Make the card while the sentence is still in front of you. A word captured on Tuesday and turned into a card on Sunday arrives without the sentence it came in, so what gets built is a word and a translation, which is the format everybody agrees does not work and is what a batch always produces.
The problem is not laziness. It is that the capture and the card are two different jobs on two different timescales, and trying to do them in one sitting means one of them happens late.
Eight rules for cards that actually work
- one fact per card. A card asking three things is failed for one reason and reviewed for all three.
- put the word on the back, not the front. Prompting with the meaning and answering with the word is production; the reverse is recognition and is much weaker.
- no card without context. A bare word pair teaches a translation rather than a usable word.
- include a sentence you wrote, about something real, which outperforms a dictionary example by a wide margin.
- cloze deletion, a sentence with the target word blanked out, which tests the word in place rather than in isolation.
- add a picture rather than a translation where the word allows it, because it attaches the meaning directly.
- make your own cards. The act of making one is a large part of the learning, which is why bought decks underperform so consistently.
- keep the deck small. Two hundred good cards reviewed daily beats two thousand reviewed occasionally.
Split the capture from the making
Capture has to be fast enough to happen mid-page: the sentence, pasted or typed, with no thought about the card. Making the card can wait for the evening or the weekend, because by then you still have the sentence and the decisions are quick.
That split is the whole technique and it is what separates a deck built from your own reading from a deck downloaded from somebody else. The downloaded one has better cards, in the sense that they are more consistent, and every item in it arrived attached to nothing.
One fact per card, or the failures multiply
A card that asks for three things fails if any one of them fails, so the chance of getting it wrong is the chance of missing the word, plus the chance of missing the gender, plus the chance of missing the preposition. A card built out of three near-certainties is not a near-certainty.
That is why a card asking for a word, its plural and its article turns into the card you dread. It is not a hard item, it is three items chained together, and the chain is only as strong as the weakest one. Split it and each part settles on its own schedule.
The same arithmetic applies to a card whose front does not narrow the answer to one thing. Several correct answers means several ways to be marked wrong, which is a failure generated by the card rather than by your memory.
Where it stops: making them in a batch
The stall is Sunday. A learner collects words diligently all week in a notebook, sits down at the weekend, and turns forty of them into cards in an hour. The session feels productive and the deck grows.
What was lost happened days earlier. The sentence, the reason the word mattered, the situation it appeared in, and often the sense it was carrying, because a word met in one meaning is written down as a word and looked up in a dictionary that lists five.
Nothing signals it, because the cards look fine. They are neat, they are consistent, and they behave exactly like the word pairs that this whole practice is supposed to have replaced.
The buffer has to be faster than the card
The notes app already on your phone is enough, and its virtue is that it is two taps away. Share a sentence into it, or type five words, and go back to what you were doing. The requirement is speed rather than structure: anything that takes longer than the lookup did will not be used at the moment it is needed.
It replaced making cards in a batch at the weekend, by which time the sentences have gone.
The limitation is that a buffer that never gets processed is a graveyard, and a note file with four hundred untouched lines in it is a common and demoralising object. Process it weekly, keep the ones you still care about, and delete the rest without guilt: a word you no longer recognise from your own note was not a word you needed.
What it looks like when it goes wrong
The recognisable failure is a card you get right in the deck and cannot produce in a sentence. That is a card built from a word rather than from a use, and it is the most common thing in most decks.
The second is the card everybody has: the one that has been failed a dozen times and is still there because deleting it feels like giving up on the word. It is almost never a hard word and almost always a card asking for more than one thing.
How often any of these come back, and why the intake rate matters more than the schedule, is reviewing a word exactly as you are about to lose it, and the reason a card built from a sentence survives when a word pair does not is the curve everyone is fighting, and what bends it.
What no card can do is put the word into your mouth under time pressure, which is a different mechanism entirely and needs another person in the room: finding someone to speak to, and what to do in the hour.