There is a card in your deck that you hate. You know the one. You've seen the front of it so many times that your stomach tightens before you flip it. You've failed it, relearned it, felt a flicker of triumph, and failed it again a week later. Somewhere along the way you started to suspect the problem was you: your memory, your focus, your age.
It probably isn't. That card is almost certainly broken, and the fix is rarely "try harder."
What a leech actually is
Spaced repetition users have a name for this card: a leech. It feeds on your study time and gives nothing back. Anki, the best-known spaced repetition program, flags a card as a leech by default after it has been forgotten eight times. The number is a convention, not a law of nature. But the idea behind it is sound: a card that keeps failing is telling you something.
The usual response is to grit your teeth and keep drilling. That response feels virtuous, and it is a trap. Each failed review costs you a minute or so, adds a small dose of frustration, and teaches you to associate studying with failure. A handful of leeches can sour a whole deck.
The better response starts with a diagnosis. In my experience, and in what the research on memory would predict, leeches tend to fall into a few recognizable types.
Type one: the card asks too much
Piotr Wozniak, who built the algorithm behind the original SuperMemo and wrote a widely read set of guidelines on formulating knowledge, called one principle the minimum information principle. Keep each item as simple as possible. A card that asks you to recall a definition, three exceptions, and a date is really four cards wearing one coat.
The reason this matters is how retrieval works. Recalling something is an all-or-nothing event at grading time. If you remember three of the four pieces, you still have to mark the card as failed, and the scheduler treats it as if you'd forgotten everything. Your memory is partly working and your deck is reporting that it isn't.
The fix is to split the card. Take "What are the causes of the French Revolution?" and break it into one card per cause, each with a specific prompt. Many leeches stop being leeches the moment they're cut into pieces small enough to fail cleanly or succeed cleanly.
Type two: the card is a near-twin of another
If you're learning two similar things, such as two chemical names that differ by one syllable or two Spanish verbs that mean nearly the same thing, the memories compete. Psychologists call this interference, and it is one of the best-documented reasons we forget. When you try to recall one, its twin comes to mind instead and you answer with confidence and wrongly.
The trouble is that more repetition of either card in isolation often doesn't help, because the confusion lives in the relationship between them. What helps is putting the contrast into the card itself. Write a card that asks directly, "How is X different from Y?" Or add a distinguishing cue to each: a sentence, an image, a rule of thumb that only one of them satisfies. You are giving your memory a fork in the road instead of hoping it picks the right turn.
Type three: the card has no hooks
Some facts are arbitrary. A phone number, a drug name, a vocabulary word with no resemblance to anything you know. Raw, disconnected information is the hardest kind to hold, because there is nothing for it to attach to.
Research on levels of processing going back to Craik and Lockhart suggests that thinking about meaning produces more durable memory than dwelling on surface features. Elaboration is the practical version: connect the new thing to something you already know, and make the connection yours.
For an arbitrary vocabulary word, that might be the keyword method, where you invent a vivid image linking the sound of the foreign word to the meaning. For a medical fact, it might be a one-line story about a patient you can picture. The image doesn't have to be elegant. It has to be yours, and it has to be specific. A silly picture you made yourself usually outperforms a tidy one you copied from someone else, partly because the act of making it is itself a form of processing.
Type four: the card tests recognition, not recall
A subtler leech: the card feels familiar when you see it, so you think you know it, but the answer isn't actually retrievable on demand. You see "Mitochondria" on the front, your brain says "yes, I know that," and then the back reveals a detail you can't produce. That feeling of familiarity is not the same thing as being able to retrieve, and the gap between the two is a classic source of overconfidence.
If you keep failing a card whose front seems so familiar, rewrite the prompt to demand a specific answer. "What does X do?" is weaker than "Which structure produces most of the cell's ATP, and how?" The question should leave no room to coast on a sense of recognition.
Type five: the card isn't worth knowing
This one is uncomfortable. Some leeches survive because you made them in a hurry, copied from someone else's deck, or because the fact simply doesn't matter to anything you care about. You are fighting for a memory you have no reason to keep.
Deleting a card is allowed. It's not a confession of weakness. It is the same editorial judgment a good teacher uses when deciding what goes on the syllabus. A smaller deck you actually review beats a perfect one you dread.
What the scheduler can and can't do
Modern spaced repetition algorithms, including FSRS, model each card's difficulty separately from its stability. A card you keep lapsing on gets marked as hard, and its intervals stay shorter. That's the algorithm doing its job: it adapts to the card.
But an algorithm can only schedule the card you wrote. It can't tell that the card is really four facts, or that it's confusable with another, or that it has no hooks. Scheduling can soften a bad card. Only editing can fix it. That part is yours.
Your next moves
- Find your worst five today. If your flashcard app shows lapse counts or flags leeches, sort by most failures. If it doesn't, write down the cards that make you groan during your next session.
- Diagnose each one in under a minute. Ask: Is it asking for more than one thing? Does it have a lookalike? Is there nothing to attach it to? Do I just recognize it? Do I even need it?
- Apply one fix per card. Split it, add a contrast, write a mnemonic image, sharpen the question, or delete it. Don't do all five at once; pick the single change your diagnosis points to.
- Make the replacement hard to misread. Phrase the front so only one specific answer is correct, and keep the back to a single short fact.
- Review your leech list weekly. Set a fixed ten-minute slot, for example Sunday morning, to deal with whatever got flagged. Left alone, leeches pile up; handled in small batches, they disappear.
A last thought on the card you hate
There is something quietly freeing in realizing that the card you've been failing isn't a verdict on your intelligence. It's feedback about a piece of writing. You wouldn't blame yourself for not understanding a confusing paragraph; you'd rewrite it, or you'd decide it wasn't worth reading.
That's the spirit behind Recall, the flashcard app we build. It uses FSRS to schedule reviews around how hard each card really is, and it makes editing a card in the middle of a session quick enough that fixing a leech takes seconds instead of a chore you put off. It also imports your existing Anki and Quizlet decks, so your problem cards can come along. If you'd like to see how it feels, you can find it at recall.lumenlabs.works. But the real work happens in the diagnosis, and you can start that tonight with whatever deck you already have.