Learning advice today often sounds like productivity advice. Study faster, optimize your routine, squeeze more into less time. We track minutes, streaks and sessions as if learning were a logistics problem. But the science of learning points somewhere else. What determines whether knowledge sticks is not speed, effort saved or even time spent. It’s what your mind is actually doing while you learn.

The concept of deep processing shifts the question from how hard or how long you try to what you’re doing in your brain. You can be fully focused and still learn very little. You can feel mentally exhausted and yet build fragile memories. What matters isn’t effort alone but the kind of mental work producing that effort and whether it’s deep enough.

What is deep processing

Deep processing refers to the idea that our ability to remember information depends on how deeply we processed it when we first learned it. When you actively engage with the meaning, implications, underlying principles and connections of the information you’re learning, you’ll remember it better than if you engage in shallow analysis.

Deep processing stands in stark contrast to shallow processing, or superficial engagement with the learning material. Shallow processing involves focusing on attributes that remain on the surface or the look-alike of what you want to learn (typically at the visual or sound levels), such as the appearance or pronunciation of words. It also refers to engaging in rote activities that reinforce those superficial aspects like repetition, rereading and highlighting.

These ideas are hardly new: they come from a landmark moment in cognitive psychology. In a 1972 study by Fergus Craik and Robert Lockhart, researchers introduced the “Levels of processing framework” arguing that memory strength depends less on where information is stored (short-term memory? long-term memory?) and more on how it is processed when first encountered. Research has backed it with plenty of evidence since then. Recent studies from 2022 and 2025 confirm the theory: across many experiments, depth of processing strongly influences performance at both immediate and delayed tests.

Let’s see deep processing in action

To see the difference, imagine four ways of encountering a word, for example the word “Procrastination”:

1. You notice that the “r”s in the syllables “pro-” and “-cras-” sound funny when read together.

2. You think about its meaning: “unnecessarily postponing tasks” and how it describes certain unhealthy habits in your daily life such as spending too long on social media.

3. You notice that it rhymes with “formation” and “station“.

4. You notice that the first letter is capitalized.

All four require attention. Only the second one reliably creates durable memory. All the other ones stay in levels of processing that aren’t deep enough (structure and phonemic levels). These also create memory traces, but they aren’t very strong.

Carl Hendrick provides another great example in his article, “Correct answers but no learning.” He recounts an experience in which his daughter quickly completed a phonics puzzle matching images with words because the correct matches were color-coded. Although the task was designed to develop reading skills, the color coding undermined the learning outcome by causing the learner to avoid cognitive engagement and resort to shallow processing (matching colors) to complete the task. This is a great example of a poorly designed task that missed its goal.

Is it the same as being concentrated or “in the flow”?

As you probably suspect, the answer is no. That’s one of the most common sources of confusion regarding the idea of deep processing: that it simply means concentrating harder or being “in the flow”.

Concentration is about attention. It's the ability to sustain focus and resist distraction. Deep processing is about what you do with that focus. You can concentrate intensely on a shallow task. Anyone who has ever memorized a list word for word knows this feeling. It takes effort. It feels demanding. And yet the information often disappears quickly. This distinction also helps explain why productivity culture often clashes with good learning.

What does deep processing look like?

Deep processing can happen in many different forms, but the most common one is semantic processing (that is: handling information by working on what it means). But there are other kinds of deep processing on top of semantic:

• Elaborative processing: A recent study showed that simply asking students “why?” after introducing new information made them remember that information incredibly better. The reason is elaborative interrogation: studies show that students who generate explanations remember more and transfer knowledge better than those who simply read explanations. This is also explained by the generation effect.

• Organizational processing: Organization matters just as much as connection. Studies show that information linked to a highly structured framework is remembered especially well. It forces you to question the role of the new information within the structure you know and how it relates to other parts.

• Self-referential processing: Interestingly enough, one such organizational framework is the self. Research shows that relating information to yourself provides a powerful scaffold to organize information, not just emotional relevance. In practice, this means that tying information to personal experiences, goals or beliefs helps encoding by making it feel relevant to one’s own life. Does this mean that you should make learning about you? Well … yes!

• Other kinds: There are other kinds such as relational/associative processing (linking concepts to each other) or critical/analytical processing (assessing, evaluating, interpreting and so on) but all of them have one thing in common: they happen when your brain is actively engaged in managing the information cognitively.

Where deep processing can fail

Deep processing is powerful, but it is not magic. Here are some limitations:

• Working memory: If material is too complex and poorly structured, learners become overloaded. In those cases, simplifying content and sequencing ideas is a prerequisite for depth.

• Poor quality elaboration: Not all connections are useful. Random associations or idiosyncratic stories can actually weaken understanding if they do not reflect underlying structure.

• Motivational barriers: Deep processing feels harder and less efficient in the short term. Without guidance, learners often revert to habits that feel productive but produce fragile knowledge.

• Not always necessary: For simple facts or procedural refreshers, brief exposure or spaced review may be sufficient. The goal is not depth everywhere, but depth where understanding matters.

Deep processing and microlearning

Microlearning (short, flexible lessons delivered in quick bursts) has grown popular in a world of fragmented attention and busy schedules. At first glance, it seems at odds with deep processing, which requires sustained engagement to connect ideas, build structure and integrate new knowledge. Short bursts can limit room for elaboration and sense-making, though this tension is often overstated.

The issue is not that microlearning cannot support deep processing. It can: well-designed micro-modules can push learners to wrestle with meaning and make connections. Yet this approach doesn’t suit every topic. Some material resists easy breakdown and requires longer periods of focused engagement. Complex domains such as medical diagnostics, legal reasoning, systems thinking or advanced mathematics involve multiple interacting concepts that must be understood in relation to one another, not as isolated facts. In these cases, depth comes from sustained interaction until structure emerges.

Conversely, simpler facts, procedural refreshers or easily separable content often do not need intense depth. Deep processing here can be overkill, adding effort without improving retention.

The efficiency paradox

This leads to an efficiency paradox. Short learning bursts feel faster, but they can slow down the path to durable understanding when they prevent deep engagement. Ironically, learners often reach long-term retention sooner by spending time processing meaning deeply than by spreading attention thin across many brief sessions. We encounter a similar paradox regarding the use of AI in learning.

Practical tips to learn deep, not hard

For individual learners

1. Get used to going deep. Great learners make it a habit to wrestle with the meaning and connections of anything they encounter. Cultivate a mindset that’s allergic to shallowness.

2. Make learning yours. Treat explanations as something you generate, not something you receive. When you do read or watch an explanation, rephrase it to make it yours. Make connections by asking “why?” relentlessly or finishing the sentence “This is like…”.

3. Learn to sit with discomfort. Confusion is often the entry point to durable learning. Don’t give up when it feels hard, channel your efforts to engage with the meaning and organization of information.

For teachers and learning designers

1. Design tasks that require transformation. Summaries, comparisons, predictions and explanations force semantic work.

2. Align assessments with meaning. If tests reward surface recall, students will study shallowly.

3. Scaffold deep strategies. Model how to elaborate, organize and self-explain instead of assuming students know how. Teach them to go deep.

4. Use microlearning strategically. Short modules work well for topics that can be easily broken down, but developing expertise often requires deep processing.

You might end by reframing learning itself

Instead of treating learning as the management of information, techniques or time, it can be understood as refining the quality of your thinking. The methods that work according to science (taking guesses, testing yourself, applying knowledge, struggling with difficult problems, etc.) work not because they are clever hacks but because they force the mind to engage in specific kinds of processing. They require organization, interpretation and integration with existing knowledge.

Seen this way, deep processing is not one strategy among many. It is the underlying mechanism shared by strategies that reliably produce durable understanding, transfer and insight. Whenever a technique works, it does so because it pushes processing beyond surface features.

The real question is not which tool to use, but whether the tool creates the conditions for the mind to go deep enough.

Share

Keep learning

Prompt suggestions. Always ask follow-up questions

• I want to improve long-term retention of a topic. Can you help me turn a lesson into deep-processing activities like elaboration, organization and self-referential processing? Ask about the topic and audience first.

• Explain shallow vs deep processing with examples from common study habits. Then guide me step by step to transform a shallow review activity (like rereading or highlighting) into a deep-processing task.

• I’m designing microlearning modules. Can you help me decide which content needs short bursts versus sustained engagement for deep processing, and give examples of activities for each case?

Links

• ▶️ Levels of processing theory (explained in 3 minutes): This short video summarizes the core ideas I tried to convey in this article in a very succinct and approachable way. It only hurts that, by the end, it mentions “learning styles” as a valid criticism of the levels of processing theory when it’s not.

• 📄 Levels of Processing: A Framework for Memory Research (1972): If you found the topic interesting, I’d recommend reading the foundational article of this framework, it goes into a lot of detail and is still today a relevant read.

Related articles

Videos were once the future of learning. They’re now its present and, although they have lost their novelty, there is no sign of them going away.

According to data from 2017, learning-related videos alone get 500 million views per day on YouTube. That’s half a billion. Per day. The popularity of this format is beyond question, and the reasons seem pretty evident: a video can do what text and images can’t. It moves, it talks, it shows.

That richness still tricks people into thinking any video lesson will naturally improve learning outcomes. Anyone who has sat through a weak explainer knows the opposite. The effectiveness of a learning video has little to do with the fact that it’s a video and has everything to do with how its design fits the limits of human cognition and how we learn.

Subscribe now

The principles of multimedia learning

Richard Mayer’s cognitive theory of multimedia learning explains why some videos teach and others merely look modern. His 12 evidence-based principles apply beyond video, shaping all forms of multimedia.

Mayer first presented these principles in the early 2000s before learning videos became mainstream. YouTube didn’t even exist yet. He drew on decades of studies and led his own experiments to identify which choices in audiovisual design support learning. He presented them in his now-classic work Multimedia Learning, and later in the Cambridge Handbook of Multimedia Learning.

What’s fascinating about them is how specific they are: they give practical guidelines, such as placing words next to images, and avoidances, such as reading out loud text that is already on the screen. This clarity explains their popularity: Mayer’s principles are now an indispensable reference for every instructional designer.

The 12 principles rest on the concept of dual coding: learning improves when verbal information (processed through the ears) and visual information (processed through the eyes) work together. Each channel has limited capacity, so how we split information between them matters.

These ideas connect directly to cognitive load theory, which splits mental effort into three types. Mayer mapped his principles onto these types: five principles for reducing unnecessary load, three for managing complexity and four for strengthening productive effort. Let’s look at them!

Reduce extraneous load: remove distraction

Extraneous load is the bad kind of mental effort: it comes from mental work that doesn't support the learning goal. Poor multimedia design creates a lot of it. Five principles target this problem:

• 1) Coherence principle: remove extraneous details, visuals or sounds.

• 2) Signaling principle: add cues that highlight essential structure.

• 3) Redundancy principle: avoid combining narration, identical on-screen text and graphics.

• 4) Spatial contiguity principle: place text and visuals close together.

• 5) Temporal contiguity principle: present words and pictures at the same time.

Coherence principle

We often think of decorative visuals or background music as a nice-to-have. They aren’t! Studies show that irrelevant sounds, images that are mere decoration or mismatched text and visuals, cause learners to lose the capacity needed for comprehension. Ironically, videos with more material result in less understanding!

Coherence goes beyond sounds or visuals. Adding seductive details like fun facts to make a video more engaging can backfire because learners remember the trivia instead of the lesson. Mayer showed this talking about a study on teaching how viruses cause colds: including a note that "people who make love once or twice a week have better immunity” reduced learning rather than improved it.

Unsurprisingly, Mayer explains, people find topics such as sex or death to be the most interesting, but adding fun facts about these juicy topics doesn’t carry over to the learning of other topics, even if they’re somewhat related. Too bad!

Signaling principle

Think of the signaling principle as the instructional designer’s equivalent of putting a neon arrow over “The Important Thing™” so learners don’t wander off. Good cues act like breadcrumbs for the brain, guiding attention to the structure that actually matters. Don’t just describe a part of a visual on screen and expect the learner to find their way to that part, use an arrow or a circle to highlight what you’re talking about, use section headings, overviews, summaries, etc. Without cues, your screen is a visual maze for non-experts!

Redundancy principle

The redundancy principle is another counterintuitive case. It seems logical to think that narrating and showing the same text would reinforce learning. After all, aren’t we all used to reading captions on Instagram reels and subtitles on Netflix series nowadays? Yet when a lesson includes graphics, research consistently shows the opposite.

This isn't a contradiction of dual coding. Written text and images share the visual channel, while speech uses the auditory one. Learning improves when channels provide complementary information and declines when both present the same verbal content.

There are cases where redundancy can support learning: if the text appears next to the corresponding visual and if it highlights only key elements (rather than duplicating the narration). That means: this principle doesn’t exclude the use of key terms, cues, headings or summaries that support understanding.

Subtitles on video platforms such as TikTok, Instagram or Netflix aren't counterexamples. These services aim for attention or accessibility (in noisy or multilingual settings), not for learning.

Split-attention principles

The two principles on spatial and temporal “contiguity” address split attention (they’re often treated as one principle): when visuals and their explanations appear far apart or out of sync. The effort learners have to put in trying to look for the explanation that matches a diagram or graph is simply wasted and takes a toll on understanding. Use the audio channel (voiceover) to explain the visual as it’s directly on screen. In other words, different sources of information should be physically and temporally integrated.

Manage intrinsic load: support complexity

Some material is simply complex in itself. Intrinsic or essential load depends on the topic and how much the learner already knows about it (prior knowledge). Three principles help manage this complexity:

• 6) Segmenting principle: break content into learner-controlled units.

• 7) Pre-training principle: introduce key elements before the main lesson.

• 8) Modality principle: use narration rather than on-screen text with visuals.

Segmenting lets learners control the pacing. Segmented videos following the principles of microlearning let learners pause, rewind and consolidate, without a forced processing speed. Nowadays, most video platforms offer their users video controls, so the key lies in the length and the alignment with learning outcomes (proper chunking).

Pre-training prepares learners by introducing the key elements before the main lesson. When learners know the names and functions of parts in a system, they can devote more capacity to understanding how the system works.

The modality principle (what I call the anti-PowerPoint principle) sounds similar to the redundancy principle but isn’t the same. It says that when explaining a visual, audio narration is better than on-screen text. Say goodbye to those long blocks of text explaining a visual! Your audience will thank you.

Foster germane load: deepen understanding

Once extraneous load is minimized and essential load is manageable, designers can encourage deeper cognitive processing. Another four principles support this stage:

• 9) Multimedia principle: combine words and relevant pictures.

• 10) Personalization principle: use conversational style.

• 11) Voice principle: use a human voice instead of a synthetic one.

• 12) Image principle: adding the speaker’s image rarely helps.

The multimedia principle states that people learn better from words and relevant pictures than from words alone. This principle sums up the idea behind dual-coding in a nutshell. If you combine words and visuals, learning increases.

The personalization and voice principles reflect social agency. Learners engage more deeply when instruction feels like a human speaking directly to them rather than a formal script or artificial voice. The language used should be conversational and the voice shouldn’t sound robotic! Even LLMs apply these principles now.

The image principle is a surprising outlier! Showing the instructor’s face doesn't reliably improve learning and often harms it. A talking head consumes visual capacity and attention without adding instructional value. Don’t take it personally, but the movement of your face won’t add to the understanding of any topic, unless we’re discussing anatomy, and even then…

Myths, misconceptions and reversals

Mayer’s research overturns multimedia myths by showing counterintuitive findings: “more” media doesn’t always mean better learning, subtitles can hinder unless needed, talking heads don’t necessarily improve outcomes, etc.

But do his principles always work? Mayer often talks about an important boundary condition: the expertise reversal effect. These principles apply mainly to beginners because once learners have strong mental models, extra guidance becomes redundant and uses up cognitive capacity.

Also consider: more recent revisions of the 12 principles add additional ones: embodiment principle, generative activity principle, immersion principle, collaboration principle, etc. So the list definitely isn’t a closed one and keeps evolving as new research emerges.

The modern landscape: videos, AI tutors and beyond

Here are some examples of platforms and channels that excel at implementing Mayer’s principles:

• Kurzgesagt - In a nutshell: Clear scientific explanations paired with high-end motion design. Strong signaling, tight narration and solid storytelling, though the playful style can add some extraneous load.

• Art of the Problem: Deep conceptual focus with minimal, purposeful visuals. Strong alignment with Mayer’s principles through clean coherence, synchrony and controlled pacing.

• Khan Academy: Function over flair. The handwriting approach keeps load low, uses effective modality and signaling, and supports learners with short segments and practice.

As we’ve seen, videos remain important in digital learning, even if multimedia design now includes interactive animations, simulations, AR, VR and increasingly AI tutors. Mayer’s principles apply across these formats because they tie technological possibility to how learning actually works within the limits of human cognition.

Educational innovation succeeds when it follows cognitive science. The key question for any new tool is not whether it’s new but whether it aligns with how we learn.

Because tools evolve, but our working memory doesn’t.

Checkable list - 12 principles (Multimedia learning, Mayer)

103KB ∙ PDF file

Download

Here's the full list of the 12 principles of multimedia learning as a PDF to download and use as a checklist.

Download

Keep learning

Prompt suggestions. Always ask follow-up questions:

• I want to make an educational video. Can you help me apply Mayer’s 12 multimedia principles to my script and visuals? Start by asking about my topic and audience.

• Explain the difference between extraneous, intrinsic and germane cognitive load using short examples from video learning. Then test me with 5 retrieval questions, one at a time.

• Many creators think “more visuals = better learning.” Can you show me how this belief conflicts with Mayer’s research and help me redesign a short explainer script to follow the coherence and signaling principles instead?

Links

• ▶️ A lecture by Mayer at Harvard explaining the principles: This 2014 video contains the full, hour-and-a-half-long recording of a lecture by Mayer himself, in which he explains an early version of the principles in a very accessible way. I can’t think of a better way to spend 90 minutes than watching this!

• 📑 Introduction to Multimedia Learning: Here you can read for free the introduction to the 3rd and most recent version of the Cambridge Handbook of Multimedia learning. It’s a great first step to start reading directly Mayer’s works. Spoiler: he starts by showing an example of multimedia instruction from 1657!

Related articles

How many books on leadership make you a leader? Is there a learning app for empathy or a course that guarantees better judgment? Can an online course really make you better at handling difficult coworkers? You already know the answer.

Theoretical knowledge offers direction and insight, but it doesn’t automatically translate into behavior. These skills play by different cognitive rules than maths, physics, coding or medicine.

Traditional education has long favored hard skills and neglected soft ones. However, research shows that soft skills are a strong predictor of success in life, often rivaling if not surpassing technical knowledge. With AI automating more of the “hard” work, we can expect that gap to grow.

Subscribe now

But how do we learn each type? Hard skills depend on memory: the brain’s ability to store and retrieve explicit information. Soft skills depend on experience: the brain’s ability to detect patterns, interpret context and act appropriately amid uncertainty.

So what’s the best way to learn each? And is the divide as clear as it sounds? Let’s see what the science says.

Know-what and know-how… but also know-when

Cognitive science says knowledge comes in two flavors.

Know-what (declarative knowledge) is about facts, concepts and rules. You can explain it, write it down and test it.

Know-how (procedural knowledge) is about doing. An expert dancer might pull off a moonwalk perfectly yet struggle to teach it step by step: they have the know-how, not necessarily the know-what.

OK, so hard skills are declarative knowledge and soft skills are procedural knowledge, right? Not so fast. Most hard skills include procedural knowledge too. A mechanic, nurse or IT specialist doesn’t just memorize steps: they execute them smoothly. Even calculating the result of a division or a square root is procedural knowledge.

In fact, most vocational training programs are focused on developing the kind of procedural hard skills that make you a professional expert in a certain area.

Soft skills require a third layer:

Know-when (conditional knowledge) is the judgment of when and under what conditions to use a specific skill or strategy. In other words, it’s about acting at the right time and in the right way. Giving feedback, leading a meeting, motivating a team or calming a customer all depend on it.

You probably haven’t heard about conditional knowledge because it’s not just “a third type of knowledge”, but rather a metacognitive function. It requires retrieving facts (declarative knowledge) and executing skills (procedural knowledge) simultaneously, mediated by highly complex executive control processes. In essence, it’s basically the two types of knowledge working together to manage themselves.

As a metacognitive function, its relevance is backed by scientific research. However, because it’s not a separate type, conditional knowledge hasn’t really made it into the mainstream: there isn’t even a Wikipedia page for it.

Hard skills: memory and automation

Hard skills are technical, measurable abilities: coding, maths, budgeting, biology, data analysis, chemistry, surgery or playing a musical scale. They require precision, consistency and rule-following.

Traditional education has focused almost entirely on them but usually in the least effective way. Rote repetition and rereading don’t build lasting skill. If you follow Kognitivo, you already know that cognitive science points to something better: desirable difficulties.

These methods reshape memory traces instead of just refreshing them. But don’t forget about generation, transfer or delayed feedback. Conditional knowledge also plays an important role in hard skills: experts know when to apply which rule or method, but memory (long-term declarative knowledge) remains the foundation.

The final step for any hard skill is what scientists call proceduralization: when rules become reflexes. Think about learning to drive: at first you juggle mirrors, pedals, and panic. Then, one day, you’re singing along to music and changing lanes automatically.

Soft skills: deliberate practice and context

Soft skills are behavioral and interpersonal abilities: communication, empathy, adaptability, teamwork, leadership. They’re what make workplaces functional and relationships sane. Memory still matters, but it’s not the main engine.

If conditional knowledge was important but not the main character regarding hard skills, with soft skills the situation is different. The complexity, nuance and diversity of the contexts in which soft skills are required makes knowing when and how to act in a certain way or another the most relevant ability to develop soft skills. But how do we achieve that?

Progress depends on deliberate practice (DP): focused repetition with clear goals and feedback. You refine your timing, tone and reactions the same way an athlete refines footwork, through repeated exposure, reflection and correction.

The concept of deliberate practice was introduced by cognitive scientist K. Anders Ericsson and colleagues in 1993 to underline the idea of expertise as the outcome of structure, goal-oriented, feedback-rich practice, and not innate talent. Deliberate practice builds conditional knowledge: knowing when to act and when to hold back.

The cognitive principle of transfer appropriate processing (TAP) suggests that we learn best when practice feels like the real thing (i.e. when retrieval context closely matches the encoding context). For soft skills this means: training should mirror real conditions, including stress, stakes and social dynamics.

If someone needs to learn how to manage conflict or stay calm under pressure, practice should recreate those pressures. A mismatch (calm classroom versus tense boardroom) kills transfer. That’s why simulations and role plays work: they teach your brain what performance actually feels like.

Does this mean that we can neglect memory or that we should give up on theoretical approaches when working on our soft skills? Certainly not: theory still matters. In the article about microlearning I mentioned how a recent 2025 study has shown that microlearning interventions also support the development of soft skills. The nuance is in how they support it rather than drive it.

As with the proceduralization of hard skills, the more you practice, the faster soft skills will become automatic. That frees up mental space for control: judgment, empathy, adaptability and metacognition. The irony is that becoming more thoughtful starts with making some actions thoughtless.

The hybrid case: language learning

Language learning mixes both worlds. Grammar and vocabulary are hard-skill territory: rules, structures and definitions. Conversation, tone and interaction are soft-skill territory: timing, nuance and communication. This dichotomy is usually referred to in language learning literature as fluency versus accuracy.

You can memorize every tense and still freeze in conversation if your learning never leaves the page. This is also why most language learning apps ultimately fail at making you speak the language naturally: they focus only on hard-skill drills, not soft-skill fluency.

Interestingly, modern language teaching has found a way to measure the soft-skill side of communication. The Common European Framework of Reference (CEFR) organizes language proficiency around communicative competence: what you can actually do with a language. From A1 (introducing yourself or asking for help) to C2 (resolving conflict diplomatically or interpreting nuance), each level describes real-life actions, not just vocabulary size or grammar knowledge.

Let’s make it practical: how to learn hard and soft skills?

• For educators and instructional designers: effective learning aligns method with mechanism. Hard skills thrive on memory-building strategies; soft skills demand experience, feedback and realistic context. Don’t lecture empathy: let people live through it in actual social contexts. Design learning that mimics the situations where the skill will actually be used.

• For learners: alternate study with doing. Drill when precision matters, but seek messy, unpredictable situations when adaptability does. Study the manual, then step into the cockpit. Memorize the script, then improvise. Learning works best when you teach the brain the way it naturally learns: through memory for facts, and through practice for behavior.

Why traditional education gets it wrong

Let’s not forget that strategies that benefit hard skills are also important for soft skills and vice versa. Unsurprisingly, deliberate practice is the best way to develop procedural hard skills. And memory work is also key for soft skills: knowing the facts and context of a negotiation is, more often than not, the key to success. The important question is which of these strategies plays a more prominent role in developing hard or soft skills.

In any case, it’s important to bear in mind that our perception of what learning and education means is strongly influenced by the model of hard skills. Schools and universities have spent decades perfecting how to teach maths, physics or grammar because hard skills are easier to measure. Standardized tests reward what can be graded: facts, formulas and tidy answers.

This is also why the limited soft-skill education we receive in school is often boiled down to their “hard-skill components” without really exploring soft-skill territory. For instance, we may be asked to “define decision-making” or “list the criteria for good judgment,” yet we aren’t required to make decisions or demonstrate good judgment.

And in a world where AI can crunch numbers and write texts faster than any human, it’s the human skills that will set us humans apart.

Keep learning

Prompt suggestions. Always ask follow-up questions:

• I want to improve both hard and soft skills in my learning routine. Can you help me design a weekly plan that balances memory-based study (for hard skills) with deliberate practice and feedback (for soft skills)?

• Explain “know-what,” “know-how,” and “know-when” in simple terms, then help me identify which one I’m currently neglecting the most in my learning or work.

• Act as a teacher and test me using retrieval practice on “the cognitive science of learning hard and soft skills.” Ask me 6 questions, one at a time, only continuing when I answer. Make them progressively harder.

Further reading

• 📖 The making of an expert: The scientist who coined the concept of deliberate practice shows how top performers are made, not born. Ericsson’s research dismantles the myth of talent and shows the science behind real, measurable mastery in any field. The article is very accessible and really juicy.

• 📑 The power of soft skills beyond academic achievement: This long but accessible article from 2025 sums up the research on the topic of soft skills and their importance. It’s a great next read after this article!

Related articles

Have you ever tried explaining why popcorn pops to a five-year-old? You can’t exactly talk about vapor pressure. Instead, you’d probably say something like this: “It’s like a balloon that bursts when it gets too hot.” Or consider photosynthesis. You might say, “Plants eat sunlight.” Technically false, but cognitively brilliant.

If you’ve ever done any of these (and I’d be very surprised if you haven’t), congratulations, you’ve used one of the oldest teaching tools in the world: analogies.

Analogies make complex concepts accessible, relatable and intuitive. They’re often the reason behind many of our most memorable learning moments. But analogies don’t always land. Some mislead, others fall flat. How can we make sure that they lead to the desired “aha” moment? And what does the science say?

What’s an analogy?

An analogy compares two things that share key features. We use one to explain the other because those shared features are easier to notice in the second. The resemblance runs deeper and more systematically than looks or labels: it’s about how the parts relate and what roles they play.

A few things to remember:

• The two elements in an analogy don’t have to be physical objects. They can be situations, dynamics or actions: like photosynthesis and eating.

• Analogies don’t need a perfect match to work. Most share only a few features, and that’s enough. In fact, if the match is perfect, it’s probably not an analogy, but an equivalence: “The sun is like a star,” isn’t an analogy because the sun is a star.

• Researchers often define analogies not by referring to shared features but to a “common structure of relationships” or “relational mapping.” While technically more accurate, I feel like that phrasing can make something intuitive sound unnecessarily abstract.

  Subscribe now

A comparison? A metaphor?

Not quite. These terms overlap, but they’re not interchangeable. Analogies are a type of comparison and metaphors often express analogies, but there are non-analogical comparisons and metaphors that don’t convey analogies.

• A comparison contrasts features or quantities. Non-analogical comparisons don’t imply shared features, just contrast: “Berlin is colder than Barcelona” (it is) or “She speaks better Japanese than I do.”

• A metaphor is a linguistic device (“X is Y” or “X is like Y”) that can express an analogy but not always. “Time is a thief” or “He has a heart of stone” are metaphors, not analogies since they’re evocative, not structural. In any case, as you can see, the boundaries can get blurry.

Why analogies matter for learning

Analogies are cognitive shortcuts that help grasp new ideas using the mental models we already have. They let the brain recognize familiar relationships in unfamiliar territory, fostering knowledge transfer.

They trigger that “Oh, so it works like that” moment where comprehension clicks. Analogies are a natural way to use knowledge transfer for memory encoding. Cognitive science has long supported their use for learning and teaching, with early work dating back as far as 1966.

A now classic study from the 80s examined how many medical students could solve a medical problem involving stomach tumors. To help them figure out how radiation therapy destroys tumors, some students were given a military analogy. The analogy was based on the story of a general who captured a fortress by dividing his army into small groups and deploying them on different roads. The goal was for the students to come up with the “ray convergence solution”: small low-intensity rays from machines positioned around the patient converge to destroy the tumor in the same way as the small army groups took the fortress.

More recent studies and meta-analyses confirm it: comparing cases and then making principles explicit, consistently leads to deeper, more transferable learning.

The danger of analogies: misconceptions

Research shows that analogies improve learning. But analogies are an inductive learning strategy. They help learners infer general principles from examples and that means that they can also spread misconceptions. If students generalize the wrong kind of similarity, the analogy can backfire.

Poorly chosen analogies can lead to featural distraction, that is: irrelevant features overshadowing the ones the analogy was intended for. Other factors that play an important role include, crucially, students’ prior knowledge and metacognitive abilities, as well as effective teaching. In fact, some studies have shown that using analogies without further context can even hinder less experienced learners.

To prevent this, show where the analogy holds and where it breaks. Emphasize explicitly the structural correspondences and state which features don’t match. Making limits explicit ensures that students with fewer cognitive resources aren’t left behind. The best way to counterbalance the potential dangers of inductive techniques (such as analogies) is to provide the formal rule right after the generalization is made. This approach achieves the generation effect while preventing misconceptions.

Interestingly, research also suggests that well-designed analogies that lower the cognitive load with specific cues, can actually help struggling students. The key is in how the analogy is designed and what role it plays within the learning experience.

Original analogies

The secret ingredient that makes analogies truly stick, is, however, originality. Some analogies are so popular that they’re almost the standard way to explain certain concepts: mitochondria being the powerhouse of the cell, a fraction being like a piece of a pie, electricity flowing like water through a circuit, an atom being like the solar system or the east-west European division in the late 20th century being an iron curtain.

I call these the “greatest hits” of analogies. These well-worn analogies have even become part of everyday language in many cases, and they do a solid job because they’re familiar and easy to recall.

The fresh, custom-tailored comparisons that match the exact concept and the prior knowledge of the audience you’re dealing with, make learning more vivid. Such as the analogy of the general for the medical students treating tumors.

Here are some examples of analogies I personally like: the immune system is like a detective agency (immune cells are detectives finding suspects) or inflation is like watering down soup (there’s more quantity, but the value is less). These work for a rather general audience: depending on who the learner is, one could use even more specific ones.

Original analogies are more likely to turn abstract lessons into “aha” moments. They’re harder to come up with, but that’s where the artistry of teaching lives.

Using analogies to learn and teach

1. Avoid featural distraction: Pick examples with clear, relevant shared features.

2. Highlight the roles. Don’t just say “A is like B”: explain how.

3. Compare two cases. Side-by-side analogies reveal patterns.

4. Name the principle. Once learners see the link, give them the rule. As a learner, make sure to make the rule explicit.

5. Use LLMs. ChatGPT, Perplexity or Claude are a great help when brainstorming analogies.

6. Come up with your own analogies: both for teaching (original analogies are more memorable) and learning (creating analogies triggers the generation effect). Second-hand analogies can’t beat tailor-made ones!

7. Make it, then break it. Don’t just take or make analogies, discuss where they fail, that act of critique deepens understanding.

You’ll find an extensive example of this last point in my discussion of AI and the calculator analogy. Another discussion of an analogy that I also highly recommend is psychologist Robert Epstein's questioning of the brain-computer analogy. These two articles exemplify the principle:

The guiding role of analogies

Analogies don’t just explain concepts: they frame how we think about them. A well-chosen analogy can set the stage for how students approach an entire field of study, guiding their mindset from the very start.

Let’s take language learning as an example. A common analogy is that learning a language is like learning math: grammar rules (formulas) and vocabulary (variables, numbers) are combined to get the correct result. This analogy sets a logical, formulaic frame around language learning. It highlights the “hard skill”-side of learning languages: vocabulary, syntax, grammar…

But change the analogy, and you change the learner’s perspective. Imagine telling students that learning a language is like playing a sport. Yes, you need to know the rules, but rules alone won’t make you fluent. It’s the practice (the repetition, the real-life usage) that builds true skill. This analogy shifts the focus from memorizing rules to active, hands-on practice. It highlights that learning a language involves gaining practical or procedural knowledge, needed to develop soft skills such as communication, expression, conversation, interaction, etc.

Similar analogies can be used in different fields and will make learners focus on different aspects of the learning field: in psychology the brain as a computer vs the brain as a muscle, in sociology the economy as a machine vs the economy as an organism, or in software engineering coding as writing vs coding as architecture. Analogies steer thinking before the first lesson even begins.

Analogies are cognitive scaffolds

Analogies are not decorative comparisons: they’re cognitive scaffolds. (That’s an analogy of an analogy or a meta-analogy!) They turn confusion into coherence by connecting the new to the known, and building new, higher layers on the foundation of what we already know. Like any cognitive tool, analogies also have limits. Their instructional power depends on how well they’re designed and implemented. The scaffold needs to be built with care, stable enough to support understanding as it grows.

When we teach through analogy, we’re not just transferring ideas; we’re transferring structure. When analogies work, they turn chaos into clarity, letting knowledge take shape. When the scaffold holds, ideas don’t just stand, they rise.

Keep learning

Prompt suggestions. Always ask follow-up questions:

• I want to teach a difficult concept using an analogy. Can you help me design one that’s accurate, memorable and avoids featural distraction? Start by asking me what topic I’m teaching.

• Act as a teacher and test me using retrieval practice on “analogies in learning.” Ask me 6 questions, one at a time, only continuing when I answer. Make them progressively harder.

Further reading

• 📑 Teaching by analogy: from theory to practice: This incredible article published in Mind, Brain and Education gives a detail-rich, extensive explanation of analogies and their use in teaching. It explains the study about the military analogy to fight cancer in a very engaging way including visuals. If you’re only going to read one more thing about analogies, let it be this!

• 📑 Analogy entry on the MIT Encyclopedia of cognitive sciences: This article explains in depth the concept of analogy as viewed by cognitive science, its different kinds and limits. It’s a great next article to read after this, as it goes deeper into more technical grounds, while staying accessible.

• 📖 Metaphors we live by: This classic book analyzes the structural role that metaphors (both analogically and non-analogical, but mostly analogical) have in shaping our language and therefore our thinking. For instance, they analyze (among others) how we usually speak about time using the language of money (invest time, lose time, spend time, save time). Highly recommend!

Related articles

Why can we recall the plot of a forgettable Netflix series from 2012, yet blank on last month’s compliance training? That’s not a memory flaw, that’s poor design.

Ten thousand years ago, humans gathered by a fire, learning to survive through stories. Today we sit by different fires: screens, classrooms, books. The medium changes, the technique doesn’t. Stories aren’t just entertainment, they’re memory technology.

What do we mean by storytelling?

In instructional design and teaching, storytelling means using narrative structure, setting or characters to help learners build a coherent mental model of the learning content. That coherence is the cognitive scaffold that makes information stick.

Storytelling is a powerful way to convey the relevance of a learning topic. It answers the learner’s question: why should I care? It helps learners understand why a concept or learning outcome matters, what it means in practice and how mastering it can be useful in real life. Storytelling highlights or creates a need, a sense of urgency for learning that the learning material (explanations, exercises, etc.) addresses.

The mechanism most often studied is narrative transportation. When people feel absorbed in a story, attention, emotion and imagery line up and memory improves. That’s why the same facts wrapped in a narrative are easier to recall and apply.

How much better do we learn with stories?

The idea that stories supercharge memory is hardly an undiscovered truth. But how much? Marketing blogs and LLMs often cite psychologist Jerome Bruner as saying that facts told as a story are remembered up to 22 times better. Well, that’s internet folklore: Jerome Bruner never said that!

However, a made-up exaggeration doesn’t disqualify the underlying reality. The real data is still striking: when learners wove items into a story, their long-term serial recall was about 93% vs 13% for rote study, roughly 6-7× better.

Is 10-12× ever realistic? Depending on the baseline and the task, some storytelling or story-based formats produce higher gains, but effects aren’t constant. Treat the headline number as a direction, not a law.

But why the advantage? Stories reduce cognitive load. They create cause-and-effect links, inferences that are easier to recall than isolated facts. They hold attention, evoke emotion, and promote transfer. Recent studies keep finding the same thing: when the story structure fits the goal, outcomes rise.

What counts as a story?

When we think of “stories” we often picture a character arc, like a hero’s journey. But character-driven plots aren’t the only kind of story, and in education they’re often overkill. Here are kinds of stories that work in education:

• Linear stories. A simple beginning–middle–end is great when learners need coherence more than control. Typically, there’s a challenge or twist in the middle that’s resolved at the end. Explanations and exercises usually fit between middle and end.

• Setting-based storytelling. Rather than a hero, a situation is the anchor. The setting makes the learner feel the need to learn something that the material (explanations, exercises) conveniently fulfills.

Setting-based storytelling can be further divided into different categories depending on the kind of arrangement.

• Branching stories. Learners face decisions that shape the path: if you choose A, you go to scenario 2; if you choose B, you go to scenario 3. This is useful for judgment and “it depends” skills, and it aligns with simulation and situated learning research.

• Case-based stories. Real or fictional cases for analysis are strongly supported in health and professional education, showing gains in problem solving and communication.

• Problem or mystery structures. Start with a puzzle, reveal clues and drive to an explanation. Problem-based learning meta-analyses report gains in critical thinking versus lecture, especially for complex domains. The learning content (the skill to be learned) should be the key to solving the problem.

• Transmedia arcs. Spread the story across formats, from short video to chat simulations. Transportation theory predicts higher absorption when imagery and coherence are well designed.

Summary: Linear stories work best for coherence, branching for judgment, cases for professional practice, mysteries for critical thinking and transmedia to create immersion.

How much storytelling is too much?

I have worked extensively with teachers, content writers, materials developers and instructional designers, and one thing is clear: education professionals love storytelling. As creative experts crafting learning experiences, they typically invest themselves quite intensely in the stories they build. But this can be counterproductive.

Story can help or hurt. The seductive details effect shows that interesting but irrelevant flourishes lower learning outcomes. Two meta-analyses (2020, 2012) and newer replications find small-to-moderate negative effects when details don’t serve the goal. The bottom line: trim the fluff.

Do we need characters?

One common myth is that every story needs characters. Not always. Characters can boost empathy, but they often add noise. A bare situation is often enough. The real test is: are characters necessary for the outcome?

In some cases, they are necessary. Medical training, for example, requires people because the learning outcomes are about treating people. Consider: “Two patients arrive at the doctor’s with fever and cough. They suspect covid.” Clear, simple, on target. Adding names might help distinguish them, but that’s all we need.

On the other hand, consider: “Miguel and Oliver, lifelong friends who party hard, wake up sick after a late night…” That background is wasted space. It adds nothing to the learning goal.

Guideline:

• Use characters only if the outcome requires them.

• Add background only if it serves the goal.

• If generic, use “you” as the protagonist (“Imagine you…”)

A common universe across modules

There is little research on using a common world or universe across modules. This is popular in language learning books (but not only) with recurring characters, settings and locations. In theory, this reduces cognitive load.

In practice, many learners skip modules and miss parts of the world. Topics often don’t fit naturally either. Writers often end up bending learning goals and content to match the story instead of the other way around. While a consistent world can help with branding and emotional connection, it also risks seductive details. Often, the effort isn’t worth it, but at the end of the day, it’s a matter of weighing pros and cons.

More practical tips for your stories

On top of the tricky topics of characters and world-building, here are some tips on how to keep your storytelling efficient and spot on:

1) Wrap micro-tasks in micro-stories.
Swap sterile prompts for tiny incidents. Instead of “Translate this sentence,” try “Your friend texts this. How do you reply?” It cues prior knowledge and invites inference.

2) Use branching for judgment, not procedures.
Branching shines where choices are ambiguous and feedback can show shades of gray. For step-by-step procedures, branching is overkill. Tie decisions to consequences learners can feel. Pair with short debriefs to surface the rule, ensuring the correct information sticks.

3) Start with a mystery, end with a payoff.
Open with a question or anomaly, then resolve it. Mystery structures sustain attention and give you a natural knowledge check at the reveal.

4) When stakes are high, simulate.
For complex skills, use narrative simulations. The evidence base for simulation-based learning is unusually strong. Even short scenario videos with branches show gains in applied knowledge.

5) Invite learners to tell the story.
Have learners generate brief narratives of a solution or error they encountered. Generation plus narrative chaining boosts recall and transfer.

Not all storytelling is focused on learning

By now we know that storytelling and learning don’t always go hand in hand. A famous example comes from Homer’s Iliad. In Book 20, Homer describes in unexpected detail how Aeneas’ spear cuts through Achilles’ shield’s many layers of metal (cuts through bronze, stops at a layer of gold). For centuries, scholars assumed Homer provided a technical description of shield construction. More recent scholarship shows the design would make the shield useless in battle. The point was never metallurgy but symbolism.

This is a reminder for education: not all storytelling is focused on learning. A beautiful story can carry symbols, emotion and cultural significance without teaching a skill. Instructional storytelling has a narrower brief: it should support outcomes without weighing the learner down.

Learning time is story time… with a “but”

Storytelling isn’t window dressing. It shapes what learners notice, feel and remember.

Design the narrative to fit the goal, keep details coherent and right-size the arc for your audience. When you do, stories do what bullet points can’t. If you want learners to remember, give them a story worth retelling.

Or in one line: good stories teach because they serve the learning, not themselves.

Share

Keep learning

Prompt suggestions. Always ask follow-up questions:

• I want to design a short lesson using storytelling. Can you help me decide whether I should use a linear story, a branching scenario, or a case study for my topic?

• Stories stick, but seductive details don’t. Can you give me examples of how to trim distracting background info from a learning story while keeping it engaging?

• Act as a teacher and test me using retrieval practice on “storytelling in education.” Ask me 6 questions, one at a time, only continuing when I answer. Make them progressively harder.

Links

There are plenty of online resources on storytelling and teaching that provide tips on how to craft stories. Often, these resources don’t consider whether the advice is excessive, which it usually is. I have two recommendations: a tool to help you efficiently build great stories and research on the seductive detail effect to help you keep your stories on track.

• Storyteller Tactics: Pipdecks produces decks of cards with tactics for tackling specific issues, such as productivity, team management and… storytelling! Each card provides a tactic to help you develop a narrative, such as “Downfall” or “The Dragon and the City.” These tactics are structures that can be applied in multiple settings. Sadly, they’re not available for free, but I wholeheartedly recommend them.

• Seductive details hamper learning even when they do not disrupt: If you’re going to read one thing after this article, let it be this paper on the empirical findings related to how seductive details are detrimental to learning. I found other meta-analyses on the topic, but they’re sadly behind paywalls.

Related articles

You sign up for an online course, eager to learn. But then the syllabus hits like a brick: fifty hours of content, dozens of modules and long video lectures. You start with good intentions, but soon life gets in the way. Work emails, family plans… and suddenly that course is gathering digital dust.

Microlearning claims to fix this by breaking down large topics into small, easy-to-digest pieces. You can squeeze a lesson into a coffee break instead of blocking out an evening. Apps like Duolingo, YouTube explainers, TikTok clips, and podcasts all work this way. This method has also become the standard in modern corporate training. Microlearning delivers quick hits of knowledge, no stamina required.

But does this convenience have a ceiling? These tools deliver bursts of information, but do they provide the depth needed for lasting skills or memory? They scratch the itch for instant progress, but do they help build true mastery? Can microlearning genuinely replace long-form courses, or is it just another corporate buzzword? The answer, as always, is in the details. Let’s find out.

What is microlearning?

Microlearning is usually defined as the process of delivering content in small, highly-focused units, each designed to achieve a single learning objective and to be readily available when the learner needs it. These short bursts of information, called “microunits”, “nuggets” or “snippets”, are typically completed in under 10 minutes and are meant to be a standalone piece of learning, making something like a minimal meaningful learning unit.

While definitions vary (some argue that more than 5 minutes is too long), most agree that the key isn’t just the length, but also the design. Microlearning works when the units are purposeful, targeted and tied to outcomes. There is scientific evidence to support this: a 2024 meta-analysis of 600+ higher education students found that thoughtfully designed microlearning led to measurable gains in academic performance compared to traditional long-form courses. While the research base is still relatively small, these findings are consistent with those of most other studies.

The science of learning small

The real strength of microlearning lies in chunking, a core cognitive strategy. Chunking means breaking down complex subjects into smaller, digestible bits, so your brain isn’t overloaded. Each module tackles one idea, making it easier to process and remember.

The word “micro” makes it sound like we’re pandering to short attention spans, but the real target is cognitive overload. According to Cognitive Load Theory, working memory can only handle a few ideas at a time. If you dump too much information at once, you overwhelm this capacity.

Chunking reduces cognitive load and works both top-down (breaking down big topics) and bottom-up (grouping small bits into patterns). Think of how we handle phone numbers: instead of memorizing ten separate digits, we often group them into chunks like 012-456-7890. The same principle works in language learning: for a beginner it’s often easier to learn “How much is it?” as a single phrase than to learn each word in isolation.

But microlearning isn’t only about chunking. When delivered continuously or in “drip” form, it also taps into spaced repetition. By revisiting material at intervals, it strengthens memory. This is a key difference between learning within a well-designed course and passively watching videos as they are suggested by an algorithm.

How micro is microlearning?

For instructional designers, defining the smallest meaningful learning unit isn’t an easy task. How short is too short? When is a topic too big? How much can you chunk together? Simply cutting a 30-minute video into five-minute segments isn’t microlearning, that’s just a chopped-up mess. The key is intentional design and focus.

For example, if you’re teaching Excel, “How to use Excel effectively” is far too broad. But a five-minute module on “Creating a pivot table”, “Using VLOOKUP to merge data” or “Applying conditional formatting” is more achievable and focused.

Each one of these is a discrete, achievable skill. How do you identify these skills? The time factor becomes a useful, though not rigid, guideline. Ask yourself: Can a non-expert realistically achieve this single outcome in under ten minutes? If your answer is yes, you've likely found a solid microlearning topic.

The real power lies in the focus, not the length. The time limit just helps you zero in on a single, digestible outcome. If you’re feeling more ambitious, use an even tighter time frame as a reference (2-5 minutes) to sharpen the focus.

From knowledge to action

Microlearning isn’t just about consuming information, it’s about using it quickly. One of its greatest strengths is delivering knowledge at the exact moment it’s needed, which makes knowledge transfer to real tasks far more likely.

This is the essence of the just-in-time model. Imagine a salesperson about to call a client: instead of recalling an entire negotiation course, they watch a three-minute module on “handling objections.” Or think of a nurse checking a short refresher on dosage calculations before administering medication. The knowledge is immediately tied to action, which strengthens memory and confidence.

Learners don’t just know something, they use it at the right time. That’s why microlearning often feels more practical than traditional courses. It collapses the gap between theory and action, turning learning into performance support.

Where microlearning falls short

Every strength has a shadow, and microlearning is no exception. Let’s look into where it tends to fall short.

• Fragmented design: Microlearning is often criticized for fragmented design. When content is chopped into scattered nuggets, learners lose sight of the bigger picture. Chunking should reveal structure, not hide it. Done poorly, microlearning leaves people with trivia instead of a mental model. However, that’s not a flaw of microlearning itself, but rather of weak design.

• Learning skills that need depth: Microlearning works best for tasks that can be split into clear, measurable parts: terminology, compliance rules, onboarding steps. But complex fields like law, medicine or architecture need context, immersion, complex judgment calls and the grind of deliberate practice. Microlearning still has a role to play: it can reinforce foundational knowledge, deliver just-in-time reminders and help prevent forgetting, but it can’t replace the sustained effort and depth needed to build true expertise.

• Soft skills: One controversial take comes from learning scientist T. Maddox. He argues that microlearning works well for hard skills like software use or coding, but struggles with soft skills such as communication, leadership or empathy, which require behavioral practice in varied scenarios. However, recent research challenges this idea, such as a 2025 study in Frontiers in Psychology. It shows that well-designed, discipline-specific microlearning can improve some soft skills (unlike the generic “one-size-fits-all” approaches that Maddox criticized). That said, Maddox’s caution still holds. Even the best microlearning can’t match the impact of role plays, simulations or practical interaction when it comes to soft skills. Microlearning helps, but it’s not the optimal way to develop behavioral skills.

To micro or to macro… that is the question!

Microlearning has its limits, but the overall balance tilts in its favor. Studies back it as a credible alternative to long-form training, provided it’s well designed. What matters is how deep the knowledge needs to go and the type of skill you’re teaching.

Comparing microlearning to traditional long-form courses is somewhat unfair because, for many learners and organizations, the long form was never an option. Thanks to its accessibility and flexibility, microlearning makes learning possible where it wasn’t before. This achievement alone can’t be overstated.

Yet the real risk of microlearning isn’t inefficiency, it’s complacency. Microlearning is so convenient it can steer you away from the harder methods that build deeper expertise. Think of the daily Duolingo streak that feels rewarding but delays booking a real conversation class. Or the endless YouTube tips on AI that spark insights but stop you from starting a demanding project.

Microlearning is like espresso: strong, quick, easy to get and perfect for a hit of focus. But if you want to build expertise, sometimes you still need the whole pot.

Share

Keep learning

Prompt suggestions. Always ask follow-up questions

• I want to design a microlearning module for [topic]. Can you help me frame a single, focused learning objective that could realistically be achieved in under 10 minutes?

• Microlearning feels great, but I worry I’m just collecting trivia. How can I tell if my microlearning is building a bigger picture instead of fragmenting knowledge?

• Act as a teacher and test me using retrieval practice on “microlearning: strengths, limits, and design principles.” Ask me 6 questions, one at a time, only continuing when I answer. Make them progressively harder.

Related articles

Consider the long and ugly German word “Aschenbecher.” (ashtray) I think even my German friends would agree it’s not the most beautiful word in their language.

My sister, a non‑smoker learning German while working in a Berlin restaurant, met this word twice. She learned it first in a German class, but the word didn’t stick. Later, when she started working at the restaurant, customers kept asking for an “Aschenbecher”. Suddenly it wasn’t just a list item anymore, it became a job requirement. The word instantly stuck in her memory.

Why didn’t it stick in the classroom but did at the counter? Would a smoker have remembered it better at the German course? What does this anecdote tell us about how learning and memory work? And most importantly: How can you recreate this process in your study sessions to boost your learning?

High-value information

The guiding idea is value-directed remembering: when your brain tags information as high value, it gets more attention and sticks better. This happens when the information helps with current goals, needs and interests, fits identity or carries rewards. Lower‑value material is not erased, it just slides down the brain’s priority list. Attention is a budget, relevance pays the bill.

Value from solving a need is the real lever. When information is instrumental to a task, the brain upshifts attention and engages reward systems, which deepens processing. That extra processing strengthens the memory trace at encoding. In practice, learning follows task demands more than study intention.

Make it useful on purpose

You can use the principle behind value-directed remembering in your study sessions. The job is to make the target information feel high value to the brain. But that’s easier said than done: how can you make a random page in your textbook feel like high value information? Most of the time it’s not relatable information at all. You can’t fake value. The brain notices.

That’s why most students choose a different but flawed strategy: they intentionally set their brains to directly focus on learning the information. Ironically, simply trying to understand is a poor way to achieve understanding. The brain does not follow commands like “Please, understand this.” Intention is a weak instruction set.

Rereading, rewatching and underlining without a goal other than “learning” will boost familiarity, not mastery. “Drill and kill” looks busy, but it does little for encoding information into your memory.

The key to leveraging the concept of high-value information is to focus on performing cognitive tasks that require the information, as opposed to focusing on the information in the abstract.

By performing a cognitive task that requires an understanding of the information, you’re actually creating the conditions for your brain to see the information as high value within the learning process, even without real-life needs having to be met (such as my sister’s job at the restaurant).

Learning is a byproduct of cognitive work

The counterintuitive reality is that learning and understanding aren’t cognitive tasks. Learning isn’t something we do. It’s rather something that happens when we do the right things. More precisely, learning is a byproduct of specific cognitive actions that force deep processing of the target ideas.

Your brain learns new information when processing it is necessary to complete a cognitive action, something you do. But what is a cognitive action? Deciding, generating, simplifying, grouping, comparing, justifying, those are actions. Internal actions are invisible; we only ever see the outcome of our mental work (the text, the slides, the diagram…). But it is the unseen work that causes learning.

The shift is simple: stop measuring exposure, start measuring the number and quality of cognitive actions performed.

What cognitive tasks produce learning?

Summarizing

Simplify to essence. Put the same idea in fewer, plainer words, without the fluff. Summarizing is prioritizing: it forces a decision about what is central and what is garnish. If a concept can’t be said cleanly, it hasn’t been understood cleanly. Summaries are not old‑school busywork, they force you to engage meaningfully with the information. Example: while reading, give a title to every page on your textbook in 1 sentence.

It goes without saying that if you offload summarizing to an AI, it doesn’t work: it’s your cognitive effort that matters. Also consider: do different rounds of simplifying, summarize the summary again. It’s about the process, not the outcome.

Grouping ideas and concepts

Group to see structure. Placing items together pushes you past surface features toward shared logic, traits or implication. Grouping forces you to confront the pieces at a cognitive level. Example: cluster ideas by function or mechanism, then name each cluster in three words.

This is the logic behind mind maps: they are membership tests. Does this belong here? Why? Also: Groups do not have to be final. Regroup as new information arrives. The point is not the map, the point is the mapping.

Comparing and connecting

Make differences and bridges explicit. Comparisons force criteria, connections force an account of inner parts and logic. Analogies can help here. Example: “X is like Y because of A, B, C.” The juicy bits are the A, B and C.

When analogies break, that fracture is a gift: it reveals what the concept truly is (and is not). First drafts can be wrong, it’s not about getting it 100% right, since revising the draft is the superior task. Questioning analogies can be even more productive than creating them. You can also build mind maps that group, connect and clarify hierarchies.

Judging the information

Here is one place where being judgy pays off… not people, but information! Judging is about categorizing information as more or less important, as good or bad, as high priority or low priority. When you judge, you tie information to a principle, which is a small act of transfer. Judging means building hierarchy and this is another way of organizing, regrouping and sorting out information. But remember: judge information, not people! Example: Rank the five core ideas high → low and note the rule used.

What is encoding?

Summarizing, grouping, comparing and judging are simple cognitive tasks that trigger durable memory by making the information instrumental, but of course there are many more: rephrasing, justifying, storytelling... These tasks are a simple way to implement the generation effect, which is a desirable difficulty in learning.

Great learners do this almost intuitively, even while reading. You don’t necessarily need to write or produce any output (you can also summarize and make connections in your head, even though writing helps). And remember: do not stop at one pass. Summarize the summary, regroup the grouping, revisit comparisons, question past judgments again, and then again… especially as new information arrives.

The technical term for this process is encoding: converting information held in working memory into a durable long‑term trace. This information is then stored and can later be retrieved from the long-term memory.

Things to pay attention to

The kind of encoding task isn’t the whole story (of course, that would be too easy). Here are a few factors that also have a big impact and that I’ve already discussed here in Kognitivo:

• Are you at the right level of challenge? If these tasks feel frictionless, they are too easy. The purpose of a cognitive action is to push past surface recognition into deep processing and rich details. Aim for the optimal challenge point. If you don’t feel the struggle, you’re likely not learning.

• Don’t overdo it: Working memory is limited. Design tasks so attention goes to meaning, not juggling details. Reduce cognitive load: segment material, spotlight what matters, keep prompts sharp. Spend bandwidth on meaning, not clutter.

• Retrieve, space, interleave: Encoding is the beginning, not the end. To keep information, pull it back out. Replace review with recall, then check, spaced over time. Mix topics so the mind stops coasting on sameness and starts noticing differences. Retrieval makes memory stronger; spacing makes it durable; interleaving makes it transferable.

You will have learned

Better encoding requires sitting with discomfort and making a habit out of thinking hard. The tasks above are the exercises, the burn is the point. Memory adapts to the quality of cognitive work it is asked to perform, again and again.

And yes, I know it feels strange to see it this way, but learning doesn’t happen because it’s intended, it happens because it’s required by the task one sets for the mind. Learning is not an action, but the result of an action. Strictly speaking, you’re never learning, as in “doing the act of learning”. Learning is what “will have happened” if you do certain actions the right way.

The confusion may stem from the fact that we use a verb (to learn) to refer to something that’s not an action, but a result. We understand this better if we use a noun and talk about producing learning, leaving the verb-form for the tasks that produce it: to summarize, to group, to connect, to judge…

If you found this topic interesting or if this article made you reflect, make sure to subscribe to receive more insights and practical tips on how we learn. And share this with someone who totally “will have learned” about this!

Subscribe now

Keep learning

Prompt suggestions. Always ask follow-up questions:

• Turn low-value info into high-value tasks: Give me a boring textbook fact and then show me how to transform it into a task (summarizing, grouping, comparing, judging, etc.) that forces my brain to treat it as high value.

• Value-directed remembering in action: Help me design a study activity where I can make abstract textbook material feel “high value” to my brain. Use concrete examples from science or history, and guide me to turn them into tasks that require summarizing, grouping, or comparing.

• Diagnose my study habits: Ask me to describe how I usually study for an exam. Then evaluate which parts are just “exposure” and which involve true cognitive actions. Suggest two specific changes that would make my encoding deeper and more effective.

Related articles

ChatGPT’s Study Mode hasn’t lived up to expectations.

Only a month after its launch, Study Mode faded from the conversation. Not just because GPT-5 stole the spotlight or because it’s summer and students are on break, but because it hasn’t delivered on its promise to transform learning.

In this article, I’ll show why Study Mode falls short and how you can instead apply key insights from cognitive science to maximize learning with AI. It’s not about hacks or perfect prompts. It’s about intentionality and… science!

Where Study Mode falls short

At first glance, Study Mode appears to be a learning upgrade. However, in practice, it feels more like a slightly repackaged version of ChatGPT than a genuinely new approach. Here are some issues that many users have flagged:

• Answers before the right time: Although it provides step-by-step guidance with Socratic questioning, it often provides explanations and solutions too early (sometimes in the first exchange despite its system prompt). This takes away the productive struggle of guessing and figuring things out on your own.

• Shallow structure: It works best with homework-type problems. For concept explanations, it loops through superficial explanations that are then simply rephrased as questions. It doesn’t build toward a coherent, goal-oriented lesson.

• Missed optimal challenge point: By making things too easy, it undershoots the “desirable difficulty” needed for growth, leaving the learner just outside the optimal zone of challenge. Add ChatGPT’s sycophantic habit of telling you what you want to hear and it’s easy to feel like the sessions are ineffective.

One could argue these study modes are just the start of more learner-driven LLM experiences (likely true), but for now, they’re still only a superficial layer over the usual LLM interaction. For deeper critiques, see Dr. Philippa Hardman’s test and Daniel Nest’s comparison. While also imperfect, many agree Gemini’s Guided Learning mode performs better.

Better than Study Mode: own your learning

There’s an uncomfortable truth here: every AI use involves a choice. Either we use AI to learn or we use AI to bypass learning. No preset mode, prompt, custom GPT or strategy will magically turn the avoidance of learning into learning, because it’s actually a choice, and it’s yours to make.

If you choose to take ownership of your learning, research shows you should also know how the AI is teaching you. Metacognition, or knowing how you learn and adapting your learning accordingly, is one of the strongest predictors of academic success. Before your next AI session, ask yourself:

• What exactly do I want to learn?

• What strategy will I use to get there?

• How will I know if it’s working?

The first is straightforward; the next two are less obvious. Most AI advice focuses on cheating or making AI outcomes “look real”. But what about using AI well to actually learn?

Here’s a cheat sheet with three science-backed strategies that I’ve covered in previous posts and how you can leverage them into a full-fledged AI-powered lesson.

Three science-backed strategies

1. Knowledge transfer

Principle: Applying knowledge to new contexts builds flexible, adaptable understanding: the hallmark of expertise.

How can I implement that?

• Examples: Ask for real-life situations where the concept applies and come up with your own: “Give me two real-life examples of how supply and demand works, then I’ll try to come up with my own.”

• Analogies: Request comparisons from similar concepts in different contexts, then build your own: “Compare how electricity flows in a circuit to how water flows in pipes and prompt me to come up with my own analogy.”

• Problem-based learning: Ask for a problem that can be solved by implementing the concept, then solve it. “Describe a realistic problem that can be solved using the Pythagorean theorem, then let me figure out the answer.”

• Storytelling: Have the AI place the concept in a story, then become the storyteller. “Tell a short story where Newton’s first law is essential to the plot and now I’ll write my own story using the same law. Give me feedback.”

Evidence: Transfer is hard but critical; research demonstrates that prompting learners to apply concepts to new domains improves problem-solving ability.

2. Retrieval practice

Principle: Repeatedly pulling information from memory strengthens your understanding far more than passive review.

How can I implement that?

• Ask for questions, not answers: “Quiz me on the three main causes of the French Revolution.”

• Define how many questions and ensure focus: “Ask me 10 questions, one at a time, only proceeding when I answer each one.”

• Set the difficulty and scaffolding: “Make the questions progressively harder.”

• Add subject-specific requests: “Ask me to connect causes to specific historical events.”, “Ask me to explain each vocabulary item in the target language.”, “Ask me to justify which formula I used.”

• Follow up before giving up: If you get stuck, ask for a hint. If the retrieval isn’t very effortful, ask to increase it. This keeps you close to the optimal challenge point.

Evidence: Practicing retrieval can boost retention by 10–20% compared to re-reading, according to research by Karpicke and Roediger.

3. Generation effect

Principle: Generating your own explanations, even imperfect ones, leads to deeper understanding and better retention.

How can I implement that?

• Explain in your own words: After learning, rephrase the idea yourself and ask, “Here’s my explanation in my own words. Did I miss anything?”

• Write your own questions: Think of what you’re unsure about and prompt, “Here are three questions about the French Revolution. I’ll give you my attempts at answering each of them one by one. Can you provide feedback?”

• Summarize from memory: Close your notes, summarize what you remember, and ask, “This is what I know about the water cycle. Did I get it right?”

• Teach the topic to the AI: Walk through the material as if teaching and ask, “If I taught you about mitosis, what mistakes or missing parts would you point out?”

Evidence: Studies show that generating content (such as summaries, analogies or examples) improves both comprehension and recall by 20-40%.

Play the AI game

Implementing knowledge transfer, retrieval practice and the generation effect will ensure your learning is maximized. However, LLMs weren’t designed for the core use of education. Here are four simple AI-specific strategies to steer your LLMs towards an outcome that radically increases your learning.

Reduce sycophantic behavior

Even though recent versions like GPT-5 have improved a lot in this regard, you still need to ensure that your LLM is giving you the friction you need. Add this explicitly to your prompt:

Don’t praise me just for asking questions or simply attempting to answer. If my answers are too simple or not developed enough, don’t praise me for them. Rather, ask me to try harder. Specifically tell me where my answers need improvement, where my arguments are weak and where I’m missing evidence. Give me direct criticism without softening your feedback.

Role prompting

By now, you've probably heard that assigning a role and personality to your LLM improves the quality of its outcomes. This is because it adds consistency to the LLM's predictive text power in terms of knowledge, tone of voice and reactivity. You can also use this technique to reduce sycophantic behavior. Here’s a sample prompt:

You are a strict history professor with extensive experience evaluating advanced student writing. You expect critical thinking, depth and originality, not simply recitation or superficial engagement. You don’t give credit for merely asking questions.

Context engineering

Providing context has a massive effect in the quality of the interaction with your LLM. Many argue that context engineering overshadows prompt engineering as a core AI skill. There are many ways to do this: from uploading PDFs or pasting an article on the topic to following this simple interaction:

Request context on the topic from me until you feel confident enough to start a study session on it with me.

Here’s an additional tip: providing links doesn’t always do the trick well, due to reliance on web fetching features. Copying and pasting the content of a website onto the chat works best.

Personalize learning

Don’t forget to tell your LLM what kind of learner you are, your background, and your knowledge level. Study modes might ask up front, but their personalization is usually limited. Don’t hesitate to remind the LLM regularly to adapt whatever reply it gives you to your level, background and current level of knowledge.

Putting it all together: A sample session

Imagine you’re studying cell biology. Here’s how a research-backed AI session might look:

1. Set the stage: Here’s where you want to reduce sycophancy, try role prompting, provide context and introduce yourself. Set a goal: “I want to be able to explain the five stages of mitosis without my notes.”

2. Retrieval practice: A great way to start a lesson is by recalling what you know on the topic. Prompt the AI: “Ask me 5 questions on the stages of mitosis. Start simple, then make it harder. For example, ask how errors in each stage might affect the cell.”

3. Generation effect: After answering, write your own summary and paste it to the AI: “Here’s my explanation. Please check it and suggest improvements.”

4. Knowledge transfer: Now make things messy on purpose. Create a struggle that you need to overcome. Ask the AI for a scenario: “Describe a situation in which mitosis goes wrong, and ask me to explain the consequences in terms of the stages we discussed.”

5. Review and reflect: At the end, ask yourself, “Did I recall and apply these concepts? What was hard? What should I revisit?”

This is just a sample lesson, you can adapt it depending on your case. For instance, if you struggle with understanding, starting with knowledge transfer (analogies, examples or problems) also helps you grasp concepts fast.

Beyond the AI: integration and adaptation

• Mix tools: Combine AI prompts with flashcards (e.g., Anki), handwritten notes, and practice problems for a blended learning system.

• Track progress: Use a spreadsheet or journal to note what you’ve learned, what’s challenging and how you’ve adapted your approach. It’s easy to lose track inside a chat.

• Space your reviews: Ask the AI, “Based on our session, what are the most important concepts for me to review?” Then, set reminders to revisit those topics at increasing intervals.

• Privacy and copyright: Be mindful of the data you share with AI tools and check their policies before diving in.

It’s about ownership!

Use these strategies in one of your next study sessions. You can download the visuals in this article to keep a cheat sheet. Save and bookmark this article for future reference. Track your progress. Adapt as you go. Let me know in the comments if you’d like more sample sessions or AI strategies that are more tailored to specific subjects.

And remember: AI can be transformative, but only if you take charge. Intentional, reflective learners will gain the most. Use science-backed methods, guide the AI deliberately and make learning your choice.

Share

Picture this: you meet someone, exchange banter, then hit the “how old do you think I am?” trap. Social dread kicks in. You throw out a too-kind number. Of course, you’re wrong. They tell you their real age. You just lost a new friend. Or you never had them to begin with.

Weeks later, you realize that you don't know the exact age of many of your friends. However, you perfectly remember the age of the friend who "could have been." According to science, there’s a reason for this: it’s because you had to guess it first.

Subscribe now

What is the generation effect?

The generation effect is the principle that you remember ideas better when you come up with them yourself rather than having them handed to you. Guessing, predicting, recalling, speaking or even running an inner monologue all do the trick.

People often think of “generative AI” when they hear “generate.” But the truth is: our brains invented this centuries ago. Want to learn faster? Start generating!

The generation effect in learning is one of the most reliable findings in cognitive science. This memory boost you get from producing information yourself (instead of passively receiving it) was first identified in the late 1970s and has been confirmed in dozens of contexts since. It explains why taking a stab at the answer, even if you’re wrong, often makes it stick.

Experts also call this a “desirable difficulty.” It’s related to retrieval practice but goes deeper. Generating boosts focus, forges stronger connections and layers new knowledge onto what’s already rattling around in your head. It has a strong creative element.

How the generation effect works

Here’s how it plays out. When you generate an answer, your brain cycles through many different steps and considers different ”factors”. These factors include focusing on the answer and relating it to other similar answers (for example discarding the answers that don’t fit). You also relate the information to the context in which you need it.

If you’re learning French and a friend hands you coffee, you grab for “merci,” but you pause and search your memory before you finally say it. You link it to similar words (s’il vous plaît, bonjour, gracias, Danke) and the situation, so next time you’re in that context, merci comes to mind more easily.

These factors (item-specific processing, relational processing and contextual processing) don’t just check what you know, they strengthened your knowledge through focused attention, rich connections and a match between learning and recall situations. This is explained by the three-factor theory.

When the generation effect works and when it doesn’t

The generation effect works best when:

• The challenge is moderate: tasks are effortful, but achievable.

• Timely feedback is provided: you receive confirmation or correction.

• Questions invite explanation, prediction or inference, not just facts.

It’s weaker when:

• You have zero prior knowledge (pure guesswork with no foundation).

• There’s no feedback, especially for complex material.

• The task is purely perceptual, with no conceptual connections.

Without feedback, there’s also a risk of false confidence: you risk thinking you’ve learned something because you generated it, but it may be wrong. Oddly enough, feedback that’s slightly delayed sometimes does the trick better. It gives your brain a moment to stew, then the real answer lands for good.

Without prior knowledge or context, information can stick if it's impressive or unexpected, but ultimately, it's not an efficient strategy. Especially for beginners, it’s important to balance generative strategies with providing information (direct instruction). This explains why you don't remember all the answers to every Trivial Pursuit question you've ever been asked, though you probably remember some.

Even wrong answers help

One of the most surprising findings about the generation effect is that you don’t need to be right for it to work. In fact, producing a wrong answer, as long as it’s corrected, can strengthen the memory of the correct information more than if you’d simply been told the right answer from the start.

In a nutshell, the idea is that you remember better a correction than information you can't relate to. A correction is always more relatable to you, because it's based on something you produced.

And this is a very concrete reason why you should totally overcome your fear of making mistakes. Consider that there is a more radical version of this technique that also works, according to science: the derring effect, making mistakes on purpose.

The “pause and guess” approach

The generation effect can be implemented in many different ways. The most traditional way is when you “explain it in your own words”, which means extracting the ideas and generating a your own explanation. But you can also use it before getting the information: pause strategically as you read and try to guess what comes next.

You can apply this in any domain: guess words before checking, solve problems before seeing examples, predict the next recipe step or estimate a bill. While “pause and guess” may feel awkward, it usually takes less than a minute and helps you encode information much more deeply, saving a lot of time later.

Other ways to use the generation effect: active note-taking, Socratic dialogue, question generation, concept mapping or summarizing. Many advanced learners implement the generation effect naturally by internally questioning and discussing ideas as they read. This is possible because they are used to inner monologue and generate information to themselves.

Practicing the generation effect with AI

The best uses of AI for education create or support desirable difficulties and generation is a prime example! AI can be a great generation partner:

• Ask it to quiz you without revealing answers. You can use role prompting for this to improve the accuracy of the replies: "Act as an expert in XYZ. Quiz me on the topic ABC".

• Generate your own answer first, then request feedback. You can prompt it as simply as: “Consider this statement and provide feedback: XYZ”

• Have it increase difficulty as you improve: "Make questions progressively harder."

• Use "curveball prompts": Ask AI to give you a riddle, a “what happens if…” scenario or an odd but plausible analogy for a topic you’re studying, then try to resolve or complete it before the reveal.

The AI will give you feedback afterwards anyway, so don't be afraid to make mistakes (although, always double-check that it's not hallucinating!). This will help you remember the information much better than simply asking direct questions.

The best generative engine you have is your mind

That person’s age stuck with you because you guessed it first. The generation effect shows us that if you're actively involved in the production of information, you learn best. Who would have said that the best thing you can learn from an AI is to actually do what the AI itself does best? The generation of the information, rather than the information alone, is a technique that boosts your learning.

It also shows us in a concrete, science-backed way how mistakes are not the enemy of learning. Rather, it's the fear of making mistakes who's the true enemy. That fear leads to inaction and passivity, which truly hinder learning.

Next time you’re faced with a question, resist the urge to peek, take a guess. Use the world’s best generative engine: your own mind.

Share

Nerd out with AI

Prompt suggestions. Always ask follow-up questions

• Act as an expert in [subject]. Give me a series of questions where I must guess the answer first. Wait for my response, then provide detailed feedback, including why my answer was right or wrong. Make each question slightly harder than the last.

• Describe a scenario, experiment, or problem in [topic]. Ask me to predict the outcome or next step before revealing it. After I answer, explain the correct result and the reasoning behind it.

• Give me a tricky or unusual example related to [topic], like an analogy, ‘what if’ scenario or riddle. Have me generate a solution or explanation first, then provide feedback, highlighting connections to prior knowledge.

Links

• 📑 The generation effect: delineation of a phenomenon: Full text of the research paper that introduced the concept of the generation effect in the 1970s. Well worth reading even today.

• 📊 Theories of the generation effect and the impact of generation constraint: A meta-analytic review: This article summarizes the research up to 2020 of the generation effect. I recommend reading the short section on “Context memory theories” explaining how associative and contextual strengthening of memory works.

Related articles

When calculators were introduced, people said the same thing: students would stop thinking.

You've heard the analogy a thousand times. Just as calculators freed us from arithmetic, Artificial Intelligence will free us from the tedious parts of thinking, such as summarizing, proofreading and outlining. Supposedly, the important stuff will stay untouched: strategic thinking, decision-making, metacognition…

Early in the boom of Large Language Models, the analogy quickly caught on. But does it hold up? Can we really divide thinking into distinct layers, with the tedious tasks at the bottom and meaningful tasks at the top? What do we lose when we skip the “unglamorous” parts of thinking? It’s time to revisit the analogy.

What’s really at stake here is the idea of cognitive offloading.

Subscribe now

Cognitive offloading: what the analogy gets right

Cognitive offloading is the use of a tool or technique to reduce the mental effort required for a task that would otherwise rely on memory or internal reasoning. The concept emerged in 2016, well before LLMs became widely available.

Arguably, the most iconic argument against cognitive offloading was made by Socrates, in one of Plato’s dialogues. Socrates claimed that writing would make people forgetful, because they would no longer recall things by themselves. This reasoning sounds almost ridiculous today, but it echoes in modern fears around AI.

Millennia later, calculators raised similar concerns. They take over arithmetic, reducing cognitive load and allowing students to focus on problem-solving. This shift helped math and science education move away from rote drills and toward conceptual understanding.

As with writing, education systems didn’t merely adopt calculators as new tools; they rewrote their curricula around them. LLMs offer the same promise beyond numbers. While calculators took over the numbers, LLMs are coming for everything else.

Why the analogy breaks down

Here’s where the analogy stops working.

• Calculators don't hallucinate. LLMs generate text based on statistical patterns, not truth. When they happen to be accurate, it’s usually because their training data was. However, they're built to sound plausible, not to be right. Recent research (published this week) suggests that hallucinations are here to stay. Calculators, by contrast, are reliable. They do one thing and do it right.

• Skill erosion is harder to spot with LLMs. When you forget how to do long division, you know. However, losing your ability to write or reason clearly is more difficult to detect, especially if LLMs make you feel like you're still capable. This is the paradox of outsourced thinking. As long as the output looks good, you assume the skill is intact. Until you have to perform without an LLM.

• The boundaries are blurry. When using an LLM, you might think you're just getting help with phrasing, but you've probably handed over the structure, tone and even the authorship of the ideas.

Also, LLMs don’t have a fixed domain. In theory, there are no limits to the kinds of cognitive tasks you can delegate to a LLM. And that makes accountability slippery. This is true in classrooms, at work, even in your own mind. Who’s thinking? Who’s learning? With calculators, this wasn’t a question.

And that’s just the cognitive side. I haven't touched on copyright, environmental costs or the messy job market. I won’t dig into those, but they’re hardly background noise and let’s not pretend any of that looks like the calculator era either.

Cultural shift towards higher-order thinking?

Zooming out reveals something more fundamental. Tools not only change how we work, but also what we admire as a society. Calculators shifted the value placed on arithmetic. AI is shifting the value of many thinking tasks we used to consider signs of mastery and intelligence: writing, explaining, analyzing and remembering (though memory has been losing ground for decades).

If a machine can do it, we start to wonder whether it’s worth learning at all. This is not just a technical shift. It’s cultural. The idea of cognitive effort itself is being re-priced. And faster than we’re ready for. Optimists argue that this is the whole point: let LLMs handle the grunt work. Humans can then focus on strategic thinking or decision-making.

All that hinges on a fragile idea: that thinking tasks come in neat layers. That you can automate the base and keep the top. That the pyramid won’t collapse.

The counterpoint: the hierarchy of thinking skills is a myth

Higher-order thinking needs lower-order skills. The notion that thinking is organized in a neat hierarchy (often depicted as a pyramid) is unfounded. Experts have pointed this out for years.

Knowledge of facts and memory are necessary for critical thinking and creativity. They are intertwined. As Carl Hendrick so elegantly puts it: “You can’t connect the dots if you don’t have any dots”. You can’t analyze or be strategic about something you don’t understand. You can’t reflect on thoughts that aren’t grounded in factual knowledge. If automation erodes those, the rest falters too.

This is why teaching facts, practicing recall and making learning purposefully difficult still matter. Especially now. A student who relies on AI for summaries or arguments won’t develop the mental framework needed to critique ideas or apply them in new contexts.

LLMs can help build these skills too, but their default use as chatbots doesn’t encourage it. The most beneficial uses of AI for learning are ones that create desirable difficulties: the productive cognitive struggles that make learning stick. These difficulties are also what leads to higher-order thinking in the first place.

When the tool becomes the skill

Although “higher-order thinking” isn’t necessarily winning the AI cognitive wars (as the calculator analogy implies), another type of knowledge is clearly gaining ground. Mastering LLMs is becoming a valuable skillset in itself. Whether it’s prompt engineering, context engineering, agentic workflow design or GPT-building, the ability to steer LLMs toward precise or scalable outcomes is already an ever-growing body of knowledge separating amateurs from experts.

Yes, calculators also required basic competence. But it stopped there. There was no ecosystem of advanced calculator design for the general public, no body of theory, no careers built on fine-tuning them beyond those of calculator builders. The knowledge forming around LLMs is different in scale, complexity and cultural significance. This isn’t just hype.

In education, this raises new questions about what Generative AI literacy should look like. I've written about the paradox of openness and the need for scaffolding in AI-driven learning design. Recent research also suggests that advanced students may actually benefit from using passive AI chatbots, because it fosters higher metacognitive awareness. The way forward likely lies in a careful mix of structure and autonomy: teaching students not just with AI, but about how to master it.

The end of the analogy is just the beginning

The AI-calculator analogy is comforting, but that comfort hides its flaws.

Calculators changed how we approach math. AI is taking it further by reaching into how we communicate, the professional skills in demand and the kinds of cognitive effort we still consider meaningful.

This isn’t just a shift in work and education. It’s a shift in what we value as thinking. The real question isn’t what we think with AI.

It’s about what we now call thinking at all.

Share

Keep learning

Nerd out with AI

Prompt suggestions. Always ask follow-up questions

• Everyone compares the boom of AI to the time when calculators came out. Can you break down where that analogy holds and where it completely falls apart, especially when it comes to learning?

• I want to avoid skill erosion when using AI tools. Can you help me design a workflow where I still do the hard thinking myself, but use the AI in a way that supports learning?

• Act as a teacher and test me using retrieval practice on “cognitive offloading and the AI-calculator analogy.” Ask me 6 questions, one at a time, only proceeding when I answer. Make them progressively harder.

Links

• Cognitive offloading: This article from 2016 is credited with introducing the concept of “cognitive offloading,” even if no one at the time anticipated the rise of LLMs. The idea has gained new urgency, because it captures what’s quietly happening when we let AI handle the very tasks that shape how we learn, reason and remember.

• The most important memory is still the one inside your head: In this excellent piece, Prof. Dr. Carl Hendrick unpacks the consequences of AI-driven skill erosion, with a sharp focus on memory. Even though his outlook is pessimistic, his argument is compelling: it is detailed, grounded and clear-eyed about the cognitive risks AI introduces.

Related articles

If I asked you, what's the best predictor of academic success, what would you say?

Most people would point to IQ. Or maybe how hard someone studies or how much time they put into preparing for exams.

But research shows something more surprising: one of the strongest predictors of academic success, often more so than raw intelligence or effort alone, is metacognition.

Subscribe now

What is metacognition?

Metacognition is the ability to think about and manage your own thinking. It’s how you reflect on what you know, how you’re learning it and what you should do next. It’s thinking about thinking or learning about learning.

To make the idea clearer: I have a shopping bag full of other shopping bags at home, and I bet you have one too. I jokingly call it the "meta-bag", a bag that holds other bags or a "bag of bags". Similarly, metacognition is thinking that holds thinking. Cognition about cognition. The brain stepping back to look at what the brain is doing.

It involves being aware of your cognitive processes, recognizing when you understand something and, critically, when you don’t. It also includes knowing how to adjust your strategies accordingly. In short, metacognition turns learning from a passive experience into an intentional one.

Meta... what?

Metacognition sounds abstract, but it's deeply practical. And it’s probably the thing you’re not doing when you say, “I studied for hours and still didn’t get it.”

Let’s debunk a few myths:

• Myth 1: “I know how I learn best.”
  Usually false. Without conscious reflection and experimentation, people tend to repeat what feels comfortable, not what works. We also tend to gravitate towards non-scientific ideas such as learning styles.

• Myth 2: “If it feels easy, I must be getting it.”
  Nope. That’s the illusion of fluency again. Real learning often feels clumsy and effortful.

• Myth 3: “Some people are just naturally good learners.”
  There’s no such thing as an effortless genius. What looks like born talent is often the result of metacognitive habits that have been so skillfully mastered that they appear innate. But they were built, not born.

The process of learning is a deeply deceptive experience. How it feels often has little to do with how well it works: learning often feels productive without being productive. But once you start seeing learning as a system you can monitor and steer, not just endure, you unlock a different level of cognitive control.

Let's break it down

The cognitive scientist J. H. Flavell introduced the now-classic distinction of metacognition into three overlapping types:

1. Metacognitive knowledge: What you know about your learning. For example: “I learn better when I quiz myself,” or “Math problems take more effort to stick.”

2. Metacognitive monitoring: What you notice while you’re learning. “This part is confusing,” or “I think I remember this, but not well.”

3. Metacognitive control: What you do with the insights gained from monitoring: changing tactics, skipping ahead or slowing down to focus where it’s needed.

Each of these metacognitive types reinforces the others, and studies suggest that while control often has the most visible impact on performance, it’s hard to develop without first building knowledge and monitoring skills.

I know what you’re thinking: all of this sounds obvious, doesn’t it? Check that you’re understanding, monitoring your study… It’s pretty commonsensical in theory. But in practice, you probably skip this entirely as most students do according to research. You probably treat learning like a checklist: read chapter, highlight text, solve problems, repeat. When pressed for time, you’re likely to skip reflecting on your learning altogether. And that’s when your brain tricks you again into unconsciously mistaking activity for progress and confusing effortlessness with understanding.

Why we're bad at metacognition

Like other signs of mastery, such as knowledge transfer, metacognition is rare. Beyond the fact that we tend to avoid it simply because it takes time and effort, there are a couple of additional and arguably more interesting reasons.

The judge and the judged

We often find it awkward to ask ourselves questions and answer them honestly. It can feel like performing on an empty stage, with only our own applause as a response: shouldn't the judge and the person being judged be two different people? We're so conditioned to seeking external validation, such as grades, feedback and exam results, that trusting our own judgement can seem like cheating.

However, this is a consequence of years of learning focused on performance (exams, presentations, etc.) rather than long-term growth. Learning to question ourselves isn’t just about study habits. It's about developing the kind of honesty and self-confidence that leads to real success, not just when we're studying.

The unreliable narrator

We also tend to be unreliable narrators of our own learning. We tell ourselves neat stories about what worked, when in reality we’re often mostly guessing. The more we master a skill, the harder it is to remember how we got there, and the more likely we are to craft a beautiful story that hides the reality of an often messy, effortful process. Without reflection, we end up trusting that superficial story instead of the actual process.

How to get better at metacognition

So what can we do to develop this elusive skill? The answer isn’t a one-off magic trick. It’s a set of habits, reflections, and small shifts in perspective that compound over time. Let's get into it.

Practical techniques

Let’s start with the practical stuff. Here are three simple techniques you can implement in your learning routines to start developing metacognition.

• Ask yourself questions. Not rhetorical ones, real prompts like: “What do I already know?” or “How will I know if I’ve understood this?” These small check-ins build metacognitive awareness.

• Delay your feedback. Come back later and try to recall or explain what you learned without notes. Struggle shows you where the gaps are.

• Use retrieval, not review. Retrieval practice is one of the most efficient study techniques out there and it fosters metacognition by the very nature of how it works: it makes you aware of what you don't know.

Habits and routines

An effective way to develop metacognition is to turn reflection into a habit. You can do this at different times: through structured post-task routines, short reflections after study sessions or daily check-ins.

This isn’t busywork, it’s how experts learn from the process, not just the outcome.

Narrative shift

Finally, change your identity. Stop being just the learner and become your own coach. The goal isn’t to feel smart while studying. It’s to improve your learning gradually through reflection. And that means learning to sit with discomfort, normalizing your internal dialogue (talking to yourself and asking yourself questions) and being at ease with regularly dealing with your own gaps or failures, as this is the only path to improvement.

The good news? Once you start thinking about how you think, it gets easier to notice when you’re fooling yourself. And that’s when real learning begins.

It comes down to honesty

Our brains don't come with preset glowing progress bars for learning. Metacognition is how we bridge that gap. It won’t make learning feel easier. But it will make it more honest, and ultimately, more effective. Once you know how to check in with yourself (not just coast on feelings) you stop mistaking comfort for progress. You stop narrating your learning like a fable, and start living it like a practice.

In a sense, metacognition is a concrete way of embodying the ideal that the ancient Greeks fervently advocated: ‘Know thyself’ (γνῶθι σεαυτόν), or as Socrates somewhat radically put it: an unexamined life is not worth living. Become the master of your own thoughts and learn to regulate them to achieve true freedom and mastery.

Share

Keep learning

Prompt suggestions. Always ask follow-up questions

• I want to improve my metacognition while studying. Can you help me build a short routine with reflection questions before, during, and after a study session?

• Act as a learning coach. I often feel like I understand something, but later realize I don’t. How can I tell the difference between actual understanding and the illusion of fluency?

• Act as a teacher and test me using retrieval practice on “metacognition and its role in learning.” Ask me 6 questions, one at a time, only moving on when I answer. Make them progressively harder.

Links

• 📑 Metacognition and cognitive monitoring: A new area of cognitive–developmental inquiry: This is a classic article on the topic of metacognition, widely understood to have laid groundwork for the field and introduced core distinctions in metacognitive processes. It’s short and easy to read.

• 📄 Metacognition by Cambridge Assessment: Here’s a simple yet detailed explainer PDF on the concept of metacognition by Cambridge International Education. It includes an extensive list of literature for further reading. It's a great read to follow on from this article.

Related articles

Everyone’s got a study buddy now: ChatGPT, Claude, Gemini… even your calendar app pretends to be an AI coach now! But if you’re serious about learning, there’s a question that matters more than the next app adding ‘AI-powered’ to its tagline: which one actually knows how to teach?

Last week, I talked about three paradoxes of AI in education: those structural challenges that show up when we use AI to learn. Every model handles these paradoxes in its own way. But which one actually does it better?

A new benchmark has finally given us a real answer. And no, it’s not based on vibes, popularity or yet another LinkedIn post claiming to have finally found “the” model.

It’s based on pedagogy.

The first benchmark that actually cares about learning

Most AI benchmarks test content knowledge. They’re good at asking: does this model know biology? Or law? Or whether Napoleon was short? What they don’t ask is: can this model teach any of that stuff well?

Subscribe now

A new study, by Lelièvre et al. published only some weeks ago, changes that. It introduces the Pedagogy Benchmark, a dataset made up of real teacher certification questions from Chile’s Ministry of Education. These aren’t trivia questions, they assess teaching strategies, classroom management, assessment design and even special education knowledge.

There are two parts:

• CDPK (Cross-Domain Pedagogical Knowledge): 920 multiple-choice questions focusing on learning theories, instructional design and teaching practices for learners of different ages.

• SEND (Special Education Needs and Disability): 223 questions focused on inclusive education and specific pedagogy for learners with special needs.

Want to know the kind of questions AI had to face? Here’s the general idea. Some questions described a task and then asked “Which performance of the student would demonstrate the achievement of the learning objective?”, then giving four multiple-choice options. Some others described a learning objective and asked “What activities are most appropriate to evaluate it in Year 2?”, etc.

This isn’t only about facts. It’s about understanding what learning looks like. And it’s a lot harder to fake!

Who topped the leaderboard?

The benchmark tested 97 large language models. Some were big and shiny. Some were scrappy and efficient. Here’s the top 10:

You’ll find more models on the leaderboard than in the study, as the benchmark is regularly applied to newly released ones! The leaderboard also includes a benchmark on content knowledge.

But the results are clear: Gemini, ChatGPT and Claude top the list, with Google’s Gemini-2.5 Pro taking the first place! You might be wondering: how good is 88.77%, really? Consider that the reference of an average teacher’s accuracy is 50% based on past national exam scores.

Not to gloat over anyone's misfortunes, but here I drop that at the bottom of the ranking is Meta’s AI Llama-3.2 1B, which only scored 28.03%. In their defense, the newer version Llama-4 Maverick performed much better at 81.65%. Somewhat disappointing is Perplexity’s Sonar at 70.75%, not even making it to the top 25.

Also note: when it came to Special Educational Needs and Disabilities (SEND), performance dropped. Even top models did worse here than on general pedagogy, dropping by an average of 3%. If we want AI to support all learners, this gap needs to close.

OpenAI’s o3 vs GPT-4o: a cautionary tale

One of the most unexpected results? OpenAI’s o3 outperformed GPT-4o by nearly 10 points. Interestingly enough, o4-mini also outperformed GPT-4o! That’s not just surprising… it’s a bit embarrassing for the flagship model.

• 87.88% – o3 (OpenAI)

• 81.98% – o4-mini (OpenAI)

• 78.31% – GPT-4o (OpenAI)

You’d expect GPT-4o to be better. It’s newer. It’s faster. It’s got the marketing push. But o3 (an older, slower model) was smarter when it came to pedagogy. The authors don’t spell out why, but they hint at the real culprit: reasoning. o3 had better fine-tuning for chain-of-thought and instruction-following tasks, which matter a lot when you’re asked to identify learning strategies, not just spew facts.

Let this be a lesson: newer ≠ more pedagogically competent. And definitely faster ≠ better. Here we can clearly see the consequences of the paradox of efficiency.

Reasoning beats raw power

Another trend stood out: the best models were all “thinking models”, fine-tuned for step-by-step reasoning. Even though the benchmark used multiple-choice questions, those that encouraged internal reflection (a.k.a. “think before you guess”) scored higher.

This is exactly the paradox of outsourced thinking in full action. In a world that rewards hot takes and instant answers, it’s refreshing (and kind of poetic) that the better teachers were the ones who took a moment to think.

What this means for learners

So, if you’re using AI to learn, what should you take away?

• Pick tools that know how to teach, not just what to say. A model that helps you discover the answer beats one that blurts it out in the most convincing way. This new benchmark is gold when it comes to picking your AI study buddy!

• Some AIs are better teachers in specific areas. The dataset includes questions per subject. While Gemini-2.5 Pro beats everyone in Maths, Science and Technology, o3 actually tops the Social studies leaderboard! You can use the filter per subject on the leaderboard (top left corner).

• Performance isn’t just about model size. Don’t fall into the trap of “it’s newer and more expensive, therefore it must be better”. Smaller, smarter and cheaper might be better for your needs.

Don’t forget that, at the end of the day, you also need to do your part! No matter what AI you use, if it’s not making you think, then you know it’s not doing the job. You want an AI that gives you desirable difficulties, the productive struggle that makes learning stick!

What this means for educators and edtech teams

If you’re building tools, not just using them, here are 4 takeaways:

• Open-source is rising. Deepseek R1 and the Qwen series (from Alibaba) delivered strong performance at a fraction of the cost. You don’t need a $100 million model to build a smart tutor.

• Cost vs performance matters. The benchmark included a full value frontier showing that models like Qwen 8B get you >70% accuracy at $0.035 per million tokens. That’s wild. And budget-friendly. You’ll find a breakdown graph of accuracy vs cost on the leaderboard (scroll down).

• On-device models are catching up. Gemma 3n E4B (Google) and LFM-7B (Liquid AI) are optimized for phones and laptops and scored >60%. That means local, offline AI tutors may be here sooner than you think. This is relevant regarding user privacy, as these models don’t send the user’s information to any servers.

• There’s no way around fact-checking. Even if the study shows promising results, every edtech professional should be worried when reading that no model managed to avoid 10% of its answers being complete nonsense. Sadly, the core issue might be bigger than we think.

Warning: the benchmark isn’t about practice

There’s a small “but”. The authors of the benchmark openly acknowledge that it’s based on pedagogical knowledge, not on pedagogical practice. A properly scientific benchmark would require experts set up experiments and assess the actual outcomes of learners using every AI model. This benchmark, however, is built on what every AI model “claims” about pedagogy in multiple choice questions, which certainly isn’t the same as assessing the actual practice of an AI teaching students.

The authors explain that such a setting wouldn’t be scalable, as it would be too difficult to replicate and update for every new AI model. This is indeed relevant when it comes to tackling the paradox of openness, which can hardly be observed by asking direct questions to the AI models.

In any case, the questions used are very praxis-oriented (as we saw) and the benchmark certainly provides a solid guideline, grounded in actual pedagogy. The value it provides by giving us a dynamic, ever growing leaderboard is certainly worth the compromise.

Choose wisely

If you’re wondering whether I paused mid-article to re-rank my own study tools… obviously! This benchmark gives us more than a scoreboard that will get progressively updated. It gives us a compass. I’m not saying I printed the benchmark leaderboard and taped it to my wall, but… I’m also not not saying it!

Share

Keep learning

Prompt suggestions. Always ask follow-up questions

• I want to pick the best AI model to help me study. Based on the pedagogy benchmark, which one should I use for my specific subject or learning goal?

• Explain why older or smaller models like o3 or DeepseekR1 might be better for learning than newer, more powerful ones. How should that change my choice of study buddy?

• Act as a tutor and test me on the key findings of the pedagogy benchmark study for AI models. Ask me 6 questions, one at a time, only continuing when I answer. Make them progressively harder.

Links

• 📑 Benchmarking the Pedagogical Knowledge of Large Language Models: Here’s the link to the study I’ve discussed in this article. You can access it on “View PDF”. The last version of the study, at the moment of writing, is from July 1st 2025. The article is shared under a cc-by license.

• 📊 The Pedagogy Benchmark leaderboard: This is a good website to bookmark on your browser and keep coming back to whenever a new AI model comes out!

Related articles

"It was the best of times, it was the worst of times." Charles Dickens wasn’t talking about AI, but he might as well have been.

AI is already changing education. According to OpenAI, college students use ChatGPT more than any other user group. Some use it as a study assistant, others to generate flashcards or rewrite essays. Edtech leaders talk about boosts in “efficiency” and “personalization.” The possibilities are endless. It’s hard to not see this as the biggest revolution in educational thinking in the last 100 years (maybe in thinking, full stop). But something doesn’t add up.

A recent MIT study found that students who relied on AI scored worse on critical thinking. The Flynn Effect, or the trend that describes how average IQ has been increasing globally across the last century, seems to have peaked and started reversing in several developed countries. AI feels like a gift to education, but it often works against how learning actually happens. Not because it can’t support learning. Because learning isn’t its core design.

Barbara Oakley wrote in a recent study about the “memory paradox”: by offloading our knowledge onto tools such as AI chatbots, we are actually preventing long-term memory formation and, with it, decreasing mental flexibility. Carl Hendrick’s comment on it made a sharp point: memory is still the backbone of learning.

I found the framing around “paradoxes” to be quite on point at capturing something essential about how AI is used in education: well-meaning tools often produce contradictory outcomes. I’d like to unpack three additional paradoxes I’ve observed and what we can do about them.

Subscribe now

1. The paradox of outsourced thinking

The MIT study found that students using AI to write essays couldn’t recall what “they” had written. What happened here is the AI version of the illusion of fluency: when a chatbot hands you the answer, it feels like your brain did something. It didn’t. When you have to solve that problem later, you realize that you're actually dependent on the AI and that you didn’t really learn.

Even retrieval practice, one of the most effective learning techniques, loses its power if you let AI do the recalling. However, when students do use AI to help them retrieve (not the other way around), they learn more. Because your doing the retrieval was the whole point all along!

This raises a few uncomfortable truths:

• Getting to an answer isn’t learning.

• Completing a task isn’t learning.

• Performance isn’t learning.

Real learning is about overcoming the right kinds of struggles that make us think, stay cognitively engaged and activate deep processing. If done properly, that thinking gets encoded and reinforced as long-term memory. This is what cognitive scientists call desirable difficulties.

Paradox of outsourced thinking: by delegating thinking, you're also undermining learning.

2. The paradox of openness

The openness, flexibility and versatility of AI chatbots is one of their powers: they can solve virtually any issue or provide support on any kind of problem. However, that very openness becomes a problem when it comes to learning.

Proper learning needs scaffolding: presenting simpler knowledge first, offering a warm-up problem or layering concepts before introducing complexity. It’s by adjusting the scaffolding that we can get to the optimal challenge point. Structure is fundamental to great learning experiences. But chatbots don’t scaffold by default. They don’t pace unless asked to.

Certainly, this can be improved with good prompt engineering or by providing context to the AI. And motivated learners do exactly that. However, a study from last year showed that learners with less experience using AI struggle more with the implementation of AI chatbots than those who were proficient. The study also showed that the struggle is considerably reduced if the chatbot offers scaffolded or guided interactions.

Expecting every learner to proactively shape their own learning experience and master prompting to access quality learning isn’t just unrealistic. It misses the point of what kind of user most learners are. Learners should benefit from any learning tool without needing to master the art of tweaking it. Otherwise, we’re creating accessibility barriers without realizing it. Students with lower metacognitive abilities and technical skills are being left out. And these students are precisely the ones who need more support.

Paradox of openness: radical flexibility without guidance makes learning harder, not easier.

3. The paradox of efficiency

Efficient learning is the objective of any learning tool. This means that every learning app, website or material should aim to provide (1) the highest possible quality in (2) the shortest amount of time. When we say that we want to optimize efficiency, we usually mean increasing speed without sacrificing quality.

But when it comes to AI in learning, optimizing for efficiency often leads us to streamline every step in the process to get to the learning outcome faster. Streamlining and optimizing typically mean removing obstacles and blockers. And as we've seen, it's difficulties (of a certain kind) that precisely make learning happen.

If we define efficiency as speed plus quality, and AI already handles speed, then the real challenge is improving the quality to match non-AI learning experiences. Ironically, that means creating more friction and adequate learning challenges, not less. Supporting effort and productive struggle, not eliminating it. So, even though efficient learning is the ultimate goal, focusing on efficiency might be what prevents achieving it.

Paradox of efficiency: by optimizing for efficiency, you might be learning less efficiently.

How to handle these paradoxes

Regarding AI in education, it’s now a commonplace to hear people say, “It depends on how you use it.” But when pressed on what that actually means, those same people struggle to provide a clear, straightforward answer.

Here’s the simple answer:

All good uses of AI in education either create or support desirable difficulties. That is, AI is well used when it gives us learners the kinds of cognitive struggles that keep us mentally engaged.

Use AI to support and shape effort, not skip it. Let it quiz you, space your practice, challenge you, interleave topics or give hints, not answers. Learning happens when you wrestle with ideas, not when you copy them. But not every struggle is useful: AI tools shouldn't offload the responsibility of designing the learning experience onto learners. That’s an undesirable difficulty. Scaffolded, structured learning should be the default.

For AI to have a positive impact on our learning experience, learning tools need to be designed to guide us by default into the cognitive struggles that help us learn, even when we don’t know what to ask (which is the most common starting point!).

As it happens, OpenAI is already working on a "Study with me" mode for ChatGPT and Anthropic is introducing Claude for Education as well. Other tools are experimenting with Socratic dialogue as an alternative to prompt-heavy interaction. And many learning apps are starting to embed AI-driven experiences into structured learning plans, where the plan provides the pacing and scaffolding. So, there’s real movement in the right direction.

The real lesson behind the hype

AI in education isn’t a miracle. It’s a mirror.

It reflects back all the tensions we never resolved: speed vs depth, freedom vs guidance, ease vs effort. The good news? It also gives us a genuine chance to rethink learning from the ground up.

Not by removing difficulty. But by designing it on purpose.

Share

Keep learning

Prompt suggestions. Always ask follow-up questions:

• I want to use AI to study more effectively. Can you help me design a study session where the AI supports desirable difficulties instead of bypassing them?

• Act as a learning coach. I’m worried that using AI is making me too passive. How can I tell if I’m outsourcing too much of my thinking and what should I change?

• Explain in simple terms the paradox of efficiency related to the use of AI in education. Then help me come up with one concrete way to use AI that doesn’t fall into that trap.

• I want to consider multiple perspectives. Find 3 experts with different points of view on the impact AI is having in education and compare their opinions.

You probably grew up thinking mistakes were bad. Deep down, you know they’re somehow part of learning, but if you’re honest with yourself, you know you still try to avoid them.

Turns out, they’re not just useful. They’re one of the most effective ways to learn, especially when you make them on purpose. That old saying about learning from your mistakes just got an upgrade!

New research just confirmed it

This just in: a new study for the September 2025 edition of Contemporary Educational Psychology titled “Learning from errors: deliberate errors enhance learning” backed up earlier findings with stronger evidence. Deliberate mistakes fall under what cognitive scientists call desirable difficulties: methods that feel harder but improve long-term memory.

When you knowingly write a wrong answer, then fix it, your brain notices the conflict. That mental friction sharpens how you store the right answer. The trick is doing it deliberately. You need to be intentional about making mistakes. Mindless ones won’t cut it.

Compared to rereading or trying to get everything right the first time, deliberate errors force you to engage. You monitor your own thinking. You dig deeper. And you remember more.

What did the new study find out?

The 2025 paper didn’t just replicate earlier results. It widened the scope. It compared deliberate error-making against restudy and retrieval practice, which is one of the most evidence-backed learning techniques out there.

The effects were not evident in immediate tests, but they were clear in delayed testing (a week later). Even better, when the errors were close-but-wrong (“semantically confusable”) students got sharper at telling similar concepts apart: no more tripping over terms that sound right but mean something else entirely.

Why does this work? The researchers suggest that fixing an error creates a special kind of memory trace, a unique mental tag that makes the correct version easier to recall later.

But don’t get clever with random guessing. The study underlines that gains only showed up when the mistakes were intentional. If students scribbled something thoughtless or didn’t reflect, the whole thing fell flat.

But here’s the twist: students didn’t trust the method. Even after the deliberate-error group outperformed the rest, they still believed strategies like restudy were more effective. Classic brain betrayal. That mismatch between perceived and actual learning? It's a classic metacognitive illusion.

It’s not a new thing: the derring effect

This isn’t a brand-new idea. Wong and Yap’s 2024 study already showed that deliberate mistakes helped students retain more, stay engaged and transfer their knowledge better.

Some researchers had doubts. Early results were inconsistent. But the new 2025 study cleared that up: the derring effect is a real thing.

The derring effect: When students intentionally make errors and correct them, they retain more, engage more deeply, and transfer knowledge more effectively than with error-free practice.

Traditional studying is like using GPS. You get there, but forget the way. Deliberate mistakes are like getting lost on purpose, then learning the whole map.

How to fail forward on purpose

I know, if you're a perfectionist (and if you're reading this, chances are you are), the idea of making mistakes on purpose sounds like hell, so this technique may feel like cognitive anarchy. Here are some tips on how to proceed:

• Frame it as a game: Tell yourself, “Ok, now I’m going to make a mistake on purpose and then correct it. How can I answer this question incorrectly?”. This will help reduce the cringe idea of “you being wrong” (drama).

• Start small. Try writing one wrong answer per study session. Fix it right away. The correction is the whole point.

• Explain the correction out loud or in writing. What was wrong? Why does the right version work better? If you're stuck, ask an AI to play devil’s advocate.

• Keep a record of your “best mistakes”. Review them. It’s hard to forget something you caught, fixed, and filed under something that only makes sense to you and your prefrontal cortex.

If you name that record “Chronicles of Regret (Educational Edition),” I won’t judge. Unlike your ex, your mistakes are honest!

Stop fearing the mess

Deliberate mistakes work. Scrubbing them out may keep your notes tidy, but it won’t help your brain. This new study also showed us one of the biggest challenges out there: your own perception of learning. Like the students in the study, you probably doubt the efficacy of making deliberate mistakes.

So here’s the real question: you say you want to learn deeply. Are you willing to feel stupid for five seconds to make that happen?

And remember: it's not about tolerating errors. It's about using them and building around them.

Share

Keep learning

Nerd out with AI

Prompt suggestions. Always ask follow-up questions

• I’m trying to study a complex topic (ask me which one), and I want to use deliberate mistakes to understand it better. I will deliberate make 3 example mistakes, can you help me correct them one by one?

• I find it really uncomfortable to make mistakes on purpose, but I’ve read it’s an efficient study technique. Can you help me reframe that discomfort and suggest how to start small without it feeling chaotic?

• Act as a teacher and test me using retrieval practice on “deliberate mistakes and the derring effect.” Ask me 6 questions, one at a time, only proceeding when I answer each one. Make them progressively harder.

• Act as a learning coach. I want to try the “derring effect” while studying. Can you walk me through a step-by-step routine using deliberate errors for a concept I’m learning?

Movies and TV series love a good 'that's how they learned to be a master' scene. One minute the protagonist is falling all over the place, the next they're slicing arrows mid-air.

Mulan becomes a warrior in a musical number. In Legally Blonde, Elle Woods studies with scented highlighters, flips her hair and becomes a legal powerhouse. The Queen's Gambit's Beth Harmon plays a few games in a basement and suddenly humiliates grandmasters. Neo straight up downloads kung fu into his brain in The Matrix like it’s an app update.

Many of us know the feeling: you understand something perfectly, then fall flat the moment the setting changes. Like remembering a grammar rule from your textbook but forgetting it mid-conversation. Or following a detailed recipe, then freezing when the spoon and the eggs are in your hands.

The real challenge is knowledge transfer, and it explains why we can't just copy and paste knowledge from one setting to another.

What is knowledge transfer?

Knowledge transfer is the ability to apply knowledge to new and unfamiliar situations. It’s one of the most valuable, yet elusive, learning goals according to cognitive science.

There are two kinds: near transfer and far transfer. Near transfer happens when the new situation looks a lot like the learning context: for instance, typing a vocabulary item into the app where you learned it, just in a different exercise. Far transfer means using your skills in situations that feel nothing like where you learned them, like speaking French in a noisy café in Paris when all you’ve practiced is Duolingo.

Far transfer is more demanding and far less likely to occur without intentional design and practice. Not because you don’t care enough. Because your brain encodes knowledge with the original context anchored to it, and detaching it takes effort.

Subscribe now

Your brain is a context junkie

When you learn something, your brain doesn’t just save the main idea. It also stores contextual details: the setting, emotions, instruction style and even irrelevant stuff like background noise.

This dependency on context explains why students may ace exams but struggle to apply the same knowledge in professional or informal settings. Retrieval becomes tied to contextual signals that aren’t always present.

How to support transfer

Here are some science-backed strategies to make you a transfer pro:

• Practice the real thing. Make the learning context resemble the one where you’ll apply it. This helps bridge the most common transfer gap: between learning environment and real-life practice.

• Change up your conditions. Try different ways, sequences and contexts of study, otherwise your brain may fall into the recognition trap.

• Find the principle. After every exercise, ask what the underlying rule is. Don’t just mimic. Extract.

• Then test it somewhere weird. Learn negotiation? Try getting your landlord to fix the heater without raising your rent. Study logic? Apply it to your friend’s dating drama. If it works where you don’t expect it to, it’ll work anywhere.

Transfer signals deep learning

Transfer often feels harder than initial learning, but like other desirable difficulties, that discomfort is a good sign. It means your brain is reorganizing knowledge for real use.

So no, you can’t just download kung fu into your brain, but you can train it to think like a scientist. That’s what the subscribe button is for. Also, share this with your friend who thinks watching TED Talks counts as professional development.

Share

Keep learning

Nerd out with AI

Prompt suggestions. Always ask follow-up questions

• I want to make sure I actually transfer what I’m learning. Can you help me design a study routine or practice plan that builds far transfer into my learning process?

• Why is it that I understand something in theory, but freeze when I try to use it in real life? How can I fix that?

• Act as a teacher and test me using retrieval practice on “knowledge transfer.” Ask me 6 questions, one at a time, only proceeding when I answer each one. Make the questions progressively harder.

Links

• 📑 When and where do we apply what we learn? A taxonomy for far transfer: This landmark study by Barnett and Ceci from 2002 introduced a structured framework describing nine dimensions of transfer (domain, context, time, etc.). It also explains why far transfer is rare and how to design tasks to promote it.

• 📃 Teaching for transfer: This classic article by Perkins and Salomon from 1984 introduced the foundational model to study knowledge transfer. It’s still a pleasant read today.

The cover photo includes one of Yayoi Kusama’s iconic pumpkin sculptures on the island of Naoshima, Japan.

Last week's article was about desirable difficulties: the kinds of challenges that boost your learning (because not every difficulty is desirable). But there’s another key question we haven't tackled yet: not just what kind of difficulties we should aim for, but how difficult they should be (their intensity).

So, how hard is hard enough?

Many of us struggle to adjust the difficulty of our learning tasks. Trying to read a novel in a foreign language before you’ve mastered basic grammar probably isn’t a winning strategy. But I’ve yet to meet anyone who became fluent by sticking to Duolingo alone. Finding the right level of challenge isn’t obvious.

But here’s what most people don’t know.

When scientists analyzed how fast learning happens under specific conditions, they discovered something striking: the golden zone is where you’re right about 85 percent of the time and wrong 15 percent.

In that range, they observed that learning accelerated. If learners drifted into harder territory (more errors) or easier territory (fewer errors), progress slowed. In other words, how often you get stuff right is a decent clue for the optimal difficulty of a task.

Subscribe now

What is the optimal challenge point?

That 85 percent success rate isn’t just a curiosity. It points to a deeper principle known as the optimal challenge point.

The optimal challenge point is a concept from cognitive science which defines the level of task difficulty that leads to the most effective learning. It sits between two extremes: tasks that are too easy, which encourage repetition without growth, and tasks that are too difficult, which lead to disengagement or non-constructive failure.

At the optimal challenge point, learners operate just beyond their current level of mastery. Success is possible, but effortful. Errors are frequent enough to trigger adaptation, but not so frequent that they become discouraging. The result is what researchers call productive struggle, a state where attention, memory and motivation are all engaged.

The term comes from motor learning research, particularly the work of Guadagnoli and Lee (2004), who showed that learning is most effective when task difficulty is a little harder than the learner’s current skill level. While originally applied to physical performance, the same principle holds across domains: from language learning to math, from music to memory.

This framework also echoes a classic learning theory: the zone of proximal development. Lev Vygotsky argued that the area where you can grow further next (or your “zone of proximal development”) is just beyond what you can do without support. The overlap is clear: both models point to a narrow window of optimal challenge, just past your comfort zone.

How to find the sweet spot

There’s no universal formula for hitting the optimal challenge point. But there are signs that you’re in the right zone.

The first guideline you can safely use is the accuracy rate of your answers. The 85% principle is a great reference, as it's backed by scientific evidence. However, consider that it was tested with specific kinds of tasks: direct (yes/no answers), with clear feedback (after each trial) and with incremental progress (such as categorizing or language drills). It might not be 100% applicable to all types of practice.

Prior knowledge is the single biggest factor in determining the right level of difficulty. The more mental models you’ve built, the less working memory a task demands, and the more cognitive space you have to engage with new material.

Other variables matter too. A motivated learner can handle more complexity (like you, am I right?). A well-scaffolded task (i.e., one that starts easy and becomes more difficult step by step) lowers the difficulty without dumbing things down. Even your level of focus or fatigue shifts where the “just hard enough” line falls.

Use it wisely

The idea of the optimal challenge is a qualitative framework backed by empirical data, but unlike other theories in cognitive science (such as cognitive load theory), it’s hard to quantify universally. Think of it as a guiding principle, not a precise formula.

And remember: the kind of difficulty matters too. If the challenge isn’t desirable, there’s no optimal point… just wasted effort. You’ve heard “work smarter, not harder”? Try this instead: work smarter and harder... enough.

By the way, you might not be at the optimal challenge point just now, but you’re definitely at the optimal subscribe point. So, go ahead: it only gets better (harder) from here.

Share

Keep learning

Nerd out with AI

Prompt suggestions. Always ask follow-up questions

• Act as a coach. I want to apply the idea of the optimal challenge point to a skill I'm learning. Can you help me design a practice routine that keeps me in the right difficulty zone?

• I’ve just read an article on the topic of “difficulty in learning”. I’ve always thought that making learning easier and more accessible was the point of good education. Let’s discuss this topic: provide arguments supporting the idea of making learning strategically harder.

• Act as a teacher and test me using retrieval practice on the concept of the optimal challenge point. Ask me 5 questions, one at a time, only proceeding when I answer each one. Make them progressively harder.

Links

• 📄 The eighty-five percent rule for optimal learning: This article explains the research findings on the 85% rule. As mentioned, it focuses on “binary classification tasks”, where feedback is immediate, and tasks require one of two responses. This may or may not fully apply to the kind of practice you need, but in any case, it’s an invaluable reference!

• 📑 Challenge point: A framework for conceptualizing the effects of various practice conditions in motor learning: You can download here the paper where the idea of “optimal challenge point” appeared for the first time. The graph shown in this article is a simplified version of one from the original paper.

The cover photo includes some artworks from the series “milch” (2000) by Carsten Nicolai, 21st century museum (Kanazawa).

We humans are terrible at assessing how well we’re learning. If learning feels smooth, we think it must be working. If it feels hard, we think we must be doing something wrong. We perceive effort as poor learning.

Cognitive scientists call this the misinterpreted effort hypothesis. You probably remember the illusion of fluency and the recognition trap: these happen when we mistake familiarity for true understanding. Like when you think you know a song’s lyrics... until karaoke exposes your lies. These are different manifestations of the same illusion.

In reality, mental effort is what produces learning. But not just any effort. The kind we need has a name: desirable difficulties.

What are desirable difficulties?

Desirable difficulties are learning conditions that make you struggle in the short term but help you remember and apply knowledge later. It’s an actual concept from educational psychology: Robert Bjork coined the term to describe challenges that look inefficient but lead to stronger memory and deeper understanding.

But wait, we need to talk. How can a difficulty be desirable? Isn’t this a contradiction? Indeed, the difficulty must engage attention, force the brain to make connections and stay within reach.

You could still ask: yeah, but why is effort necessary for learning? The reason lies in how memory works. Craik and Lockhart's Levels of Processing framework shows that deeper, semantic processing leads to stronger memory than shallow exposure. The brain has limited resources: it prioritizes what seems meaningful. Struggle is a signal: when something feels challenging, the brain tags it as important and kicks learning into gear. Desirable difficulties create just enough friction to trigger that response.

What counts as a desirable difficulty?

Let’s be clear: this isn’t about glorifying pain or being a masochist (although…). If you’re investing a lot of effort into learning, but it’s not the right kind of effort, your difficulty is not desirable.

For example: arranging your desk, color-coding your flashcards or having your eyes fly over the words of a textbook on autopilot. You might be putting in a lot of effort (=difficulties), but it’s not the kind that will improve your memory (=not desirable).

To tell apart which difficulties are indeed desirable, you shouldn’t just look at the kind of activity you’re doing (reading, writing, summarizing, etc.). Instead, you should focus on what’s going on in your brain while you do those activities.

If your brain is actively engaged (retrieving, connecting, predicting, restructuring), a passive-looking activity like reading can actually be more effective than a flashy one like drawing mind maps or writing summaries on autopilot.

In a previous article, we talked about the different types of cognitive load or mental effort. Indeed, Cognitive Load Theory helps here. Desirable difficulties increase germane load (effort spent on learning) while avoiding extraneous load (effort wasted on confusion or bad design).

Give me some examples

If you follow Kognitivo, you know what we’re talking about. Remember how retrieval practice, spaced repetition and interleaving work precisely because they require effort, not despite it? These three are classic examples of desirable difficulties. But there are other learning strategies that are widely accepted to be desirable difficulties:

• Generation: Trying to come up with an answer before seeing it improves memory, even if you're wrong.

• Delayed feedback: Feedback is more powerful when you wait, because your brain must reconstruct what you learned.

• Transfer: Practicing skills in different formats or contexts boosts flexibility.

Everyday applications: how to make learning usefully harder

Desirable difficulties teach us to trust the process, not the feeling. Here are some thoughts to help you master them:

• Normalize the struggle: train your brain to seek the right kind of difficulty, the one that hurts so good. Avoid the reflex of “ohh, this is getting hard, it must mean I'm doing it wrong.” Learn to sit with the discomfort.

• Challenge yourself constantly: don’t trust the feeling of comfort. If it’s getting too easy, you should get a warning signal in your brain. Get used to turning easy tasks into productive ones: try to predict answers, delay feedback, etc.

• Find the intensity that works for you, but don’t overdo it: your study routine should be sustainable long-term. Overdoing it for some days and then being burned out won’t help you. Find the optimal challenge point.

Isn’t learning supposed to be fun?

We’ve all heard the “edutainment” mantra that learning should be fun and enjoyable. Isn’t that true? Well, we need to be specific. In the cases where we can find fun in the productive struggle that produces learning, that’s ideal, but we can’t reasonably make it a general requirement.

The reason is simple: fun doesn’t produce learning, cognitive struggle does (and only the right kind). When effective cognitive struggle happens to be enjoyable, that’s great, but the enjoyment itself isn’t what drives the learning.

Consider high-stakes fields like medicine or aviation. Of course it would be wonderful if medical school or flight training were more enjoyable, but fun cannot be the non-negotiable requirement: learning must be, or else our doctors won’t heal and our pilots won’t fly planes.

This matters also for self-directed learners, who often feel discouraged when they can’t find an enjoyable way to approach difficult topics. Without external accountability (when only personal motivation keeps you going) it’s tempting to quit the moment learning stops being fun.

But embracing a different mindset is far more effective: learning to sit with discomfort and push through productive struggle. This isn’t just better for tackling tough material, it’s a fundamental skill for self-directed learning. The ability to persist through difficulty, rather than seeking constant enjoyment, is what separates those who make progress from those who abandon challenging subjects whenever the excitement fades.

The promise of zero-effort learning

It’s hard to accept, but it’s true. The reality is that there’s no such thing as zero-effort learning. We keep dreaming of one day being able to swallow a pill or plug in a memory card and then, all of a sudden, we know everything about microbiology, speak Korean and can build microprocessors. But these ideas are contradictory in nature.

The promise of “effortless learning” is a popular marketing fantasy that education products keep coming back to. “Learn Chemistry in your sleep.” “Master a language by playing a game five minutes a day.” These claims fundamentally misunderstand what learning is all about.

This also has great implications on the right use of AI in education and its main purpose: a good implementation of AI for learning isn’t the one that makes everything easier, but the one that supports you in setting up the right kind of cognitive struggle for you to learn best. That is, the use of AI that gets rid of undesirable difficulties (makes learning more convenient), but gives you the desirable ones (makes learning efficient).

By the way, do you know what’s something that’s desirable, but not a difficulty? Subscribing!

Share

Keep learning

Nerd out with your favorite AI-chatbot

Prompt suggestions. Always ask follow-up questions

• Act as a teacher and test me on “desirable difficulties” using retrieval practice. Ask me 7 questions, one at a time, only proceeding when I answer each one. Make the questions progressively harder.

• I’m going to be honest: after reading an article on “desirable difficulties”, I still believe that an ideal learning experience should always be easy, like a game. But I’m happy to discuss that. I’ll give you my reasons, please reply specifically to them and add sources to back up your replies.

• Can you help me redesign a personal study routine or training plan to include at least three types of desirable difficulties?

Links

• 📄 Memory and metamemory considerations in the training of human beings: In this paper from 1994, Robert Bjork coined the term “desirable difficulty”. The text is very accessible even today and goes into a lot of juicy detail.

• ⏯️ Using desirable difficulties to enhance learning, Dr. Robert Bjork: Here’s a video where you can listen to Robert Bjork talk about desirable difficulties and go into detail about the ideas you learned in this article.

“I’m just a visual learner,” Susan announces as she gets back the exam she didn’t pass.

She says it with half a smile, pushing her hair behind her ear as if she just cracked the code to her own intelligence. That’s why she didn’t perform well. She needs diagrams and videos. The test was made for... what are they called? “Text learners,” maybe. The teacher just doesn’t get her.

You’ve probably said or felt something like that at some point. Identifying as a specific learning style feels like self-awareness. Like finally understanding why school was so hard. The system simply didn’t match your style.

But here’s the cold, evidence-based truth: learning styles, as ways to categorize people, are mostly a myth. And clinging to that idea might be doing more harm than good.

Preferences are not performance

The core belief of learning styles is that people learn better when material matches their preferred sensory modality or cognitive style.

The idea that some learners learn best by looking at pictures, others by listening and others through hands-on activities contradicts all the scientific evidence we have. Researchers have found no consistent link between someone’s preferred learning style and their actual academic performance. In fact, teaching people in their “preferred style” doesn’t improve outcomes.

Like everyone else, you probably have your learning preferences too: you might like learning with charts rather than podcasts. But liking something isn’t the same as learning better from it.

You are a visual learner… but so is everyone else

Many people have had that moment when a diagram makes a concept click, and suddenly they think, like Susan, “Oh, I’m a visual learner.”

But here’s the unexpected truth is: everyone is. Scientists have proven that our brains are wired to process visual information incredibly efficiently, much more than text or sound alone. Visuals also help reduce the extraneous cognitive load of text-based explanations.

So yes, you’re a visual learner… Congratulations, you’re human. We are all incredibly good at processing visual input. That’s not a learning style, it’s a species trait. In fact, I’ve never heard anyone say they’re not a visual learner. Have you?

Multimodal learning is the real upgrade: dual coding

If you want a real learning boost, don’t double down on your supposed style. Combine them. Research shows people retain more when they engage with material in multiple formats. Read it, say it, draw it, act it out. The technical term is “dual coding.”

Dual coding is the process of combining verbal and non-verbal information to enhance learning. By engaging both linguistic and visual channels, it creates multiple mental representations. The more access points you activate, the more neural cross-connections and paths are built in the brain: this strengthens all memory processes: encoding, retention and transfer.

Match the content, not your preference

Even if learning styles were real (they’re not), they’d be kind of useless. Because in the real world, you rarely get to choose the format.

In reality, effective learning depends on aligning the format not with your style, but with the content. You can’t master tennis by reading about it. You don’t memorize poetry with a chart. You don’t learn how to ride a bike from a podcast. Learning works best when the format fits the cognitive demands of the task.

• Spatial task? Use visuals.

• Conceptual reasoning? Use verbal explanation or diagrams.

• Procedural skill? Use physical practice or demonstration.

Imagine going to the gym and saying, “Sorry, I only lift with my left arm. It’s just my workout style.” That’s what clinging to a learning style sounds like.

So why does the myth stick?

This myth is incredibly resilient. A study from 2021 showed that 89% of UK teachers believe in learning styles. Research conducted in the US in 2020 found that in 29 states, government-distributed materials for teacher training referenced learning styles. Teachers love this idea.

Why? Because it’s very attractive. It seems to offer a simple way to personalize instruction. A promise of inclusion. A shortcut to motivation. And once you've labeled yourself, it becomes a tidy explanation for past struggles.

But it’s the wrong tool for a good goal.

Also, two big theories have popularized this idea, but have shaky scientific support: VARK model and Multiple Intelligences.

What this belief might be costing you

The real danger of the learning styles myth isn’t just using the wrong study technique. It’s what it trains you not to do.

Believing you have a fixed style can make you less willing to experiment, less likely to persist when something feels unfamiliar and more likely to give up when the format doesn’t suit your preference. It creates a false boundary around your potential.

This is called a self-handicapping belief: a psychological defense mechanism that protects your ego (“I failed because the format didn’t match my style”) but prevents growth.

Worse, it can discourage the use of powerful strategies like retrieval practice, spaced repetition or interleaving: simply because they feel effortful or don’t fit your preferred “style.”

In other words: the learning styles myth doesn’t just fail to help you. It might be actively making you worse at learning.

Susan, listen

Every time this topic comes up, there’s quite a bit of resistance to accepting the truth: learning styles are simply a myth.

In fact, you probably still believe that, despite everything, you know your learning style. It’s hard to change long-held beliefs, especially when they relate to our identity and have helped us make sense of our past struggles. But beliefs like this might be holding you back from unlocking your full potential.

So if you take one thing from this article, let it be an invitation to try something you never thought could work for you. You might just surprise yourself.

You’re not a visual or auditory learner. You’re a human. And humans learn best when they use all the tools available to them. Stop looking for the format that matches your identity. Start looking for the one that matches the task.

Your task now? Subscribing! What’s the learning style for that?

Share

Keep learning

Nerd out with your favorite AI-chatbot

Prompt suggestions. Always ask follow-up questions

• Act as a teacher and test me using retrieval practice on “the learning styles myth and its implications for teaching and studying.” Ask me 7 questions, one at a time, only proceeding when I answer each one. Make them progressively harder.

• I’m going to be honest: after reading an article on “the myth of learning styles”, I still believe that I have a particular learning style because I’ve always learned better if I follow it. But I’m happy to discuss that. I’ll give you my reasons, please reply specifically to them and add sources to back up your replies.

• What’s the actual difference between a learning preference and a learning style?

• Why do so many teachers and trainers still use learning styles if the science doesn’t back them up?

• Act as a curriculum designer. How would you redesign a lesson that originally targeted different “learning styles” to instead use evidence-based techniques?

• Can you show me how to use dual coding or multimodal learning strategies for a complex topic I’m trying to master? (You can ask me what the topic is first.)

Links

• ⏯️ Learning styles and the importance of critical self-reflection: This is an incredible TED Talk on this topic by Psychology professor Tesia Marshik. She also goes into detail about the importance of giving meaning to what we learn.

• 📑 Learning styles: concepts and evidence: This landmark article from 2008 by Pashler and McDaniel is still today the reference on debunking learning styles. It’s an approachable and easy-to-read text. Highly recommended!

• 📚Multimedia Learning: In this influential book from 2001, Mayer provides 15 principles of multimedia learning on how learning works best with visuals and text. It’s a must when it comes to developing e-learning modules.

Imagine this: you’ve rehearsed your big speech like a pro: paragraph by paragraph, in perfect order, just like your outline says. Paragraph 1? Nailed it. Paragraph 2? Smooth. Paragraph 3? Oscar-worthy. You’re feeling unstoppable.

But then… someone raises their hand. A question. Off-script. Suddenly you’re blinking like a deer in headlights. Wait, where was I? Was that point in paragraph 2 or 4? Why is my mouth moving and nothing’s coming out?

It’s not that you don’t know your stuff. You do. But your brain’s only ever practiced it in a perfect, linear order. And real life doesn’t care about your outline.

Learning in order feels good, but it tricks your brain into relying on routine. When that routine breaks, so do you. Interleaving helps by mixing things up during practice. It’s harder, but it trains your brain to be flexible, not just fluent in a script.

Recognition is a trap

But before we dive into interleaving, let’s talk about what’s happening here. Your enemy has a name: recognition.

In cognitive science, recognition is the ability to identify something as familiar. It’s when you say: “Oh! I’ve seen this before”. Recognition picks up patterns and cues, but stays on the surface. It’s a distinct, low-effort cognitive process.

In learning, recognition is often a self-defeating trap! Our brains tend to unconsciously focus our learning around recognition instead of understanding, as a way to trick you into believing you truly mastered something.

It’s like a default shortcut your brain likes to take, because it’s less demanding. And it often happens without you realizing it… that’s why you need to fight against it!

As opposed to recognition, understanding (or comprehension) involves deep processing: being able to explain topics, implement them, use them in different contexts, make logical connections, etc.

When you read a flashcard and think “I know this!”, it’s often just “I’ve seen this before!” The verb “to know” is misleading. Languages like Spanish, French and German split this into two verbs.

So, how do you escape the recognition trap? There are many strategies to avoid the trap of recognition: one of the best ones is interleaving.

What is interleaving?

Interleaving means mixing different topics, skills or problem types during your practice sessions instead of blocking them by category or theme. Instead of drilling one skill over and over before moving on, you shuffle multiple skills, concepts or topics together.

Am I saying that not studying things in logical order is better? Yes, I am.

Why? Because when you learn in strict sequence, your brain gets good at recalling things in that context only. It builds fragile “if-then” links: if I just finished paragraph 2, then I say this line. When life scrambles the sequence, those links snap.

By breaking the order in which topics and concepts are presented, interleaving forces your brain to not look for regularities or patterns, but rather to deal with the topics themselves, to actually make sense of them and understand them.

Mix similar ideas

Interleaving is especially helpful when you’re learning ideas that are conceptually similar, but not the same and can often be confused, like grammar rules, math formulas, etc. The reason is that interleaving directs your attention to the differences across the concepts you learn together. Here are some examples:

• Languages: Practice different Spanish past tenses together to learn which fits the context.

• Math: Mix linear, quadratic, and systems of equations instead of doing them in blocks.

• Medicine: Symptoms can belong to different conditions. Shuffling similar case studies trains their correct diagnosis.

That said, interleaving isn’t only for similar concepts. It’s also powerful for building adaptability across different skill types, and yes, it even helps with encoding.

It feels wrong because it’s working

As with retrieval and spacing, interleaving doesn’t feel like progress. It slows you down. You make more mistakes.

In a classic study by Rohrer and Taylor students using interleaved math practice did worse during learning… but outperformed others on surprise tests. Why? They learned to think, not just repeat.

Interleaving builds three key learning skills:

• Discrimination: You get better at telling similar things apart. When concepts are mixed, you can’t rely on the label. You have to understand what makes them different.

• Transfer: You’re more prepared to use your knowledge in unfamiliar situations. That’s how real life works: problems don’t show up labeled.

• Flexibility: Instead of fixed responses, you gain the ability to adapt, pivot and improvise. You’re not just recalling: you’re applying.

How to interleave

Here are six quick ways to put interleaving into practice (you’ve already seen two):

• Switch order on purpose, shuffle your flashcards or questions.

• Mix similar ideas to get better at discerning or telling them apart.

• Mix problem types instead of practicing all of one kind.

• Alternate skills, like writing and speaking in language study.

• Tackle multiple topics in review sessions instead of doing one at a time.

• Combine interleaving with retrieval practice and spaced repetition.

Think: controlled chaos. Just enough disorder to keep your brain awake. If Marie Kondo saw your interleaved notes, she’d faint. But mess sparks learning more than joy.

Make it messy on purpose

Most learning materials are clean, organized and logically sequenced. That makes sense: it helps us get started. But too much structure creates the illusion that using that knowledge later will be just as neat. It won’t.

Interleaving fixes that. It brings a bit of real-life mess into your study routine. It teaches you to wrestle with confusion and uncertainty: and come out stronger. You don’t just learn the steps, but how to dance when the music changes. Interleaving makes our learning experiences closer to the unpredictability of real-life situations.

You might be nodding along thinking “Ooh, this makes so much sense!” Careful: that’s the recognition trap talking. If your ego can handle it, subscribe for more truths your brain doesn’t want but desperately needs. I won’t stop.

Share

Keep learning

Nerd out with your favorite AI-chatbot

Prompt suggestions. Always ask follow-up questions

• Act as a teacher. Test me on “interleaving” using retrieval practice. Ask 10 progressively harder questions, one at a time.

• Can you help me design a simple daily study plan that uses interleaving to learn two or three skills simultaneously?

• How can I convince a skeptical friend or colleague to try interleaving when they think “learning in order” is the best way?

• Act as a learning scientist. Explain what kinds of skills or subjects benefit most from interleaving and which might not.

• What are some common mistakes people make when trying to use interleaving? How can I avoid them?

Links

• The shuffling of mathematics problems improves learning: This article from 2007 by Rohrer and Taylor is one of the earliest and most influential demonstrations of the benefits of interleaving, specifically in math, against traditional reliance on blocked practice.

• Learning concepts and categories: Is spacing the “enemy of induction”?: This incredible article from 2008 confirmed interleaving’s benefits… by accident! Kornell and Bjork aimed to test spacing effects using interleaved artworks by different painters. While they credited spacing for the positive results, their experiment paved the way for later interleaving research.

• A systematic review of interleaving as a concept learning strategy: This extensive article tells the history of the research on interleaving in an easy-to-follow manner. It’s long but worth it: great read after this article!

Image

• In the main picture, you can see me in front of the artwork Komposition in zehn Farben 000–999 (1) by Horst Bartnig at the Neue Nationalgalerie, Berlin.

You meet someone at a party. His name is James. You smile, repeat it to yourself, even throw in a clever “Nice to meet you, James” to lock it in. Three minutes later, you're introducing him to a friend and your brain hands you... static. You stall. “This is... my new friend!”. The name is gone. Like it was never there. You sweat.

You see James one week later on the street. You concede defeat: “Sorry, I totally forgot your name…” Now he’s angry “It’s James”. Did you look like a jerk? Yes. Is the name going to stick now? Also yes.

This isn’t just a social fumble. It’s a sneak peek into how memory actually works. Your brain doesn’t just forget stuff because it’s rude. It forgets stuff because it thinks it’s doing you a favor. And that's where spaced repetition comes in.

The forgetting curve

If you already follow Kognitivo, you know that what strengthens memory is retrieval practice. But when should you do it?

To answer that, imagine memory loss as a curve. Not a gentle slope, but a cliff. You learn something today. Tomorrow, half of it is gone. By the weekend, you're lucky if you even remember what it was about.

This is the forgetting curve. And it’s not just a theory: it's been confirmed by over a century of research. Unless we revisit information, we forget most of it quickly. But here’s the twist:

Forgetting isn’t the enemy of learning. It’s the condition for it. Imagine remembering everything with equal intensity: every ad, every typo, every awkward silence. You’d be drowning in noise. Borges even wrote a haunting short story about a man cursed with perfect memory. It didn’t make him wise, it made him miserable.

Forgetting sharpens the mind. It’s subtraction with intent, a way of focusing on what matters now. If forgetting is the brain’s filter, then learning has to play by its rules. That’s where spaced repetition comes in.

Subscribe now

What is spaced repetition, really?

Spaced repetition is a learning technique based on strategically timed reviews. Instead of reviewing material repeatedly in a short period, you space out your reviews over increasing intervals of time.

The key principle is to retrieve information before you are likely to forget it. Each successful recall makes the memory more durable, allowing longer gaps between future reviews, and slows down the natural process of forgetting (the “rate of decay”). It’s a way of beating the forgetting curve at its own game.

Training your brain to care

Spaced repetition works because it signals to the brain that certain information is important. Each time you struggle a bit to recall, your brain interprets that effort as a cue to strengthen that memory. It’s like telling your brain "This is worth keeping."

Spacing doesn’t just preserve old memories. It changes how your brain stores them. Over time, those traces move from the hippocampus (short-term, fast-learning) to the neocortex (long-term, stable storage). You’re not just reviewing: you’re re-encoding.

That’s why spaced repetition feels harder than cramming. Struggling a bit to recall is a sign that the memory is being rebuilt, not just replayed. That friction is where the learning happens.

Let’s put it to practice

So how do you actually do spaced repetition?

Most people use a flashcard app that already has a built-in spaced repetition algorithm, such as Anki, Memrise or RemNote. But don’t forget that you can implement spaced repetition with any kind of retrieval, not just flashcards (remember: testing yourself, brain dumps…). Even without an app, the principle is simple.

The golden rule: after each successful review, you can usually double (x2) the interval until the next one.

Here’s a classic schedule:

• 1st review: Same day (10–20 minutes after learning)

• 2nd review: 1 day later

• 3rd review: 3 days later

• 4th review: 7 days later

• Further reviews: 14 days later, then 1 month, 3 months…

This is just a guideline. Harder stuff might need shorter gaps, easier stuff can wait longer. Adjust as needed. Studies show that adaptive schedules beat fixed ones.

Leitner system

Most apps follow some version of a system called Leitner, basically flashcards with a promotion strategy. You sort cards into different sets based on how well you know them. Every time you get a card right, it moves up to a set reviewed less often. Get it wrong? Back to the start.

It’s like a memory bootcamp: the stuff you struggle with gets more reps, and the stuff you know gets to chill. Simple, smart and really efficient.

If spaced repetition is so great, why isn’t everyone using it?

The science is solid. Decades of research say the same thing: spacing works. So why isn’t it everywhere?

Because our brains are drama queens.

Spaced repetition feels like failure. You forget a bit, struggle to recall and assume it's not working. Meanwhile, re-reading hundreds of pages the night before the exam gives you a sugar high of confidence. It feels efficient, even if it’s not. Cramming is a hack for now, it gives you the illusion of fluency. Spacing gives you real mastery. If you liked it, then you shoulda put a gap on it.

Also, let’s not forget that it involves planning. As with any plan, thinking about sticking to it is the easy part, but it’s in the execution where things often fail.

Thank you for reading!

Forgetting James’s name wasn’t a fail, it was a cognitive feature. He just happened to be your brain’s low-priority notification. But hey, spacing saved the day. Eventually.

That also goes for subscribing to Kognitivo. Haven’t done it yet? Fix it before the curve kicks in!

Share

Keep learning

Nerd out with your favorite AI-chatbot

Prompt suggestions. Always ask follow-up questions

• Act as a teacher. Test me on “spaced repetition” using retrieval practice. Ask 10 progressively harder questions, one at a time.

• How can I use spaced repetition techniques to get a friend to remember something very important?

• What are useful applications of the Leitner system at work?

• I fear that implementing spaced learning properly can take too long and be too complicated for me. Can you give me some tips on how to do it?

Links

• Gwern.net: Spaced repetition: In this website, writer Gwern Branwen sums up available research and literature on spaced repetition in a very accessible manner. Highly recommended as a next read after this post.

• Play it Again: The master psychopharmacology program as an example of interval learning in bite-sized portions (2nd link): This article describes a case study for spaced-repetition learning being implemented at a university program. It’s a really insightful and enjoyable read!

• The right time to learn: mechanisms and optimization of spaced learning: This article links spaced learning to specific neural mechanisms and it goes on detail into how to fine-tune study intervals for better memory.