Why mixing up your study feels wrong — and works better

There is a way most language learners would naturally organise a vocabulary study session if left to their own devices. First, all the food words. Then all the travel words. Then all the words for family, work, shopping — one category at a time, worked through to completion before moving on. It feels organised. Methodical. The cognitive equivalent of a tidy desk, every item grouped with its kind.

The research suggests this intuition is systematically wrong.

Blocked versus interleaved

Psychologists call this natural habit blocked practice — grouping similar material together and finishing each category before beginning the next. Its opposite is interleaved practice: shuffling the categories, so that a food word is followed by a travel word, then a family word, then back to food. The result feels incoherent, even a little unsettling. Something in the brain resists it. That resistance, it turns out, is the point.

The case against blocked practice does not come from theory alone. It comes from a finding that has replicated across decades, subject areas, and participant populations — one that continues to surprise people who encounter it for the first time.

The motor learning study that changed classrooms

The clearest early evidence came from an unlikely corner of psychology: motor learning research. In a 1979 experiment, psychologists John Shea and Robyn Morgan had participants learn three different arm-movement sequences. One group practised each sequence in a block — all the repetitions of sequence A, then all of B, then all of C. The other group practised the same sequences in a mixed, unpredictable order. During the practice phase, the blocked group performed better. Their error rates were lower. They looked, by every measure, more competent.

Then both groups were tested. The results reversed. The interleaved group remembered significantly more, and transferred their skills more reliably to new conditions. What had looked like slower, messier learning during the sessions turned out to be deeper learning all along.

This finding, and the many that followed it, gave researchers a name for the phenomenon: the contextual interference effect. The interference — the cognitive disruption that comes from switching between tasks before any one of them is mastered — is not noise in the system. It is signal. The difficulty it imposes is the mechanism by which the learning takes hold.

Why interference helps

The leading explanation has to do with what the brain is forced to do each time it encounters a familiar item in a varied sequence. With blocked practice, each repetition of the same type arrives so close to the last that the item can simply be refreshed in working memory — held briefly and re-activated. This is efficient in the short term, requires very little effort, and leaves relatively shallow traces.

Interleaved practice is different. By the time you return to a food word, you have encountered several other items since the last one. The earlier encounter has faded enough that retrieval cannot be merely refreshed — it must be reconstructed. And the act of reconstructing a memory, as the research on the testing effect has shown, is itself one of the most powerful ways to deepen it. What interleaving does, in part, is engineer retrieval practice into ordinary study, without requiring any deliberate effort to do so.

A second mechanism is what researchers sometimes call the discrimination hypothesis. When similar items are grouped together, the brain can treat them as a single category — one undifferentiated blob of vocabulary rather than a set of distinct, well-encoded words. When items from different categories are mixed, each encounter requires the brain to notice what makes this word different from the ones surrounding it. The work of distinguishing deepens the encoding of each item separately.

Robert Bjork, a psychologist at UCLA who has spent his career studying what he calls desirable difficulties — the class of conditions that make learning harder in the moment but better over time — describes interleaving as one of the clearest examples the field has produced. The discomfort the learner feels is a reliable indicator that something useful is being asked of them.

The gap between what learners think and what the data shows

What makes this finding especially striking is a second result that emerges consistently from this literature: people almost universally prefer blocked practice, predict they will learn better from it, and report higher confidence during it. The interleaved group tends to feel like they are struggling unnecessarily. The blocked group feels like things are going well.

In research by Nate Kornell and Robert Bjork at UCLA, participants studied painting examples under either blocked or interleaved conditions, learning to distinguish the styles of different artists. The interleaved group subsequently identified the artists more accurately when tested. But when asked which method they believed had helped them more, participants chose blocked — consistently, confidently, and incorrectly.

This is the metacognitive trap at the centre of blocked practice. The fluency you feel during study — the sense of answers arriving easily, of moving smoothly through material — is not a reliable signal that learning is happening. It is, in fact, partly a signal that retrieval is too easy to build anything durable. Blocked practice produces that feeling by design. Interleaved practice denies it — which is precisely what makes it uncomfortable, and precisely what makes it work.

The feeling of fluency during a study session is not evidence of learning. It is often evidence that the brain is not being asked to do enough.

What this means for vocabulary

Applied to vocabulary learning, the implication is direct. Studying ten words from the same semantic field — all fruits, all emotions, all office furniture — creates a priming effect: each word arrives in a context where similar words have just been active, making retrieval easier in the moment but less robustly encoded. The brain is not really distinguishing the words; it is riding the category.

Studying the same ten words interleaved with words from other semantic fields, or mixed with different exercise types, creates the conditions for genuine encoding. Each word must stand on its own — its meaning, its spelling, its sound — rather than being carried by the category surrounding it.

This does not mean semantic categories have no value. Knowing that a word belongs to a neighbourhood of related meanings is useful knowledge; that understanding should stay. What should be distributed is the retrieval practice itself — the moment of being asked to produce the word from memory, which is better done in a varied, unpredictable sequence than a grouped and predictable one.

The practical guidance that follows is simple, if slightly counterintuitive: if you have thirty words to review, resist the pull to group them by topic and work through each group. Let them arrive in mixed order. If you have several exercise types to complete, resist the urge to do all the multiple-choice first, then all the fill-in-blank. Alternate. The disorder is the mechanism, not a flaw in the design.

The harder sell

The difficulty here is that interleaving is, almost by definition, harder to enjoy. The sense of smooth progress that comes from drilling one category at a time is genuinely pleasant, and the pleasure of feeling competent is a real motivational asset. This is not a small consideration. An approach that works perfectly but that you abandon after three days is worse than an imperfect approach you sustain.

But there is a distinction worth keeping in mind. The fluency you feel during blocked practice is performance — it reflects how well you can retrieve something in the moment, in a context engineered for retrieval. What you want from vocabulary study is learning: the kind that survives sleep, survives several days without review, survives the moment when you need a word in a real conversation and cannot prime yourself first. The research consistently shows that these two things point in different directions. Feeling like you are learning well is not the same as learning well.

The only real fix is to know this going in, and to trust the discomfort. A session that felt somewhat hard, somewhat jumbled, where the answer did not always arrive immediately — that is probably the session that left the deepest marks.

How this fits with spaced repetition

There is a quiet alignment between interleaving and the other major principle of science-backed vocabulary practice. Spaced repetition — returning to a word just before you would have forgotten it — already produces a naturally interleaved queue. As words age at different rates and rise and fall in the review schedule at different times, a well-run session mixes beginners alongside familiar ones, easy words alongside difficult ones, food words alongside travel words, because each has its own due date and none of them align neatly by topic.

This is not accidental. The designers of spaced repetition systems have been reading the same cognitive literature for decades. The shuffled queue is not a quirk of implementation. It is, in a quiet way, the whole point — a design that makes interleaving automatic, so you do not have to trust the discomfort consciously. The algorithm handles the mess. You just have to show up for it.

So the next time a session feels harder than it should — the answers slower, the transitions more abrupt, the sense of smooth progress missing — it may be worth resisting the instinct to reorganise. That friction is the sign that your brain is being asked to do something harder than recall in the moment. It is being asked to remember for real.