How to make knowledge really stay?
This was the question our client, the owner of a professional learning platform, asked during consultation.
We often meet clients who focus on improving features, dashboards, or gamification – all useful components when applied with purpose. But their real value appears only when they help achieve the main goal: to make learning deliver measurable and lasting impact. The best way to do that is to understand how people actually remember and apply information and to design platforms that follow these principles.
This client wasn’t just chasing higher completion rates or improving certification outcomes. The wanted was for knowledge to retain, not for a week or until the next test, but for good, firmly rooted in learners’ minds. That focus made sense: the courses offered on their platform are essential for professionals developing real-world skills. The knowledge gained must not only be memorized but successfully applied in practice.
This approach, they realized, could help their platform stand out in a market, because learners will return to it not out of obligation, but because it genuinely helped them grow, recommend it to others, and see tangible value in every course they took.
We started by studying their platform and its competitors. During that process, I recalled a book that had once reshaped how I think about education – “Make It Stick” by Brown, Roediger, and McDaniel.
It gathers decades of research in cognitive psychology and explains why many of our usual learning habits fail us.
Rereading gives us comfort, not understanding. Highlighting feels productive but rarely changes how we retain information. The methods that truly work, such as retrieval practice, spaced repetition, interleaving, are less intuitive but far more effective in the long run.
That book became our starting point. From there, we explored more recent studies and how these well-established principles are being applied in modern eLearning platforms. And the results were surprising. Despite the research being widely available for more than a decade, many platforms still ignore it.
Of course, some major players, for example, Duolingo, Khan Academy, Udemy, or OttoLearn, have started applying these ideas in practice. But the overall market remains far behind.
After analyzing dozens of courses, we saw a clear pattern. Around 80% still follow the same outdated formula: watch a video, take a test, get a certificate. Very few are designed around retrieval and knowledge reinforcement – the processes that actually make learning last.
Read, test, forget… What indeed kills real learning
So, is there a master key to lasting knowledge? Before we can search for it, we need to understand what patterns quietly make learning fade away.
Cognitive science can shed light on this. Over the years, researchers have found several recurring reasons why traditional methods rarely build true understanding. Here are four of them.
1. The illusion of mastery
The greatest enemy of learning isn’t ignorance; it’s the illusion of knowing.
When something feels familiar, the mind relaxes. You reread a paragraph, recognize a concept, and it seems almost effortless. That sense of ease convinces you the material is learned, when in fact, you’ve only grown comfortable with repetition.
This illusion isn’t limited to reading. In a study at Northwestern University, participants watched videos demonstrating different skills, for example, throwing darts or doing the moonwalk, up to 20 times. With every viewing, their confidence grew. Yet when they tried to perform the tasks, their actual results didn’t improve at all. Watching gave them the feeling of progress, not the skill itself. The researchers called this the illusion of skill acquisition.
The principle is the same in learning. Learners move through slides, take quick quizzes, and feel certain they’ve mastered the topic. But a week later, most of it has faded. Familiarity creates the comfort of progress without the learning that lasts.
2. The learning styles myth
You’ve probably heard someone say, “I’m a visual learner,” or “I learn best by listening.”
The idea that people learn better when content matches their “preferred style” remains widespread. But decades of research show no consistent evidence that such alignment improves outcomes.
The persistence of this belief shapes how many courses are designed: adding visuals for “visual learners,” voiceovers for “auditory” ones, and so on. Yet the mechanism of learning doesn’t depend on style matching. It depends on how information is processed and recalled.
3. The fixed mindset trap
Besides “I’m a visual learner” or “I can only focus with audio,” there are other phrases often heard in classrooms and training sessions: “I’m good at math” or “I’m bad at languages.”
Sounds harmless, right? But it carries a quiet message that ability is fixed, not developed. When people believe intelligence is something they’re born with, they tend to avoid what challenges them. Struggle feels like proof of limitation rather than a step toward progress.
This belief often shapes digital learning, too. If a task feels too easy, learners stay comfortable but bored. When it’s too hard, they give up.
4. What real knowledge is
The last misconception hides in plain sight – the assumption that remembering equals knowing.
Many platforms measure progress by how much learners recall, not by whether they can use that knowledge outside the course.
But recognition and understanding are not the same. Someone may easily repeat a definition yet struggle to apply it in a real situation. That gap marks the difference between temporary recall and lasting competence.
Cognitive science principles that actually work – and how AI helps put them into practice
William Butler Yeats, popular Irish author, once said, “Education is not the filling of a pail, but the lighting of a fire.”
If traditional learning is about filling that pail, then what can light the fire? Cognitive research offers some answers; it shows what truly helps knowledge stay. Below, we explore proven approaches and how AI can make them practical and scalable.
Retrieval practice and spaced learning
People often talk about the learning curve, but ignore the forgetting curve.
The forgetting curve, first described by psychologist Hermann Ebbinghaus in the late 19th century, is a pattern showing how quickly memory fades.
After learning something, we may retain nearly 100% of it right away, but within an hour, that can drop to around 50%. Within a day, we might remember only 30%. And this is assuming we really learned something in the first place.
What truly strengthens memory is when knowledge is spaced over time and recalled with effort.
If the part about spacing is clear, then what does it mean to recall with effort?
Most courses today are built to engage and motivate learners – through streaks, badges, progress bars, and other gamified tools. These elements play an important role in keeping users active. But engagement alone doesn’t guarantee retention. When the focus shifts too much toward keeping learners entertained, the course can create the illusion of learning without actually reinforcing memory.
Putting effort means making our brain work harder – retrieving, connecting, and reconstructing information rather than just recognizing it. And yes, sometimes that’s uncomfortable. Learning isn’t always smooth or pleasant, but it’s precisely the struggle that makes knowledge last.
This principle has proven so effective that medical and nursing schools have fundamentally restructured how students prepare for high-stakes licensing exams, with retrieval practice now central to their curricula.
Why traditional systems struggle with retention
Like the learning curve, the forgetting curve is personal. The rate of forgetting varies depending on memory strength, difficulty of the material, and even physiological state (sleep, stress, and attention all influence memory).
Most platforms deliver a one-time course but do little to combat this natural decline. And if they do offer review options, those are usually generic and fixed in time, not personalized to each learner’s forgetting curve. As a result, everyone reviews the same material at the same pace, regardless of how quickly their memory fades.
How AI can make practice actually work
AI can automate and personalize both when and what to review.
Adaptive algorithms analyze learner performance, for example, how accurately and quickly they recall information or how often they need reminders, and use this data to predict individual forgetting patterns.
Based on that, the system can schedule review sessions at the right moments, vary the difficulty, and mix in different question types.
For example, you can add to your LMS an AI-powered micro-test and quiz creator. This is a module that delivers short quizzes, flash prompts, or micro-reviews embedded directly into the course. Instead of rewatching everything at once, learners get small, well-timed reminders that reinforce long-term retention.
Interleaving and varied practice
Earlier, we mentioned that real learning often requires effort. Here’s another case where the process may not feel easy, but the outcome is worth it.
Research shows that switching between topics and task types – a method known as interleaving – feels less comfortable at first but results in more flexible and durable knowledge.
In simple terms, interleaving means mixing related concepts or skills instead of practicing one at a time.
Imagine preparing for a project management certification. In a blocked format, you might study all scheduling topics first, then budgeting, then risk management. After several similar tasks, you can rely on repetition rather than real understanding.
With interleaving, scheduling questions might appear alongside budgeting or risk scenarios. This constant shift forces your brain to compare, distinguish, and recall. It feels slower, but this very struggle strengthens learning.
A 2022 study in high school science classes tested how combining retrieval practice (quizzing) with interleaving (mixing topics) affected learning.
Weekly quizzes with interleaved questions improved scores on a delayed test taken a month later: 63% compared to 54% with blocked questions. The difference came not from more study time but from how the brain was asked to work.
Why it’s difficult to use in practice
Instructors and content creators rarely have the time to design such patterns. Building interleaved schedules means mapping entire courses, finding logical links between topics, and preparing mixed assessments. Doing this for a single course takes hours; doing it for many is unrealistic.
Another reason is time pressure. Many courses aim to “cover” as much content as possible.
Revisiting earlier topics seems inefficient, as if it slows progress, even though evidence shows the opposite.
As a result, interleaving remains more of a scientific insight than a widespread teaching practice.
How AI can help implement varied practice
Luckily, this task can be delegated to AI systems that can handle the complexity of content sequencing and learner analysis.
Unlike humans, algorithms can continuously track performance, detect patterns, and recommend when it’s best to mix topics or revisit earlier material.
For example, an adaptive learning path feature can be added to an LMS. It automatically alternates between related subjects – say, a short theory quiz followed by a real-world scenario from another
module – and reintroduces older topics just before they’re forgotten.
This makes the process more adaptive and data-driven without extra work for instructors. Learners still experience the necessary challenge, but it’s structured in a way that fits their individual pace and performance.
Desirable difficulties and generation effect
The term “desirable difficulties,” introduced by psychologist Robert Bjork, describes learning conditions that intentionally make the process more effortful, but in a productive way.
The idea might sound counterintuitive, yet it’s close to what we discussed earlier: real learning often comes with a certain degree of challenge.
However, desirable difficulties also implies carefully adjusting the level of effort required to learn. The goal of this method is not to make learning frustrating, but to ensure that learners engage their memory, problem-solving, and reasoning before receiving the answer.
A related concept is the generation effect – the tendency to remember information better when you generate it yourself rather than simply reading or hearing it.
For instance, trying to predict the outcome of a business scenario before being told the answer helps you retain the logic behind it far more effectively than passively reading the explanation afterward.
Why it’s difficult to apply in real courses
The challenge is that what feels “desirably difficult” to one learner might feel too easy or too overwhelming to another. The optimal difficulty level is highly individual, depending on prior knowledge, pace, and even motivation.
For course creators, this presents a major obstacle. Designing scenarios that hit the right balance requires deep understanding of both the subject matter and the learner’s mindset. Creating these experiences manually, especially across multiple modules, is time-consuming and complex.
How AI can create the right level of challenge
Smart AI-driven tutoring systems and chatbots can create adaptive learning experiences that encourage learners to think before being shown the answer. By analyzing responses, these tools can detect when a learner is ready for a tougher question or when they need more support.
Consider this example from a customer service training module:
Here, instead of simply testing recall, AI tutor prompts reasoning, reflection, and correction in real time.
In practice, similar tools already exist. For example, a Feynman Chatbot can be integrated into modern LMS platforms.
Inspired by the Feynman learning technique, it encourages learners to explain concepts in their own words, identifies gaps in understanding, and offers clarifications until the explanation is clear and complete.
How to stand out by delivering true value in digital learning
There is a steady tension in digital education. Research shows that lasting learning requires effort – through retrieval, spacing, interleaving, or generation. Yet learners often prefer formats that feel simple and efficient: short videos, visible progress bars, and quick quizzes.
These two forces shape how courses are built. If learning feels too demanding, many learners disengage before finishing. If it feels too easy, they move quickly but retain little.
The preference for ease is partly practical. People learn alongside work and other tasks, so they choose materials that fit short time slots. Platforms, in turn, focus on metrics such as completion rates and time spent, which reward content quantity over cognitive quality.
A large content library may boost engagement, but it rarely ensures durable understanding. Cognitive design aims for retention and transfer, helping people recall and apply what they’ve learned. That means scheduling reviews, mixing practice types, and prompting learners to generate answers instead of only recognizing them.
Creating such conditions manually is time-consuming. However, AI has already become a good helper for these tasks. Adaptive algorithms can decide when to review, what to mix, and how much challenge to introduce.
Some platforms are already proving that it works. For example, Duolingo applies spaced repetition to resurface words just before they’re forgotten, leading to longer streaks and better long-term recall. Quizlet integrates memory-based review to improve retention. Coursera uses AI-driven feedback and adaptive testing, with tools like its essay grader cutting response time tenfold and raising engagement in writing-heavy courses.
The outcome isn’t only improved learner experience. Companies that build cognitive science into their platforms collect far richer learning data: patterns of forgetting, the types of feedback that work best, and how effort translates into mastery. These insights form feedback loops that continuously improve both the learner and the system.
Still, these cases are not common. Only a minority of learning platforms consistently apply evidence-based design.
In professional and vocational training, it’s even rarer: only 32% of recent LMS deployments include adaptive learning engines.
However, in vocational fields, such as medicine, aviation, engineering, decay of knowledge has real-world consequences. Here durability of learning is not optional.
This is also where the business opportunity lies. Platforms that apply learning science and AI-based personalization are not just improving education, but also shaping the next competitive standard. As more companies recognize that retention equals value, learners will naturally gravitate toward solutions that help them actually remember and apply what they’ve learned. Being early in this shift means leading the market, not catching up to it later.
I believe the future of learning lies in that space between effort and ease. Knowledge that endures doesn’t come from removing difficulty but from shaping it with care.
AI gives us the tools to do that. When used wisely, it helps design learning that challenges without exhausting, supports without simplifying, and turns short-term effort into lasting, practical expertise.
Evaluate how your LMS supports lasting learning
If you’re responsible for an eLearning product, these questions can help assess how closely your system aligns with evidence-based learning principles.
Ask yourself:
- Do we test retrieval, not just recognition?
(Multiple-choice recognition is easy to measure but weak for memory.) - Is content revisited across modules and time?
(Single exposure rarely leads to durable recall.) - Does the system respond to performance, not only completion?
(Adaptive means adjusting to forgetting, not just personalizing appearance.) - Is feedback generative, not merely corrective?
(Learners should understand why an answer is wrong and how to improve it.) - Do we mix topics and practice types instead of blocking them?
(Interleaving supports transfer and long-term retention.) - Are review intervals individualized?
(Forgetting varies between learners; schedules should reflect that.) - Do learners generate answers or explanations, not only recognize them?
(Production strengthens recall and application.) - Do we monitor where learners disengage – from overload or lack of challenge?
(Both excessive and absent difficulty reduce learning value.)
These questions show how much of what we know about learning science is actually built into your product.
If some answers raise doubts, that’s a good sign. It means there’s room to build courses that not only attract learners but help them truly remember. The next step is to translate these principles into action, and AI makes that possible at scale.
We’ve prepared a short playbook with examples of how these mechanisms can look in practice: AI-powered micro-tests, adaptive review schedules, feedback that teaches, not just scores, and other features.
Exploring them might help you see where your platform stands and where it can grow.
