How to turn a language lesson into flashcards that survive the week

Your tutor spent 10 minutes untangling por and para for you. You got it. Two weeks later you're staring at a sentence, picking one at random, and watching your tutor's face for the flinch.

Flashcards fix this, but only if they're built out of the right 9 things and phrased so they make you produce rather than nod along. Most of the work is in the choosing and the phrasing, and it takes about 20 minutes.

Here's the whole method. No tool required.

Capture: what to write during the lesson

Write almost nothing while the lesson is running. Two or three words per item, enough to rebuild it later. "por/para thing". "se me cayó". "pretérito on 'saber'?"

Full sentences during a lesson cost you the thing you're paying for. You're transcribing instead of listening, and the tutor is waiting.

What earns a jot is narrow:

• Words you reached for and missed. The gap you hit mid-sentence is the highest-value item in the whole hour, because you've already proven you want to say it and can't.
• Corrections to mistakes you keep making. If your tutor has fixed the same gender twice, that's a card.
• A one-off slip, though, is just a slip. You said el mano once at the end of a tiring hour. Let it go. Cards are for patterns.
• Phrases your tutor produced that you wouldn't have. The natural construction you'd have built clumsily out of dictionary parts.

Recording the lesson makes this easier, since you can jot even less and rebuild from audio. If your tutor is fine with it, ask.

Plenty of people arrive at this step with something other than a notepad, and the move is the same in every case: get it into one plain list, then cut. If you used italki Classroom, the private notes pane and your teacher's saved notes are already on your account, so paste both into one document. If you have a recording, scrub it rather than transcribing it, because the moments worth keeping are the ones where you stalled or got corrected, and those are audible.

A shared Google Doc or PDF from your tutor is usually a curriculum rather than a record of what you got wrong, so mine it for the 2 or 3 things you personally fumbled and ignore the rest. Same with a screenshotted whiteboard: retype the handful of items you couldn't produce today. The cut in the next section is doing the real work, and it only works on a list you can see all at once.

Cut: 30 items down to 9

Do this within a few hours of the lesson, while "por/para thing" still means something. By Friday it's a note from a stranger.

An hour surfaces 30 or more keepable items. I keep about 9, and the number is mine, not a finding. The papers further down this page say nothing about how many cards to build; the 9 is a working figure from my own lessons, and the failure I see most often is a learner who kept all 30, faced a 30-card pile on Tuesday, and never opened it again.

Cut these without ceremony:

• Words you already recognise. Recognising ventana when you read it is not the same as producing it, but if it's already coming out of your mouth unprompted, it doesn't need a card.
• Words you won't say this month. Alcachofa is a fine word. Do you talk about artichokes? Word frequency in any language is lopsided enough that your early effort belongs on the words you'll actually reach for, which is what the frequency curve is about.
• Anything a dictionary would have told you. If you could have looked it up without the lesson, the lesson wasn't where the value was.
• Grammar phrased as a rule. More on why below.

Deciding an item doesn't deserve a card requires judging whether you can already produce it. That's a retrieval attempt wearing a disguise.

Phrase: the blank is the whole design

This is the step that decides whether the cards work, and it's the step every three-click tutorial skips.

Take the sentence the item appeared in during the lesson. Delete the target word. Leave a gap. Put a short English hint underneath. That's the card.

The same item, four ways

Weak se me cayó = I dropped it A word pair. You read both sides at once and feel like you knew it. Tests nothing.

Better "I dropped it" → ? Now you have to produce something. But produce what, exactly? Several answers fit.

Right Perdona, ______ el vaso. "sorry, I dropped the glass" · One blank, one answer, and the sentence around it tells you which register you're in.

Generated Perdona, ______ el vaso. Same shape, pulled from the lesson audio: native recording of the full sentence, image of a dropped glass, article handled. Written by a machine after the call ended, by someone who wasn't you.

Two rules govern the blank, and a third thing that isn't quite a rule.

One blank per card. Two gaps in a sentence means a half-right answer, and a half-right answer is a grade you can't give honestly. That matters more than it sounds, because the scheduler believes whatever grade you feed it.

Use the sentence from your lesson, not an invented one. You already have a memory of hearing it, in your tutor's voice, in a conversation that went somewhere. That context is doing work a dictionary example can't.

The third: the card has to be answerable out loud before you look. Say the word, then check. Recognition after the flip is the weaker test. In Karpicke and Roediger's 2008 Science study, students learned Swahili-English pairs and then either kept being tested on an item or kept restudying it after their first correct answer. A week later, repeated testing had produced a large positive effect on recall. Repeated studying had produced none at all.

A two-sided card only tests retrieval if you supply the discipline every single time. A blank supplies it structurally.

What to add after the blank

Each of these is worth 20 seconds.

Put the article inside the blank. Write ______ ventana and the answer is la ventana. Gender learned separately from the noun is gender you'll guess at in conversation, and you'll guess wrong at roughly the rate you'd expect.

Get the sentence in audio. Ask your tutor to record the sentences at the end of the lesson, or record them yourself off the lesson audio. Learn a word from its spelling first and you tend to build a pronunciation you then have to unlearn. Bassetti and Atkinson measured that in 2015, on learners who'd had years of instruction: the written form was still bending how they said the word. The rest of the argument is in the native audio post.

Add an image where the word is concrete, and don't trust it. This one comes with a caveat most guides skip. Carpenter and Olson tested pictures against native-language translations for Swahili vocabulary and found pictures didn't beat translations at first. What they did do was make learners more confident than they should have been. Once that overconfidence was corrected, by retrieval practice or a plain warning, pictures helped. So draw the sketch, and then grade yourself harshly on the cards that have one. The longer argument is in the dual coding post. Skip images for abstractions: nobody draws sin embargo.

Review: the schedule that holds it

Cards you don't review are craft supplies. Four touches, about 18 minutes across the week:

The night touch is the one that looks like filler and isn't. Gais, Lucas and Born ran two experiments on high-school students learning English-German vocabulary, and tested recall 48 hours after learning. Students who slept within a few hours of learning beat the ones kept awake that first night. Everyone got a recovery night before the test, and the deprived group was still behind, which is what rules out the boring explanation that they were simply tired on test day. More in the sleep post.

Cepeda and colleagues went through 317 experiments across 184 articles on distributed practice. One finding is worth carrying around: the best gap between sessions gets longer as the deadline you're studying for gets further away. For a language you intend to speak for years, that argues for gaps measured in days and then weeks, not one long Sunday.

If a card fails 4 times running, reread the card. Usually the blank is ambiguous, or you've packed two ideas into one sentence. Rewrite it. There's more on this in the guide on why lessons evaporate.

Four cards I'd throw away

Concrete failures, because "make good cards" is advice you can't act on.

• "The subjunctive is used after expressions of doubt." You can recite this and still freeze mid-sentence. Blank the verb in No creo que ______ verdad instead.
• "hacer = to do, to make, to cause, and in weather expressions." Four meanings on one card means you'll answer with whichever surfaces first and call it a pass. Split it, or blank hacer inside one specific sentence.
• A blank with 5 valid answers. Ayer ______ al cine accepts fui, fuimos, íbamos. Add enough sentence to pin it down: Ayer yo ______ al cine solo.
• The whole lesson transcript, chopped into cards. Every sentence your tutor said is not worth 9 months of reviews. This is the failure mode automation walks straight into if nobody's filtering.

What automating this costs you

Here's the part the tutorials selling you software leave out.

Choosing which sentence to use, deciding where the blank goes, working out what hint pins the answer down: that's you processing the meaning, and it lays down a memory before you review the card once.

Slamecka and Graf named this the generation effect in 1978. Material you generate yourself is remembered better than the same material you simply read. As far as I can tell it has held up since.

So a generated card starts behind. Look at the fourth card in the figure above: it's a perfectly good card, arguably better built than the one I wrote by hand, with audio and an image I'd never have bothered to add on a Tuesday. And it cost you the encoding. The first time you meet that sentence is the first time you review it.

What automation gives back is consistency and volume. Every card gets the article, the audio and the image, on the day, whether or not you felt like it. The cards exist during the weeks when you'd have built nothing.

Which one wins depends on a question only you can answer: were you actually going to build them? Twenty minutes after every lesson, in week 7, in the week you're travelling. Most people are honest about this if you ask them in November rather than January.

Manual review breaks in week 3. Three lessons in, you have 27 cards and no signal about which 6 are fading today. So you review everything (slow, and mostly wasted on cards you know cold) or you review nothing.

That's the job software actually does. FSRS models each card's decay separately and puts it in front of you near the point of forgetting. Anki does it well, once you enable it, since Anki ships SM-2 by default and FSRS is a setting you turn on. Anki also ships no language content at all, so you're back to building every card yourself, which for this particular workflow is either the point or the problem.

Verbamor sits at the other end: it records the lesson, mines it for the words and phrases you actually stumbled on, and writes the cloze card with native audio and an image attached. That fourth card in the figure is what comes out. You lose the generation effect. You gain the 9 cards existing on a Tuesday in November.

Both trades are defensible. Pick the one that matches how you actually behave in week 7.

If you'd rather run this on paper, the free handbook has the capture sheet and the debrief as printable pages, with no purchase attached.

Questions people actually ask

---

Sources

Every load-bearing claim Verbamor makes is traced to its paper on the research page.