Reading time: about 9 minutes
Picture the night before an exam, or the evening before a big presentation. Most people do the same thing: they open their notes and read them. Then they read them again. Highlighter in hand, they move through the material, and each pass feels a little more familiar, a little more comfortable. By the end it all feels known.
Then the moment comes to actually produce the information — and it is not there. The familiarity evaporates. You recognize the material when you see it, but you cannot summon it when you need it.
This gap between recognizing and recalling is one of the most important — and most ignored — facts about how memory works. And the fix for it is almost the opposite of what most of us instinctively do. Instead of putting more information in by re-reading, the most reliable way to make knowledge stick is to try to pull it out: to test yourself, from memory, before you check the answer. Psychologists call this retrieval practice, and the improvement it produces is known as the testing effect.
What the testing effect actually is
The finding, stated plainly: trying to recall something strengthens your memory of it more than simply studying it again for the same amount of time.
In the classic style of experiment, people study a passage or a set of facts. Then one group re-reads the material, while another group puts it away and tries to recall as much as they can. In the short term — tested minutes later — the re-readers often look just as good or better. But when everyone is tested again days or a week later, the pattern reverses, often dramatically. The group that practiced retrieving the information remembers substantially more. The re-readers, who felt more confident, have forgotten far more of it.
This is not a fragile result that shows up in one lab and nowhere else. Retrieval practice is one of the more robust and repeatedly replicated findings in the science of learning, seen across ages, subjects, and formats — from vocabulary and facts to concepts and problem-solving. The effect is strong enough that many cognitive scientists consider self-testing one of the highest-return study strategies available, and re-reading one of the lowest.
It helps to be precise about the claim, because it is easy to overstate. "Testing" here does not mean high-stakes, graded exams. It means the act of retrieval — quizzing yourself, using flashcards, closing the book and writing down everything you remember, explaining an idea from memory. A low-pressure attempt to recall is the active ingredient. The grade attached to it is not.
Why pulling information out beats putting it back in
The intuitive model of memory is that it works like a recording: studying "saves the file," and remembering "plays it back." If that were true, then re-reading — re-saving the file more times — should be the best strategy. But memory does not work like a recording, and retrieval is not a neutral playback.
When you successfully retrieve a memory, you are not just reading it off a shelf. You are reconstructing it, and that act of reconstruction changes the memory itself — strengthening it and making it easier to find next time. Each effortful retrieval seems to build and reinforce the routes that lead back to that information, so it becomes more accessible in more situations. Re-reading, by contrast, mostly strengthens your sense of familiarity with the words on the page, which is not the same as being able to produce the idea on your own later.
The learning researcher Robert Bjork draws a useful distinction between two things we usually lump together: how well-learned something is (its storage strength) and how easily you can call it up right now (its retrieval strength). Re-reading nudges the feeling of familiarity — a form of retrieval strength in the moment — without doing much for durable learning. Retrieval practice, precisely because it is harder, does more for the underlying storage. This is the family of phenomena Bjork calls "desirable difficulties": conditions that make practice feel slower and more effortful but produce more durable, flexible learning. (Spacing your practice out over time and mixing related topics together are two other members of the same family — see how to use AI for spaced repetition.)
The fluency illusion: why the better method feels worse
If retrieval practice works so well, why does almost no one do it by default? Because it feels bad — or at least, it feels less productive — and our feelings about learning are a famously unreliable guide.
Re-reading produces a sensation of fluency: the material flows past smoothly, you recognize every line, and your brain interprets that smoothness as "I know this." Retrieval practice produces the opposite sensation. You stare at a blank page, you strain, you get some of it wrong, and it feels like you are failing. Studies that ask people to predict how well they will remember consistently find that people expect re-reading to work better than self-testing — the exact reverse of what actually happens on a later test.
This is worth sitting with, because it means your confidence during studying is not a trustworthy signal. The strategy that feels most effective is often the one leaving the least behind. Effort and a bit of struggle — as long as you are mostly succeeding — are signs the desirable difficulty is doing its job, not signs that you are doing it wrong. It is the same trap that makes it so easy to let an AI tool hand you a smooth answer and mistake reading it for understanding it, a point worth its own discussion in using AI without dulling your brain.
When retrieval practice works best
Retrieval practice is powerful, but it is not magic, and the details matter. A few conditions make it far more effective:
Retrieval should be effortful but mostly successful. The benefit comes from actually pulling the information out. If the material is so new or so hard that you retrieve almost nothing, straining against a blank produces little — you need enough initial exposure to have something to retrieve. Do a first read or a first explanation to get the material in, then switch to testing yourself. The sweet spot is retrieval that makes you work but succeeds more often than it fails.
Feedback multiplies the effect. Testing yourself and then checking the correct answer is considerably better than testing alone — especially for anything you got wrong or only half-remembered. Retrieval without feedback risks quietly reinforcing an error. Always close the loop: attempt from memory first, then check the source.
Space it out. Retrieving something once helps; retrieving it again a day later, then a few days after that, helps far more. Retrieval practice and spacing are complements, not competitors — the most durable results come from testing yourself repeatedly across expanding intervals rather than cramming all your self-testing into one session.
The act matters more than the format. Flashcards, practice questions, a blank-page "brain dump," explaining the idea aloud to someone, or writing a short summary from memory all work, because all of them require you to generate the answer rather than merely recognize it. This is also why teaching or explaining a concept from memory is so effective — it is retrieval practice in disguise (the mechanism behind the Feynman technique). Where you can, prefer formats that make you produce the answer (open recall) over ones where you just pick it out of a list (recognition).
The honest limits
A science-backed strategy deserves a science-honest account of where it stops.
Retrieval practice builds durable memory for retrievable content — facts, concepts, procedures, vocabulary, the material of most courses and most skills that rest on a body of knowledge. It is not a substitute for the hands-on, feedback-driven repetition that builds physical or performance skills (that is the terrain of deliberate practice), though even there, recalling the relevant knowledge helps.
It also is not a trick for learning something you were never exposed to; you cannot retrieve what was never encoded. And cramming a burst of self-testing right before a deadline will give you a short-term boost without the durable payoff — the long-term benefit depends on spacing. Finally, the low-stakes retrieval practice discussed here is different from high-pressure, high-stakes testing, which can carry anxiety costs of its own. The goal is frequent, low-stakes recall for your own learning, not more exams.
None of these caveats undo the core finding. They just define its shape: retrieval practice is a genuinely high-leverage habit for building knowledge that lasts, provided you use it as practice rather than as a one-time cram, and pair it with feedback and spacing.
How to use it — with AI as your quizmaster
This is where a modern AI tool becomes genuinely useful, because the historical friction of retrieval practice was making the questions. Writing good self-test questions from your own notes is tedious, so most people skip it and re-read instead. AI removes that excuse.
A practical routine:
-
Turn your material into questions — but keep the answering to yourself. Paste your notes, a chapter summary, or a transcript into an AI tool and ask it to generate a set of recall questions. Crucially, ask it to give you the questions only, and to hold the answers until you ask. The moment the answer is on screen, the retrieval benefit is gone.
-
Ask for recall, not recognition. Prompt for open-ended questions that make you produce and explain ("Explain why X happens" or "How would you apply Y to this situation?") rather than multiple-choice questions where you can just recognize the right option. Recognition is easier and feels nicer, and does less.
-
Answer from memory first, out loud or on paper. Do the effortful part before you check anything. Get it wrong, get it partial — that is the work happening.
-
Then check, and get feedback. Reveal the source or ask the AI to evaluate your answer and point out what you missed. This is the feedback step that makes retrieval practice so much stronger.
-
Come back tomorrow, and next week. Have the AI re-quiz you on the same material after a gap, focusing on what you missed last time. Now you are combining retrieval practice with spacing — the two together outperform either alone.
Two cautions. First, verify the AI's "correct" answers against a trustworthy source, especially for anything technical or high-stakes: a confident, fluent, wrong answer is exactly the kind of error these tools make. Second — and this is the whole game — resist the pull to let the tool do the remembering for you. The point of a quizmaster is to make your brain do the retrieving. An AI that hands you smooth answers you never had to reconstruct will leave you feeling knowledgeable and being no such thing. (For more on why forgetting is not failure and how to work with it rather than against it, see why you forget and how to remember.)
The one-sentence version
If you want to remember something, stop re-reading it and start trying to recall it — from memory, before you check, again and again over time. It will feel harder and less productive than it is. That feeling is the sign that it is working.
This article is for general educational purposes and is not medical or psychological advice. If you want to build these evidence-based learning habits into a structured practice, our AI Literacy for Everyone course shows you how to use AI as a genuine thinking-and-learning partner rather than a shortcut around the work.