 2,000 and 3,000 marks come from the research on vocabulary size: the most frequent 2,000–3,000 word families make up most of any text or conversation, and they are the ones worth learning deliberately (how many words you need). At 10 new words a day they take seven to ten months. Two cautions: a word family such as watch, watched, watching is one card in this count, and a word studied in both directions needs two cards, which halves the words per year.

When should you lower or raise the number?

• Lower it when reviews pile up. Anki's manual recommends that if you have a backlog of overdue reviews, you stop introducing new cards until you have caught up (Anki manual). Adding new words on top of a backlog makes it grow.
• Raise it after a calm month. The load in the first weeks understates what is coming: in our model it rises from about 3.7 reviews per daily new word on day 30 to 6.2 on day 90. Judge a pace only after it has run for a month or two.
• Keep the recall target near 90%. Anki's manual says higher retention means shorter intervals and more reviews per day, that the workload rises very quickly above 90%, and advises keeping desired retention below 97% (Anki manual). Before you cut new words, check you have not raised this setting.
• Choose words that will come back. In our counts of 117 books, 46% of the different word forms in a typical English book appear only once. A card for a word you will rarely meet again costs the same reviews as a card for a common one. Take cards from the words you keep meeting while reading.
• Count cards, not words. If you add a sentence card and a listening card for each word, a 10-word pace is a 20- or 30-card pace, and the tables above scale accordingly.

Our first-week plan sets up a daily new-card limit together with a routine of reading and review.

References

1. Anki Manual. Deck Options: Daily Limits (New cards/day), Lapses and FSRS (Desired retention); starting ease of 2.50 in the original scheduler. docs.ankiweb.net/deck-options.html, checked 31 October 2026.
2. Anki Manual. What spaced repetition algorithm does Anki use? (SM-2 ease changes for Again and Hard, hard interval 1.2). docs.ankiweb.net/faqs/what-spaced-repetition-algorithm, checked 31 October 2026.
3. SuperMemo. Burden. supermemo.guru/wiki/Burden, page last edited 5 May 2018, checked 31 October 2026.
4. Goulden, R., Nation, P., & Read, J. (1990). How large can a receptive vocabulary be? Applied Linguistics, 11(4), 341–363. doi:10.1093/applin/11.4.341
5. Kornell, N. (2009). Optimising learning using flashcards: Spacing is more effective than cramming. Applied Cognitive Psychology, 23(9), 1297–1317. doi:10.1002/acp.1537
6. Nakata, T., & Webb, S. (2016). Does studying vocabulary in smaller sets increase learning? Studies in Second Language Acquisition, 38(3), 523–552. doi:10.1017/S0272263115000236
7. Nation, P., & Waring, R. (1997). Vocabulary size, text coverage and word lists. In N. Schmitt & M. McCarthy (Eds.), Vocabulary: Description, acquisition and pedagogy (pp. 6–19). Cambridge University Press.