The Reason You Study Wrong (and why the system never told you)

Somewhere in the last few weeks, you probably did one of these things.

You reread your notes before a test or an important meeting. You highlighted passages in a book and felt the satisfying sense of the information going in. You crammed the night before like I frequently do. It’s usually a mix of staying up late, running through the material repeatedly until it felt familiar, and walked in the next morning feeling reasonably prepared.

All of these strategies feel productive. They produce a belief that psychologists have a specific name for: the feeling of knowing. It is basically the comfortable sense that the material is in there somewhere and accessible.

The research says that this feeling is often wrong.

The strategies that produce the strongest feeling of knowing are frequently the strategies that produce the weakest actual retention. That stinks for all of us, but especially students trying to study and/or learn new material.

And, not only that, but the strategies that produce the strongest actual retention frequently feel like they are not working at all.

This is one of the most robust findings in the science of learning, and it has been replicated across hundreds of studies, across decades, across virtually every domain of knowledge and every population of learners. It has been established so thoroughly that the researchers who study it named the gap between the feeling and the reality, the illusion of fluency.

Most people have never been told it exists…

The experiment that changed how we understand memory

In 1885, a German psychologist named Hermann Ebbinghaus did something that no one had thought to do before. Weirdly, he simply ran controlled experiments on himself.

Ebbinghaus spent months memorizing lists of nonsense syllables and meaningless combinations of letters like DAX, BUP, and REB. Then he began testing his own retention at precise intervals (after twenty minutes, after an hour, after a day, after a week). He varied his learning conditions systematically and recorded the results with the rigor that his scientific training demanded. It was boring and repetitive work, conducted entirely in his own head, with himself as the only subject.

What he found was a curve. The forgetting curve, and one of the most replicated findings in all of experimental psychology. It showed that memory decays rapidly after initial learning, losing roughly 50 to 70 percent of newly acquired information within 24 hours without reinforcement, then decelerating into a long slow tail.

But Ebbinghaus also found something more important for us as learners. He found out that the curve could be interrupted. Each time he reviewed material at the right interval, the forgetting curve flattened. If the material was reviewed enough times, at the right spacing, information moved from fragile short-term traces into something more durable.

The implication is that when you practice matters as much as how much you practice. Study the same material for four hours in one sitting, and you will retain far less of it a week later than if you had studied for one hour on four separate occasions. The four-hour session produces more fluency in the moment. The four separate sessions produce more retention across time.

Researchers Robert Bjork and Elizabeth Bjork at UCLA have spent decades developing what they call the theory of desirable difficulties, which is the finding that the conditions that make learning feel most productive in the moment are frequently not the conditions that make it stick. And the conditions that make it stick frequently feel, in the moment, inefficient, uncomfortable, and not working.

This is the central paradox of memory. And you have almost certainly never been taught to navigate it.

I definitely wasn’t.

The three things that actually work

The science of learning has converged, with unusual consistency for a social science, on a small number of strategies that produce large, durable improvements in retention. They are not complicated. They are not expensive. They require no technology. And they are almost entirely absent from the way most schools teach students to study.

#1: Retrieval practice. The single most powerful learning strategy identified in the research is also the simplest: instead of reading or reviewing material, you try to remember it from memory. Close the book. Put away the notes and ask yourself: what do I actually know about this? What can I share or explain right now without looking?

The act of retrieval (pulling information out of memory rather than putting it back in) strengthens the memory trace in a way that re-exposure does not. A landmark study by Roediger and Karpicke, published in Psychological Science in 2006, found that students who studied a passage and then took a recall test retained 50 percent more of the material one week later than students who studied the passage twice. Think about that.

Fifty percent more.

The testing group had spent less time with the material and retained dramatically more of it.

This effect is called the testing effect or retrieval practice effect and has been replicated so many times across so many contexts that it is no longer seriously contested. A comprehensive review in Psychological Science in the Public Interest by Dunlosky and colleagues, covering ten of the most commonly used study strategies, rated retrieval practice as having the highest utility of any technique reviewed. Read that review for yourself. It’s worth it.

The strategy feels harder than re-reading because it is harder. You’ll most likely feel uncertain rather than certain. This discomfort is the mechanism through which it works because the struggle to retrieve is what builds the retrieval pathway.

#2: Spaced practice. Ebbinghaus's forgetting curve points toward a second strategy. Instead of massing your study into a single session before a test, spread it across multiple sessions with gaps in between. The gaps feel counterproductive because you forget things between sessions…which means each new session begins with some rebuilding. But, that rebuilding is the point.

Each time you re-learn something that has started to fade, the memory becomes more durable than if you had never let it fade at all.

A meta-analysis covering hundreds of studies found that spaced practice produces retention advantages over massed practice with a wild amount of consistency. The effect holds across subject matters, age groups, retention intervals, and learning contexts. In some conditions, spaced learners retain twice as much as massed learners when tested at a delay.

The problem is that spaced practice requires scheduling review sessions days or weeks after initial learning, returning to material that no longer feels urgent, studying for tomorrow's exam today. It requires exactly the kind of self-regulation and planning that students like myself are rarely taught and that schools rarely structurally support.

#3: Interleaving. Most study involves what researchers call blocking: completing all the problems of one type before moving to the next type. Finish all the algebra problems, then all the geometry problems, then all the statistics problems. This feels logical and efficient. It is also less effective than interleaving, which is mixing problem types together, so that you never know in advance which type of problem is coming next.

A study by Taylor and Rohrer found that students who practiced mathematics problems in an interleaved sequence outperformed students who practiced the same problems in blocked sequence by more than 40 percent on a delayed test, despite having spent identical amounts of time on the material. The advantage held even when students preferred blocked practice and found interleaving more difficult.

The reason interleaving works is precisely because you have to identify which approach each problem requires rather than just applying the approach you have been drilling. This identification process is the cognitive work that produces genuine understanding rather than the surface belief of fluency.

Why the feeling is wrong

So why do re-reading, massing, and blocking feel more productive than the strategies that actually work?

It comes down to current performance and learning.

When you re-read a passage, its surface features become familiar. The words, the structure, the general shape of the argument flows through without resistance, triggering the brain's familiarity detection systems. Familiarity feels like knowing.

It feels like the information is in there and accessible. What it actually reflects is that the infomration has been encountered before, not that the underlying understanding is durable or transferable or retrievable under new/novel conditions.

Robert Bjork calls this the current-moment learning trap. It’s the confusion between the conditions that produce the best current performance and the conditions that produce the best future retention. His research has shown repeatedly that the strategies that produce the most fluent in-session performance are frequently the ones that produce the most forgetting over time. And the strategies that feel most difficult and produce the most errors during practice are the ones that produce the most durable learning afterward.

That feels weird…

But, it is a feature of how the brain works. The familiarity signal is reliable in most contexts because things that feel familiar usually are things you know. The educational context is unusual precisely because it creates familiarity through exposure without guaranteeing that the understanding is genuine or durable. The signal that serves you well in everyday life misleads you often when you are learning.

The psychologist Asher Koriat spent decades studying what he called the feeling of knowing, which is exactly what we are talking about here. It is the “subjective” sense of knowing something or not knowing it, of understanding something or failing to understand it. His research showed that this feeling is heavily influenced by processing ease (by how fluently information comes to mind) rather than by its actual retrievability under demanding conditions. Students who have read a chapter three times have a feeling of knowing that substantially outstrips their actual ability to retrieve and apply the material when tested.

What this has to do with AI (and why it matters now)

You might be wondering what any of this has to do with Learning 3.0.

Remember, Learning 1.0 is oral tradition of teaching and learning passed down through generations of families and elders.

Learning 2.0 arrived with the written word, and is education as now see it in most places. Full of scale, but missing the individual connection and understanding present in Learning 1.0.

Learning 3.0 aims to bring individual understanding at scale. Something not previously available before AI.

Here is the direct connection.

The strategies that produce durable learning (retrieval practice, spaced repetition, interleaving) are precisely the strategies that AI learning systems are best positioned to deliver at scale. Spacing requires knowing how much time has passed since a concept was last practiced and serving it at the optimal interval. Retrieval practice requires generating questions rather than presenting answers. Interleaving requires sequencing problems in a way that cannot be predicted in advance. All of these require an ongoing, detailed model of what each individual learner knows, when they learned it, and how much they have likely forgotten since.

This is exactly what the learning fingerprint (described in my previous article) enables. A system that tracks each learner's performance history can calculate, for each piece of knowledge, the optimal moment to resurface it. Not too soon (before meaningful forgetting has occurred) and not too late (after so much has been forgotten that relearning becomes effortful).

This is what the most evidence-based AI learning systems are now doing, and the early evidence suggests it works. This 2025 study in Scientific Reports found that AI systems integrating spaced repetition and cognitive load estimation produced a 24.6 percent improvement in learning efficiency compared to control conditions.

But here is the real risk that the previous post introduced and that deserves direct mention here.

AI learning systems are often built by companies whose success metrics are primarily organized around engagement. Things like how long students use the platform, how often they return, whether they complete sessions. And the strategies that produce the best engagement metrics are frequently the opposite of the strategies that produce the best learning…

Re-reading is more pleasant than retrieval practice. It feels productive and produces positive feelings about the platform. Blocked practice is more comfortable than interleaved practice. Massed study produces a satisfying feeling of fluency that spaced study, with its built-in forgetting, does not. If a platform is optimized for how students feel about using it, it will systematically drift toward the strategies that produce fluency illusions, because those are the strategies that feel best in the moment.

This is Goodhart's Law applied to learning (something Ben Somers of Recess and I have discussed in detail. When engagement (which can be a very good thing in a classroom) becomes the target, it ceases to be a good measure of learning. And the digital platform that maximizes engagement while inadvertently minimizing retention is succeeding in the wrong way.

The question is whether the people building these systems understand the distinction deeply enough to build against their own incentives. Some do. Many do not. And the learners using the platforms in either case have almost no way to tell the difference, because the fluency illusion means that the platform producing worse learning will often feel like it is working better.

The knowledge you were never given

I want to step back from the technology for a moment and say something that is not about AI at all.

The research on how memory works is not new. Ebbinghaus published his forgetting curve in 1885. The testing effect was first described in the early twentieth century. The evidence base for retrieval practice, spaced repetition, and interleaving has been accumulating for decades, and the key findings were well-established by the 1990s. A 2013 review by Dunlosky and colleagues in Psychological Science in the Public Interest (which was one of the most comprehensive assessments of learning strategies ever conducted) called retrieval practice and spaced practice "the most effective and generalisable" strategies available, with evidence that "is extensive, based on well-controlled experiments, and the effects transfer across many different kinds of learning materials and conditions."

That paper was published twelve years ago. The findings it summarized were established decades before that. And yet most students (like myself) go through thirteen years of K-12 education without ever being taught that rereading is one of the least effective study strategies available. Without ever being shown how to use a retrieval schedule. Without ever being told that the feeling of fluency is an unreliable guide to actual learning.

This is a structural failure by all accounts. It is a system organized around delivering content and assessing performance that has never systematically taught students how learning itself works.

The gap between what cognitive science knows and what students know about their own learning is one of the clearest illustrations of why the conversation about Learning 3.0 matters beyond technology.

The technology can deliver spaced retrieval practice at scale. But it can also give students something more important than an optimized practice schedule…it can give them an accurate understanding of their own cognition. Why the discomfort of retrieval practice is a signal that it is working. Why the feeling of fluency after rereading is not a reliable indicator of genuine understanding. Why the struggle, even though annoying, is the point.

This is metacognition and the knowledge that makes all the other knowledge stick. And it is, as the previous post described, one of the strongest predictors of long-term academic success in the entire research literature.

This is the “knowledge” that the educational system has most consistently failed to give students. Because the system was never designed to give it.

What to do with this right now

Before we get to the broader argument about what Learning 3.0 needs to do differently, I want to give you something useful for today.

You are probably a learner like me (always learning!). Maybe your are a student, a professional in a new domain, or someone trying to master anything…here are the three things the research most clearly supports:

Stop rereading. Start recalling. After reading a section, close it. Write down or say aloud everything you can remember without looking. This is uncomfortable. You will produce errors. You will encounter the gaps directly. This is the process. Each retrieval attempt strengthens the memory in ways that rereading cannot. The evidence for this is as strong as anything in educational psychology.

Spread it out. Studying the same material across multiple sessions, with gaps in between, produces dramatically more durable retention than the same amount of study time concentrated in a single session. The gaps do not represent wasted time. They are where consolidation happens. The rebuilding that begins each new session is the mechanism that makes the knowledge last.

Mix it up. If you are practicing problems or skills, resist the urge to complete all examples of one type before moving to the next. Interleave different types. Ask yourself, before each problem, what kind of problem this is and what it requires. The extra cognitive work of identification is not inefficiency. It is the work that produces genuine understanding rather than surface fluency.

None of these is easy to implement without support. Spaced repetition requires scheduling across time. Retrieval practice requires the discipline to resist the comfort of rereading. Interleaving requires tolerating a lower immediate success rate in exchange for better long-term retention. These are the strategies that feel like they are not working, and that is precisely why they work.

The conversation about AI in education is about whether the technology is built around what the science of learning actually shows, or whether it is built around only attention/engagement metrics that drift systematically away from what the science shows.

The distinction matters. And recognizing it requires knowing what the science actually says, which is what this post has tried to give you.