Two students sit down with the same vocabulary list for the same thirty minutes. The first reads each word carefully, notes its spelling, says it under her breath, and moves on. The second pauses at each word and asks a small, almost idle question — what does this remind me of? When would I actually say this? How is it different from the word above it? A week later, tested without warning, the second student remembers far more. Not because she spent more time, and not because she has a better memory. Because of what her mind was doing while the words passed through it.

That difference — what the mind does during exposure, rather than how long the exposure lasts — is the subject of one of the most influential ideas in memory research: levels of processing.

The paper that reframed what memory is

In 1972, the psychologists Fergus Craik and Robert Lockhart published a framework that quietly upended how researchers thought about remembering. The dominant model of the time treated memory as a set of storage boxes: information entered short-term memory, and if you rehearsed it enough, it got transferred into long-term memory. Repetition was the currency. Hold a fact in mind long enough, the thinking went, and it would eventually stick.

Craik and Lockhart proposed something different: a memory is not a thing you file away, but a byproduct of the analysis you perform. When you encounter a word, you can process it shallowly — what it looks like, what it sounds like — or deeply, engaging its meaning, its associations, its relationship to things you already know. The deeper the processing, the more durable the trace it leaves behind. Rehearsal that merely recycles the surface — repeating a phone number to yourself — keeps information available for the moment but does surprisingly little for long-term retention.

Three years later, Craik and Endel Tulving put the idea through a now-classic series of experiments. Participants saw words and answered one of three kinds of questions about them. Some questions were structural: Is the word in capital letters? Some were phonemic: Does it rhyme with "weight"? Some were semantic: Would it fit in the sentence "He met a ___ in the street"? Crucially, nobody was told to memorize anything — they were just answering questions.

Then came a surprise memory test. Words that had been processed semantically were remembered dramatically better than words judged by their sound, which in turn beat words judged by their appearance. Same words, same exposure, wildly different retention — the only variable was the depth of the question. And when the researchers stretched out the shallow tasks to take just as long as the deep ones, the advantage didn't disappear. Time on task wasn't the ingredient. Meaning was.

What "deep" actually means

Depth isn't effort in the gritted-teeth sense, and it isn't concentration. A student can stare at a page with total focus and still process it shallowly. Depth means semantic involvement: engaging what the material means, what it connects to, what it implies.

Researchers generally point to two mechanisms for why this works. The first is elaboration. When you process meaning, you link the new information to things you already know — and every link is a potential retrieval route. A fact connected to one thing has one road back; a fact connected to ten things has ten. The second is distinctiveness. A richly encoded memory has texture. It's harder to confuse with its neighbors, so when you go looking for it later, it stands out from the crowd of similar traces instead of blending in.

There's a memorable extension of this called the self-reference effect. In 1977, Rogers, Kuiper, and Kirker found that words judged by the question does this describe you? were remembered even better than words given ordinary semantic judgments. Information processed in relation to yourself — your experiences, your life, your opinions — gets some of the deepest encoding of all, presumably because the self is the most elaborate knowledge structure any of us owns.

The trap: studying that feels productive but stays shallow

Once you see the levels-of-processing lens, a lot of familiar study behavior looks different. Rereading a chapter is mostly perceptual processing — your eyes recognize words they've seen before, which produces a warm glow of fluency that feels like knowing. Highlighting is a judgment about which sentences look important, made at the surface of the text. Copying notes verbatim transcribes words without ever touching what they mean. All of these can occupy an entire evening while the mind performs the cognitive equivalent of Craik and Tulving's is it in capital letters? task.

This is why the experience of "I studied for hours and remember nothing" is so common, and so demoralizing. The hours were real. The processing was shallow. Memory, indifferent to your intentions, recorded only the analysis you actually performed.

How to force depth on purpose

The encouraging flip side: depth is a choice you can make, and it doesn't require more time — it requires different questions. A few that reliably push processing down into meaning:

Ask why it's true. Not just what the fact is, but what makes it so. Why does this drug cause that side effect? Why did this treaty follow that war? Explaining a cause forces you to connect the fact to the machinery around it.

Generate your own example. A definition processed shallowly is a string of words. A definition you've applied to a situation from your own life has been run through the semantic system — and picks up the self-reference bonus along the way.

Compare and contrast. Asking how mitosis differs from meiosis, or how this verb tense differs from the one you learned last week, forces discriminative, meaningful analysis — and builds the distinctiveness that keeps similar memories from interfering with each other.

Translate it into your own words. If you can't restate an idea without the textbook's phrasing, you processed the phrasing, not the idea.

One honest caveat: later work by Morris, Bransford, and Franks showed that "deep" is not absolutely superior in every situation — memory is best when the processing at study matches the processing the test will demand (they showed that rhyme-based encoding actually beat semantic encoding on a rhyming test). But since almost everything we study for real life — exams, languages, professional knowledge — is ultimately used through meaning, deep semantic processing is the right default nearly all the time.

Where depth meets repetition

Levels of processing explains what makes a single encounter with an idea stick. But even deeply processed memories fade — more slowly, but they fade. That's where deep encoding and retrieval practice become partners rather than rivals. Every time you successfully recall something from memory, you're not passively refreshing it; you're performing another act of semantic processing, reconstructing the meaning from partial cues. Retrieval is deep processing. A flashcard answered from genuine recall engages the memory at exactly the level Craik and Tulving found most durable — which is a large part of why testing yourself so reliably outperforms rereading.

And the depth principle applies before the first review ever happens: the act of writing a card in your own words, choosing an example, deciding how to phrase the question — that's elaborative, semantic work. The card is half-learned by the time you finish making it.

Depth once, repetition forever

This is the division of labor behind Recall. You bring the depth: cards phrased in your own words, questions that ask why, examples pulled from your own life. Recall handles the part no amount of insight can do for you — resurfacing each card at the moment it's about to fade, using FSRS, a modern spaced repetition algorithm that fits the schedule to how you actually remember. Because every review is an act of retrieval rather than recognition, each one deepens the trace instead of just refreshing your familiarity with it. If you already have decks in Anki or Quizlet, you can import them and keep going; everything works fully offline, so the depth you've built is always with you. If you'd like your studying to leave marks that last, you can try it at recall.lumenlabs.works.