How many new words a day is realistic, and what sets the limit

Five to fifteen new words a day is realistic for most people, and the number itself is the least interesting part of the answer. What sets it is not how many words you can absorb in one sitting — it is how many reviews those words hand back to you three weeks from now, on a day you did not choose and may not want.

That single fact explains the shape of nearly every abandoned deck. A first fortnight at thirty words a day feels easy. The fifth week arrives with a queue that is already too long before you have added anything, and something gives. Nothing went wrong with your discipline. The bill simply came due late.

A new word is a standing appointment, not a single decision

Adding a word today does not buy you one encounter with it. It commits you to a series: today, again tomorrow or the day after, then a few days later, then a week, then a fortnight, then a month, widening for as long as you keep the card and keep getting it right.

So the queue you face on any given morning is not made of today's words at all. It is made of every previous day's words, each sitting at whatever point in its own series it has reached. Ten cards added on Monday reappear in fragments across the following two months.

Add at a steady rate for long enough and that daily queue stops growing. The cards graduating to long intervals roughly balance the ones arriving from last week, and the count settles. Where it settles is your actual workload, and it is always some multiple of your daily new-card number — never the number itself.

Nobody can tell you your multiple in advance. It depends on how often you answer your own cards correctly, which depends on how good the cards are and how hard the language is for you right now. But you can measure it, and measuring takes three weeks rather than a theory. That is the last section of this article.

The bill arrives about three weeks late

The first week is misleading by construction. On day three you have almost no reviews, because almost nothing has come back yet. Nearly all your study time goes to new words, which is the enjoyable part, and the session feels short. This is the moment most people conclude they could handle more.

By week three the early cards are cycling at several intervals at once, and the queue you meet each morning is mostly reviews you scheduled for yourself a fortnight ago, at a rate you set when you had no idea what it would cost. The feedback loop is slow enough that intuition never learns from it: the day you feel capable of thirty is precisely the day you have no evidence about thirty.

This is why a review app can feel like two entirely different products in the same month, and why the honest way to choose a rate is to pick a low one and refuse to change it until the queue has stopped moving.

Your ceiling is your worst day, not your best

A rate you can only sustain on a good day is not a rate. It is a good day.

The asymmetry is worth stating plainly, because it decides everything. Reviews you skip do not disappear — they roll forward and land on top of tomorrow, which makes tomorrow worse than today, which is how a two-day gap becomes a four-hundred-card wall and then a deleted app. New words you skip cost you nothing at all. The word is still out there. You can add it next week.

So the new-card rate is the only lever in the whole system that is free to pull. Set it against the Tuesday you did not choose: the delayed train, the bad night, the evening that vanished. Whatever you can clear then, without negotiating with yourself, is your number. Not what you can clear on a Sunday morning with a coffee.

Cards you keep failing cost more than cards you keep adding

Two people can add ten words a day and have completely different daily queues, because a card you fail does not go away — it resets to a short interval and comes back almost immediately. A handful of cards you get wrong every single time can generate more daily reviews than a week of new additions.

When your queue keeps climbing at a fixed new-card rate, the volume is not usually the culprit. The cards are. And in practice the failing card is almost always failing for one of three reasons:

• It holds two things. 適当 meaning "appropriate" and 適当 meaning "slapdash, done any old way" are one word and two opposite pieces of advice. As a single card it is a coin flip forever.
• It arrived with no context. A bare word and a bare gloss, saved from a list, with nothing to recall it by. There is no scene, no sentence, no reason you cared.
• It is not actually a word you need yet. Sometimes the honest fix is to delete it and meet it again later, when something in your life puts it in front of you.

Fixing or dropping five bad cards will do more for tomorrow's queue than cutting your new-card rate in half, and it costs you nothing in progress.

Ten words is not a unit of work

Counting cards is a convenient proxy for effort, and a poor one — especially in Japanese, where two cards added on the same afternoon can differ by an order of magnitude in what they demand.

準備中 on a shutter is nearly free if you already hold 準備 and 中. You are not learning three characters; you are learning that two things you know combine into "closed for now", and the reading follows from parts you have already paid for.

ラストオーダー is cheap once, and then it teaches you something that keeps paying: how English gets rebuilt to fit Japanese sound. The word is not the work — the transformation is.

大丈夫 is a different animal. Three characters, a reading you cannot derive by looking at them, and a usage range that a two-word gloss actively misleads you about, since it covers "I'm fine", "no thanks" and "that'll do" depending on where it lands.

生ビール, 学生 and 生まれる are three cards and one character, teaching three readings of 生 that no amount of staring at the character alone would have given you — which is why the word, not the character, is the unit worth collecting.

Ten of the first kind is a light day. Ten of the third kind is not. If you want a number that transfers between weeks, think in minutes of review rather than in cards.

Words you met yourself raise the ceiling

Where a word comes from changes how expensive it is, and this is the one lever that genuinely raises your limit rather than just moving it around.

A word you met in your own week arrives with a scene attached: the station, the shutter, the person who said it, what they were pointing at. That scene is a second route to the same memory, so the card needs fewer successful recalls before its intervals start to stretch. A word taken from a list arrives bare, fails more often, and stays in the short end of the queue longer. Same card count, more work, less retained.

There is also a natural filter in it. Your own life hands you fewer candidates than any list will, and it has already run a test no frequency ranking can run: you actually met the thing. Keep what you have encountered twice, or what blocked you from understanding something you cared about, and the daily inflow tends to settle near a sustainable number without you enforcing one.

That only works if capturing a word costs seconds. This is what Kikusho is built around: point the camera at the sign, say the phrase while it is still in your head, or paste the text, and a complete card comes back — word, reading, meaning, register, example, audio — which you check before it is saved. On one photo you can highlight up to five things and get five separate cards from a single capture. The free plan caps creation at five captures a day and never caps reviewing — what you have already collected stays reviewable, which is the right way round given everything above.

Zero is a setting, not a failure

There will be weeks when you are travelling, ill, or simply out of room. The correct move then is to set new cards to zero and keep clearing the queue, for as long as it takes.

This feels like stopping. It is not: reviews are where the learning actually happens, and the deck stays alive as long as they continue. What kills a deck is the opposite trade — adding new words on a day you could not finish yesterday's reviews, which buys a pleasant twenty minutes and guarantees a worse Thursday.

The same move is the cure for a backlog. Not deleting the deck, not "starting fresh": new cards to zero, review daily, let the queue drain. It will take less time than you fear, because most of a backlog is cards you actually still know.

How to find your own number in three weeks

The protocol is dull, which is why it works.

1. Start at five a day, even if that feels insulting. You are measuring, not competing.
2. Change nothing for three weeks. Any adjustment before then is a reaction to a queue that has not finished growing.
3. Watch the daily count, not the clock. When it stops climbing week over week, you have found the steady state that five a day buys you.
4. Compare it to your worst day, not your best. If you could clear that queue on the bad Tuesday, you have room.
5. Raise by five, then hold three more weeks. Repeat until the queue is as large as a bad day can absorb. That is your number, and it is allowed to be smaller than someone else's.

Five to fifteen is the range this usually settles into, and the low end is the safer place to sit. The rate you keep for a year beats the rate you keep for a month, and the difference between them is decided by how the timing of reviews works, not by how much you can hold at once.