The ten-minute post-lesson review, and why yours never happens

The standard 10-minute post-lesson review asks you to make seven decisions before you have reviewed a single word.

Close the lesson materials. Start writing what you remember. Go back and find the source to check yourself against. Sort your misses into got it, half got it, gone. Judge which of those are worth repairing. Say them aloud. Invent a question for tomorrow morning, and remember tomorrow that it exists.

Seven negotiations with yourself, every day, forever. On day 1, after a good lesson, you win all seven. On day 9, at 9pm, after a lesson that went badly and left you feeling stupid, you lose the first one and the other six never come up.

That is why your review habit died. Not the method.

Seven decisions, every day, forever

Every guide to this prices the routine at 10 minutes, and 10 minutes is nothing. The cost that actually decides whether it happens is the standing start before those minutes, and you pay that in full whether you end up reviewing for 10 minutes or 2.

Worse, the failure correlates exactly with the lessons that most deserve reviewing. A lesson where your tutor stopped you 8 times produces the richest material you will get all week, and the least appetite for going anywhere near it.

So the days you skip are the expensive ones.

An activation-energy problem

The page currently ranking first for this question closes its diagnostic section like this: "If none of that is happening, the issue is usually not effort. It is method."

That sentence is backwards, and it is the reason I wrote this. Someone typing "how to review after a language lesson" into Google has already decided reviewing is worth doing. They have a workable method sitting in the first three results. What they have is a folder of notes from March that has never been reopened.

Their problem is activation energy: the cost of getting started, paid up front, every single time, before any of the good stuff happens. Method is what you do once you have started.

And the fix for activation energy is structural. Shrink what the loop demands at the moment you are least willing to give it, then move everything else somewhere that runs without asking you first.

The method itself is right

I want to be clear that I am not arguing with the advice. The standard loop is good, and the research backs every part of it.

It is retrieval-first, which is the single most defensible thing you can do with 10 minutes. And the instruction to fix only the highest-value misses is better than most advice on the internet, because it tells you to skip things. Advice that tells you to do less is rare and usually right.

Roediger and Karpicke (2006) had students read a passage twice, or read it once and then take a free-recall test with no feedback and no second look. After 5 minutes the rereaders looked better. After a week the testers had them beaten, recalling roughly 14% more, having spent part of their study time staring at nothing. That is the whole case for closing the notes before you review them, and I cover it properly in the testing effect.

The step everyone skips is the delayed one. Two years later, Karpicke and Roediger (2008), in Science, gave students 40 Swahili-English pairs and varied what happened after a pair was first recalled correctly. Pairs that kept being tested came back a week later at around 80%. Pairs that were only restudied, or dropped, came back at around 36%. Same students, same words, same lab time.

Karpicke & Roediger 2008 · one week later

0%

recalled when pairs kept being tested after they were first learned

0%

recalled when pairs were only restudied, or dropped, after learning

Read those two numbers against your own routine. The 10-minute review, done perfectly, is the first retrieval, the one you were always most likely to do. The 36% column is what you get when the retrievals stop there.

The top result cites no science

A note on where that first result gets its confidence. The site is called Science Based Learning. I downloaded the page on 22 August 2026 and searched its raw HTML for a citation. Not a good citation. Any citation.

Zero DOIs. Zero journal names. Zero researchers. I searched for Roediger, for Karpicke, for Ebbinghaus, for "et al.", and finally for any four-digit year in parentheses, which is the loosest possible test for a page having read a paper. Nothing matched.

A page with no citations can still be pointing somewhere, so I pulled every link and script tag out of it. Nine distinct destinations sit off-site, spread across 6 hosts: 3 files from a CDN, 2 tracking scripts (a Google ad-conversion tag and a Facebook pixel), an App Store listing, a Google Play listing, and 2 URLs on the developer's own company site.

That is the whole map. The reader is never sent anywhere they could check anything, and the only destinations that are not plumbing are places to download the app.

The page does use the phrase "illusion of competence," credited to nobody. That term is Koriat and Bjork's (2005), for what happens when you rate your own knowledge while the answer sits in front of you. It states the testing effect as folk wisdom too, flattening 20 years of experimental work into a vibe.

Uncited is not the same as untrue, and one line buried in its weekly checklist, "Decide what to drop because it is already solid," is sharper than the section around it. But the page is asking you to change how you spend the hour after every lesson you pay for, and none of its claims comes with anywhere to check it. Every number above has a link. Click them.

What to protect when you only get one minute

Back to the seven decisions. Assume the worst version of yourself: 60 seconds, taken while walking to the kitchen. What survives the cut?

Capture, and only capture. Everything else on that list is a later problem. Get the words out of your working memory and into somewhere that will come find you again, then walk away.

• Capture immediately, review whenever. The capture has a hard deadline, because your recollection of which correction mattered decays fast. The review does not. Splitting them means only one of the two is urgent.
• Capture the sentence, not the word. Write me hubiera gustado inside the sentence you were failing to build when it came up. A bare word list is a set of prompts you will be unable to reconstruct in a week, and context is doing real work here (see context-dependent memory).
• Capture your version too. The wrong thing you said, next to the right thing. That pair is the highest-value item in the whole lesson and it exists nowhere else in the world.
• Cap it at 8 items. A 40-minute lesson can generate 30 candidate words. Taking all 30 is how you end up with a backlog you avoid. Take the ones your tutor stopped for.

Then the reviews have to be scheduled by something outside your head. "Tomorrow morning" is a promise. What you want is a queue that puts the word in front of you at widening intervals until it holds, which is what the 80% column is made of, and why a single session the night before your next lesson does not substitute for it.

Say them out loud when you review, too. Producing a word aloud beats reading it silently by a margin that has been replicated for decades, which is the production effect, and you are training for speech anyway.

The industry has decided this part is your job

Here is the thing I find genuinely strange. The post-lesson hour is where the money you spent on the lesson either converts into memory or evaporates. And the platforms selling you the lessons have openly declined to build for it.

italki, one of the largest of the tutor marketplaces, says so on its own blog: "It's up to you to take the language you learned in your italki lesson and add it to a smart platform to ensure you don't forget your new vocabulary."

Read that as an admission and it is a remarkable one. The company that took your money for the lesson is telling you that keeping what happened in it is unpaid work you will be doing elsewhere, on a platform they do not name and do not provide.

I built Verbamor because I got tired of being the unpaid part. You record or type what came up in the lesson, and it turns those words and phrases into fill-in-the-blank cards with native audio and an image, scheduled by FSRS. You decide once, at capture, and the queue arrives on its own after that. It is iOS only, there is no free tier, and it teaches you nothing your tutor did not. Its entire claim is that it removes the standing start, which is a smaller claim than most apps make and the specific thing that was killing my reviews.

It also costs you something real to start here. There is no Anki import, so whatever deck you have already built stays where it is, and mine begins empty from your next lesson forward. If you have 2,000 cards you like, that is a genuine reason to stay put.

And there is a reader I would tell to close this tab: the one who keeps a notebook by the laptop, writes the words in it during the lesson, and opens it again on Thursday. A paper notebook and 10 honest minutes beats any app you will not open, mine included, and it costs nothing. If that is you, you already have the thing I am selling. The same goes for Anki, which is free and is better than anything I could sell you if you will actually build the cards.

The question that decides it is what you did last Tuesday at 4pm, and you already know the answer to that one.

---

Sources

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