IX Coach

Levels of Processing: Why Depth of Encoding Determines What You Remember

Shallow vs. deep encoding, elaboration, and the practical techniques that drive durable memory

What is levels of processing and how does depth aid memory?

Craik and Lockhart’s (1972) levels of processing framework proposes that memory is a by-product of the type of processing performed on incoming information: shallow processing (appearance, sound) produces weak, short-lived traces, while deep processing (meaning, elaboration, connection to prior knowledge) produces strong, durable ones. The implication for learning is that how you engage with material matters more than how many times you see it.

Most forgetting is not about memory capacity — it is about encoding depth. Rereading notes creates shallow traces through perceptual familiarity; explaining a concept to someone else creates deep traces through semantic elaboration and self-generation. Craik and Lockhart gave this distinction a theoretical framework in 1972, and the implication has been replicated across decades of research: it is not how often you see something but what you do with it that determines whether it stays. The practices below operationalize depth-of-processing for everyday learning.

Practices

How these practices connect

Retention depends on the number and richness of retrieval pathways laid down at encoding, not on repetition or exposure time. Structural or phonological processing (how something looks or sounds) creates a thin, isolated trace, while semantic processing recruits an existing network of associations and lays down multiple parallel routes back to the same memory. Every practice below is a different lever for forcing that deeper, more connected form of processing, whether the connection is to meaning, to a vivid image, to another person's understanding, to your own identity, or to a concrete instance.

What is deep processing and how is it different from just re-reading something?

Deep processing means analyzing information for what it means and how it connects to what you already know, rather than for how it looks or sounds. Craik and Lockhart's framework describes this as semantic-level analysis, as opposed to shallow structural (visual) or phonological (sound) analysis. Re-reading mostly re-exposes you to the surface form of the text and doesn't force meaning-based analysis, so it produces a shallow trace with few retrieval routes. Semantic encoding activates a much wider network of existing associations, which is why it produces a more durable and flexible memory.

Is making up a vivid mental picture actually better than just understanding the meaning of something?

They work through related but distinct mechanisms, and combining them is stronger than either alone. Semantic encoding widens the network of meaning-based associations a new fact connects to, while elaborative encoding (like picturing a hippo enrolling in school to remember the hippocampus forms new memories) adds a second, visual-spatial encoding pathway on top of that semantic one. The elaboration only works if the image itself carries the meaning, not just a surface pun: a hippo alone encodes nothing about memory function, but a hippo on its first day of school encodes the idea of new-memory-formation inside the picture. So a vivid image is not a substitute for meaning-based processing; it is a second parallel route to the same meaning, which is why elaborations built only on sound or wordplay tend to fail under exam pressure.

Why does explaining something to someone else help you remember it better than studying it again?

Teaching forces several deep-processing demands to happen at once: you have to retrieve the material, organize it into a coherent structure, generate examples a non-expert would understand, and monitor in real time whether your explanation makes sense. Re-reading or re-studying does not require any of that active reorganization. The Protégé Effect research found that even preparing to teach someone improves learning before any teaching actually happens, suggesting the benefit comes from anticipating the need to explain, not from the act of speaking itself. The evidence here is observational rather than a controlled trial, and the effect appears to depend on genuinely believing you will teach and actually preparing to, not just announcing an intention to explain.

Is it better to relate new information to yourself or to just understand the concept generally?

Self-referential encoding tends to produce stronger memory than general semantic understanding alone, because it uses a uniquely rich retrieval structure. Rogers, Kuiper and Kirker's research found that words judged for personal relevance ("does this describe you?") were remembered substantially better than the same words judged only for their meaning or structure. The mechanism is that self-related information gets folded into your self-concept, which is an unusually extensive, well-organized, and frequently activated memory structure, giving it more routes back to conscious recall than a generic semantic association would. That advantage only holds when the self-connection is genuine, though; a forced or unconvincing link to yourself does not produce the same benefit, so this is not simply a stronger version of semantic encoding you can apply mechanically.

Should you write down questions before you start reading, or just read the material and see what sticks?

Setting specific questions before reading converts passive absorption into active hypothesis-testing, which activates more semantic processing than reading to simply get through the material. Instead of processing each sentence in isolation, you're comparing it against a specific goal, which forces evaluation of relevance and meaning rather than surface content. This only works if the questions are specific and conceptual; a broad question like "what is this chapter about" doesn't direct attention toward any particular meaning and produces the same shallow processing as reading without a question at all. The research behind this is observational rather than a randomized trial, so the size of the effect is less firmly established than for some of the other practices here.

Why do concrete examples stick in memory longer than abstract principles?

Concrete information activates both a verbal and an imagistic (picture-based) memory system at the same time, while purely abstract information mostly activates the verbal channel alone. Dual coding theory and levels-of-processing research converge on this point: the extra imagistic encoding pathway gives concrete instances more routes back to retrieval than abstract statements have. This is the same reason concrete nouns are recalled more reliably than abstract ones in free-recall experiments, and it's why an abstract principle taught before any concrete example tends to be forgotten while the example that follows is not. The imagistic pathway only engages if you actually visualize the concrete instance; merely being told something is "like" a concrete example without picturing it does not reliably produce the same benefit.

How do you choose between these?

Pick based on the material and moment in front of you, not a fixed personal style. If you're facing an abstract or technical definition with no example yet, start by grounding it in a concrete, visualized instance before trying to explain or connect it further. If you already understand the material at a basic level but need it to stick past an exam, add a distinctive elaboration or connect it to a real situation from your own life, since both create extra retrieval routes on top of understanding you already have. If you're about to sit down and read new material cold, generate specific questions first so you're processing it against a goal rather than passively. If you have time before a high-stakes test or need to know whether you actually understand something (versus just recognize it), prepare to teach it, since that's the condition that forces you to find your own gaps. These are not mutually exclusive: a concrete example can also be the thing you elaborate on, and a self-referential connection can be the answer to a question you set before reading.

What these approaches do not settle

The original strict-hierarchy version of levels of processing (structural < phonological < semantic, in a fixed order) has been revised by researchers toward a looser "depth of elaboration" model, because the evidence didn't support a strict linear ranking, even though the core finding that semantic processing beats surface processing remains robust. Two of these six practices (teaching and question-driven reading) rest on observational evidence rather than randomized trials, so their effect sizes are less firmly pinned down than the self-reference and dual-coding findings, which do come from controlled research. Several of these practices also carry the same caveat in different words: elaboration, self-reference, and concrete visualization only produce the claimed benefit when they are genuine, specific, and actually performed (a real image visualized, a real personal connection felt, real teaching prepared) rather than a generic or ceremonial version of the same instruction.

Practice this with IX Coach

Practice this with IX Coach

Frequently asked questions

What is levels of processing and how does depth aid memory?

Craik and Lockhart’s (1972) levels of processing framework proposes that memory is a by-product of the type of processing performed on incoming information: shallow processing (appearance, sound) produces weak, short-lived traces, while deep processing (meaning, elaboration, connection to prior knowledge) produces strong, durable ones. The implication for learning is that how you engage with material matters more than how many times you see it.

How do you process new information by its meaning, not its form?

Craik and Lockhart proposed that encoding depth is determined by the level of analysis: structural (visual/orthographic), phonological (sound), or semantic (meaning). Semantic processing activates a far wider network of associations in long-term memory, creating more retrieval pathways for the new trace. The more existing knowledge a new item connects to, the more routes exist to reach it during retrieval — making it both more durable and more flexibly applicable.

What is An example of elaborative encoding to improve memory?

Elaboration extends the depth of processing by weaving new information into an existing network of memory traces. A vivid, personally relevant example creates a unique encoding context that serves as a retrieval cue later, because no two people elaborate in exactly the same way — your elaboration is distinctive to you and to that moment. The image or story also provides a parallel visual-spatial encoding pathway alongside the semantic one, creating redundant retrieval routes. Working through the hippocampus example shows what the mechanism requires: the hippo alone is a pun on the name and encodes nothing about what the structure does, while the hippo being enrolled on its first day carries the function — new memories being formed — inside the image itself. That is the difference between a cue that returns a label and a cue that returns the meaning, and it is why elaborations built on the sound of a word tend to collapse under exam conditions while elaborations built on the sense of it hold.

Explain the material to someone else: how do you real or imagined?

Teaching activates multiple deep-processing demands simultaneously: you must retrieve the information, organize it coherently (which reveals structural understanding), generate examples accessible to a non-expert, and monitor whether the explanation is making sense — all of which require semantic processing far deeper than reading or re-reading. The Protégé Effect (documented by Chase et al.) shows that even preparing to teach someone else improves learning before any actual teaching occurs, suggesting the encoding benefit is in the anticipation of explanation.

How do you link new information to your own life and identity?

The self-reference effect (Rogers, Kuiper & Kirker, 1977) showed that words and concepts judged for self-relevance ("Does this describe you?") were remembered substantially better than those judged for structural or semantic properties alone. The mechanism is that self-related information is uniquely integrated into an extensive, well-organized, and frequently activated memory structure — the self-concept — providing the richest possible web of retrieval paths.

How do you generate questions before reading to drive deep processing?

Reading with a question in mind converts passive text processing into active hypothesis-testing: the reader is searching for specific information, comparing each passage to the question, and evaluating relevance. This goal-directed processing activates more semantic analysis than reading to "absorb" or "get through" the material, because each sentence is evaluated for its relationship to a meaningful goal rather than processed in isolation.

How do you ground abstract concepts in concrete, physical instances?

Dual coding theory (Paivio, 1971) and levels-of-processing research converge on this: concrete information activates both verbal and imagistic memory systems simultaneously, while abstract information activates primarily the verbal channel. The additional imagistic encoding provides a parallel retrieval route, making concrete memories more retrievable. This is why concrete nouns are remembered better than abstract ones in free recall, and why specific examples are remembered far longer than general principles stated without them.

What is deep processing and how is it different from just re-reading something?

Deep processing means analyzing information for what it means and how it connects to what you already know, rather than for how it looks or sounds. Craik and Lockhart's framework describes this as semantic-level analysis, as opposed to shallow structural (visual) or phonological (sound) analysis. Re-reading mostly re-exposes you to the surface form of the text and doesn't force meaning-based analysis, so it produces a shallow trace with few retrieval routes. Semantic encoding activates a much wider network of existing associations, which is why it produces a more durable and flexible memory.

Is making up a vivid mental picture actually better than just understanding the meaning of something?

They work through related but distinct mechanisms, and combining them is stronger than either alone. Semantic encoding widens the network of meaning-based associations a new fact connects to, while elaborative encoding (like picturing a hippo enrolling in school to remember the hippocampus forms new memories) adds a second, visual-spatial encoding pathway on top of that semantic one. The elaboration only works if the image itself carries the meaning, not just a surface pun: a hippo alone encodes nothing about memory function, but a hippo on its first day of school encodes the idea of new-memory-formation inside the picture. So a vivid image is not a substitute for meaning-based processing; it is a second parallel route to the same meaning, which is why elaborations built only on sound or wordplay tend to fail under exam pressure.

Why does explaining something to someone else help you remember it better than studying it again?

Teaching forces several deep-processing demands to happen at once: you have to retrieve the material, organize it into a coherent structure, generate examples a non-expert would understand, and monitor in real time whether your explanation makes sense. Re-reading or re-studying does not require any of that active reorganization. The Protégé Effect research found that even preparing to teach someone improves learning before any teaching actually happens, suggesting the benefit comes from anticipating the need to explain, not from the act of speaking itself. The evidence here is observational rather than a controlled trial, and the effect appears to depend on genuinely believing you will teach and actually preparing to, not just announcing an intention to explain.

Is it better to relate new information to yourself or to just understand the concept generally?

Self-referential encoding tends to produce stronger memory than general semantic understanding alone, because it uses a uniquely rich retrieval structure. Rogers, Kuiper and Kirker's research found that words judged for personal relevance ("does this describe you?") were remembered substantially better than the same words judged only for their meaning or structure. The mechanism is that self-related information gets folded into your self-concept, which is an unusually extensive, well-organized, and frequently activated memory structure, giving it more routes back to conscious recall than a generic semantic association would. That advantage only holds when the self-connection is genuine, though; a forced or unconvincing link to yourself does not produce the same benefit, so this is not simply a stronger version of semantic encoding you can apply mechanically.

Should you write down questions before you start reading, or just read the material and see what sticks?

Setting specific questions before reading converts passive absorption into active hypothesis-testing, which activates more semantic processing than reading to simply get through the material. Instead of processing each sentence in isolation, you're comparing it against a specific goal, which forces evaluation of relevance and meaning rather than surface content. This only works if the questions are specific and conceptual; a broad question like "what is this chapter about" doesn't direct attention toward any particular meaning and produces the same shallow processing as reading without a question at all. The research behind this is observational rather than a randomized trial, so the size of the effect is less firmly established than for some of the other practices here.

Why do concrete examples stick in memory longer than abstract principles?

Concrete information activates both a verbal and an imagistic (picture-based) memory system at the same time, while purely abstract information mostly activates the verbal channel alone. Dual coding theory and levels-of-processing research converge on this point: the extra imagistic encoding pathway gives concrete instances more routes back to retrieval than abstract statements have. This is the same reason concrete nouns are recalled more reliably than abstract ones in free-recall experiments, and it's why an abstract principle taught before any concrete example tends to be forgotten while the example that follows is not. The imagistic pathway only engages if you actually visualize the concrete instance; merely being told something is "like" a concrete example without picturing it does not reliably produce the same benefit.

Research

Related concepts