The Words You Keep Forgetting Are a Card Problem, Not a Memory Problem

A word you have forgotten eight times is not telling you that your memory is bad. It is telling you that the card is unanswerable as written. Almost always the cause is on the card itself: an ambiguous prompt, two unknowns at once, or a word you have never met outside the app. Rewrite it and it usually stops failing.

The eight-failure rule

Anki, the most widely used spaced repetition program, has a name for these words. After you have learned and then forgotten the same card eight times, it tags it a “leech” and stops showing it until you deal with it. The threshold is adjustable, but eight is the default and it is a sensible one.

The word is well chosen. These cards are not just difficult, they are parasitic. A small number of them eat a disproportionate share of every session, because failing a card is what sends it back to a short interval. Fail it enough and it never leaves the front of the queue.

Most people meet their first leech without knowing there is a name for it. You recognise the word instantly. You know you have seen it, you know roughly how long you have been failing it, and you still cannot produce the meaning. That flash of recognition without retrieval is the diagnostic. It means the word has been stored as a shape you have seen before, not as a meaning you can reach.

Why the interval is almost never the problem

The instinct when a word will not stick is to review it more. More repetitions, shorter gaps, maybe a dedicated cram session for the difficult ones.

This rarely works, and it is worth being precise about why. Spaced repetition schedules when a card comes back. It has no opinion about what is on the card. If the prompt cannot be answered, because it is ambiguous or because you are missing something the card assumes you have, then the scheduler is choosing excellent moments to ask you an impossible question. Doing that more often does not make it answerable.

There is a useful distinction hiding here. A scheduling failure looks like forgetting words across the board after a break: you come back from two weeks away and everything is shaky. That really is an interval problem and it resolves on its own within a few sessions. An encoding failure looks completely different. A specific handful of words fail while everything around them is fine. Same intervals, same session, same you. The only variable left is the card.

If most of your reviews feel roughly right and five words are ruining them, the algorithm is doing its job. Something else is wrong.

What actually makes a word stick badly

Four causes cover nearly everything:

• The prompt is ambiguous. The front of the card admits more than one correct answer, so you are being graded on guessing which one was intended rather than on whether you know the word. Any word with several distinct senses does this by default.
• There are two unknowns at once. The example sentence contains a second word you also do not know. You cannot use the sentence to reach the target word, because the sentence is itself a puzzle. This one is easy to miss, because the card looks rich and well made.
• The word has no context at all. A bare headword and a bare gloss. Nothing to hang it on. No situation, no sense of who says it and when.
• You have no exposure to it. You have never met the word in reading or listening, only in the app. This is the honest one, and it is the only cause that no amount of card design will fix.

The last one deserves its own note, because it is the case where the right move is to stop working on the card. Vocabulary does not really get learned in isolation from the language it lives in. A word you have encountered only as a flashcard has exactly one association attached to it, the flashcard itself, and that association is not available when you are reading a novel or listening to someone talk.

Symptom, cause, fix

The failures look similar from inside a session, but they leave different traces. This is usually enough to tell them apart:

What it feels like	Likely cause	What to change
You give an answer, it is marked wrong, and you think “that was also correct”	Ambiguous prompt	Pin the sense: add the collocation or the part of speech you actually want
You stare at the example sentence and it does not help	Two unknowns on one card	Replace the sentence with one built from words you already know
You recognise the word but reach nothing	No context attached	Add a full sentence and the situation it belongs to
You have never seen it outside the app	No exposure	Suspend it, meet it in reading first
Everything is shaky, not just this word	Scheduling, after a break	Nothing, it settles in a few days

Note that only the middle three are card problems you can fix by editing. The first row is also a card problem but a subtler one, and the last row is not a problem at all.

Fix one: put the word back in a sentence

The single highest-value change is moving from a word-to-gloss card to a word-in-use card.

The reason is that a bare gloss stores a translation pair, and translation pairs are brittle. They give you one route to the meaning, and if that route is blocked, because you are tired or the word turns up somewhere unusual, there is no second route. A sentence adds several: the grammar around the word, the kind of situation it shows up in, the words it habitually keeps company with.

The sentence has to be a real one. Textbook examples written to demonstrate a word tend to be useless here, because they contain nothing but the word and its definition restated. “The erudite professor was very erudite” teaches nothing. A sentence someone would actually say gives the word somewhere to live.

If you want the longer argument about which of the two to store, we wrote about words versus sentences separately. For leeches specifically the answer is not close: sentences win.

Fix two: give the word a sound

A large share of stubborn words are stubborn because they exist only as spelling.

This is especially true for anyone who learned English mostly through reading, which describes a lot of adult learners. If you have never heard a word, your brain has stored a guess at its pronunciation derived from its spelling, and English spelling is an unreliable guide. So the written form and the spoken form end up filed as two separate items, and neither is strong. You will not recognise the word when someone says it, and the version in your head is not one you would ever say out loud.

Adding the sound gives the meaning a second, independent route in. It also tends to surface a specific and common failure: you discover you had the stress on the wrong syllable, and the word suddenly becomes easier to hold onto.

Every word in Vocaby ships with its IPA transcription and native audio for exactly this reason. Not as a pronunciation feature bolted on the side, but as a second encoding of the same word, so the written form is not carrying the whole load. You can see how a word page is put together on the word list.

Fix three: suspend it and let reading do the work

When rewriting has not helped, the correct move is to stop reviewing the word.

This feels like giving up and it is not. A word you have no exposure to cannot be forced into place by repetition, What is missing is not reps. It is the network of associations that only comes from meeting the word somewhere real. Suspending the card and reading more is a faster route to that than another twenty failed reviews.

The mechanics are simple. Suspend it, keep reading and listening at a level where you meet unfamiliar words regularly, and reintroduce the card when you have run into the word somewhere real. The second attempt usually goes fine, and the difference is not that you tried harder.

There is a version of this that goes wrong, worth naming: suspending a word and then never reading anything that would contain it. Then you have simply deleted the word with extra steps. Suspension only works if it is paired with input.

When to just let the word go

Some words are not worth the seat they occupy.

The test is whether you can construct a plausible situation in which you would meet the word. Not a sentence you invent for the exercise, but a real context: a book you read, a subject you work in, a conversation you have had. If nothing comes to mind, the word is probably below your frequency threshold, and there is a better use for that slot.

This is easier to accept once you notice that a leech is expensive in a way that a normal card is not. A card you know costs you a couple of seconds a month. A card you keep failing costs you a slot in every session, plus the small drag on morale that comes from meeting the same defeat repeatedly. Ten of them can make a deck feel unpleasant enough that you stop opening it, which is by far the worst outcome available.

Dropping a word is not a permanent judgement. Words come back when your reading changes.

What the app can and cannot do for you

Vocaby’s scheduler counts lapses per card and shortens the interval when you forget something, which is what a spaced repetition system should do. Spaced repetition is fundamentally a timing mechanism, and this is the timing working correctly.

But it is worth being straight about the limit: no scheduler can rewrite a card whose prompt is ambiguous, and none can give you exposure you have not had. Those are the two things that actually cause leeches. An app can put the word in front of you at a well-chosen moment, attach the sound so the spelling is not doing all the work, and get out of the way. The rest is a decision about the card, and it is yours.

I found this genuinely annoying to accept, because reviewing harder at least feels like effort and editing a card feels like admitting the deck was sloppy. But the words you keep forgetting are not evidence about your memory. They are the few cards you built badly, and cards can be rebuilt.

Get Vocaby on the App Store