Why most e-learning doesn't work - and what the research actually says
There is a moment most instructional designers recognise. You finish a course. It looks good. The client approves it. Learners click through, hit the knowledge check at the end, score 80%, and receive their completion certificate. Six weeks later, you wonder whether any of it stuck.
The uncomfortable answer, if you follow the research, is probably not much.
The tell-and-test problem
The dominant format in organisational e-learning follows a simple pattern: present information, then test whether learners absorbed it. Slides of content, perhaps with a narrator, followed by a handful of multiple-choice questions to confirm completion. This format is efficient to produce, easy to track, and almost universally used. It is also, by the standards of learning science, one of the least effective approaches available.
The issue is not that testing is bad. Testing is actually one of the most powerful tools we have. The issue is the sequence: present everything first, test at the very end. By that point, the test is measuring short-term recognition of recently seen material, not durable learning. Pass rates look good. Retention curves tell a different story.
What a century of research actually shows
The evidence that active learning outperforms passive delivery is not new, not thin, and not limited to schools or universities.
A landmark 2014 meta-analysis published in PNAS analysed 225 studies across undergraduate science, engineering and mathematics. Students in active learning conditions scored an average of nearly half a standard deviation higher on exams. Students in lecture-only formats were 1.5 times more likely to fail. The researchers described continuing to teach by lecture alone, given this evidence, as a kind of experiment on students - one without their informed consent.
In 2011, a controlled experiment at the University of British Columbia offered an even sharper result. Researchers compared an experienced, highly-rated lecturer against an instructor with no teaching experience using active methods. The active learning group outperformed the experienced lecturer by more than double on an identical assessment - 74% versus 41%. The medium did not determine the outcome. The method did.
These are not isolated findings. Effect sizes around d = 0.47 appear consistently across subjects and settings, including a 2022 meta-analysis extending the evidence to humanities and social sciences. The direction of the effect is one of the most reliably replicated findings in educational research.
The testing effect: the most underused tool in e-learning
If you had to choose a single finding from cognitive psychology to build a learning design philosophy around, it would probably be the testing effect.
In a now-classic series of experiments, Roediger and Karpicke (2006) asked students to study a passage of text using two different approaches. One group read it four times. Another read it once, then recalled as much as they could - three times, without looking at the material. On an immediate test, the re-reading group performed slightly better. One week later, the retrieval group outperformed them by 21 percentage points. Restudying led to 56% forgetting over two days. Retrieval practice produced only 13%.
The mechanism matters here. When you retrieve something from memory, you are not just checking whether it is there - you are strengthening the memory itself. Each retrieval makes the next retrieval easier and more durable. Re-reading, by contrast, creates a fluency illusion: the material feels familiar, which learners mistake for knowing it.
A meta-analysis by Adesope, Trevisan, and Sundararajan (2017) synthesised 272 effects from 118 articles. The advantage of retrieval practice over restudying: g = 0.51. Over no-activity controls: g = 0.93. The effect held in classrooms as well as laboratories.
For e-learning designers, this has a direct implication. Questions are not just assessment tools. Placed before content, they prime attention. Placed during content, they interrupt passive processing at exactly the right moment. Placed after content - especially after a delay - they do more for long-term retention than any amount of re-exposure to the material. The tell-and-test format gets the sequence backwards.
Not all clicks are equal
It is tempting to conclude from this that interactive e-learning is simply better than non-interactive. The reality is more specific - and more useful.
Michelene Chi and Ruth Wylie's ICAP framework (2014, Educational Psychologist) offers a practical taxonomy. They distinguish four modes of engagement:
- Interactive - co-constructing through dialogue or debate
- Constructive - generating new output: self-explanations, decisions, predictions
- Active - manipulating materials: clicking, highlighting, dragging
- Passive - receiving without overt response
The predicted learning outcomes follow in that order, with one critical detail: the largest jump in learning occurs between Active and Constructive. Much of what e-learning calls "interactive" - click-to-reveal panels, drag-and-drop sorting, navigation controls - falls into the merely Active category. Learners are doing something, but not necessarily thinking.
This matters because there is consistent evidence that high interactivity without cognitive purpose can actually harm learning. Extraneous interactions consume working memory resources that would otherwise be used to process the material itself. A study in medical education found that simpler "click" interactions outperformed more complex "drag" interactions precisely because the cognitive load of the manipulation competed with content understanding. Several studies now document an inverted-U relationship between interactivity and learning outcomes: moderate interactivity tends to produce the best results.
The implication is not to make e-learning simpler. It is to make interactivity earn its place. A well-designed decision point in a scenario, where learners must commit to a choice and live with the consequences, generates far more learning than ten click-to-reveal panels. A self-explanation prompt - "why do you think that approach would work here?" - activates constructive processing that passive delivery never reaches.
An honest caveat
The research is clear in direction but comes with genuine limitations worth naming.
Most studies measuring active learning advantages use short-term retention in academic settings with motivated student populations. The effect sizes in corporate, self-paced, optional e-learning - where attention is divided, stakes feel lower, and completion is often the actual goal - may be smaller. Large-scale randomised trials in workplace learning contexts remain scarce.
There is also the expertise reversal effect, documented across dozens of studies by Kalyuga, Sweller, and colleagues: instructional scaffolding that helps novices can become redundant or even counterproductive for experienced learners. A compliance course that walks a new hire through every principle step by step will frustrate a seasoned professional. The same interactivity level for all learners is always a compromise.
And perhaps most striking: a Harvard crossover study (Deslauriers et al., 2019) found that students consistently preferred passive lectures but learned significantly more from active learning. Learner satisfaction scores and actual learning outcomes were negatively correlated. This should give anyone designing for learner feedback or course ratings some pause - the formats that feel easiest often teach least.
What this suggests about design
The evidence points less toward a specific format and more toward a few persistent principles.
Make learners retrieve before you tell them the answer. Space questions across a course rather than bunching them at the end. Use scenarios that require genuine decisions, not just recognition of the obvious wrong answer. Match complexity to learner expertise. And be honest with clients about what completion metrics measure - and what they do not.
This is not a new insight. Learning designers have been making these arguments for decades. The research simply gives those arguments firmer ground.
This is the first in a series exploring the evidence behind learning design decisions - what the research actually supports, where it is mixed, and how it shapes the way we build courses at LearnBuilder.