What the data actually shows
Manu Kapur's work on 'productive failure' found that students who first attempt to solve problems they have not been taught how to solve — and typically fail or produce flawed solutions — often go on to understand the underlying concepts better than students who are taught the method first and then practice it. The initial struggle appears to prepare the mind to make sense of the eventual explanation.
Robert Bjork's research on 'desirable difficulties' makes a broader version of the point: conditions that slow learning and increase errors in the short term, such as spacing, testing before you are ready, and interleaving topics, tend to improve retention and transfer in the long term. A related effect, sometimes called errorful generation, finds that trying to generate an answer and getting it wrong — then seeing the correct answer — can lead to better memory than simply studying the correct answer from the start.
The consistent thread is that the error has to be paired with feedback. Studies of error-based learning generally find the benefit appears when learners find out the right answer shortly after their attempt. Without that correction, mistakes can simply be repeated and reinforced. So the honest claim is narrower than 'mistakes help' — it is closer to 'effortful attempts plus correction help.'
Why this feels different from how it actually is
Mistakes feel like failure rather than learning because the discomfort is immediate and the benefit is delayed. The struggle is unpleasant now; the improved retention shows up later, on a test or in real use, where it is hard to trace back to the earlier difficulty. So the methods that work best often feel like the ones that are going wrong.
Smooth study feels productive for the same reason it often isn't. Rereading and being shown the method up front produce a sense of fluency — the material feels familiar and easy — which is easily mistaken for having learned it. That fluency tends to collapse later, when you have to retrieve the information without support.
There is also a social layer. Many people learn that mistakes are something to avoid and be embarrassed by, so error feels like evidence of inadequacy rather than a normal, even useful, part of getting better. That framing makes the productive-struggle approach feel risky even where the evidence supports it.
The methods that work best often feel like the ones that are going wrong.
What the research says to do about it
Attempt before you are taught. Trying to solve or explain something before you have been shown how — even knowing you will probably get it wrong — appears to prime later understanding. The point is not to succeed on the first try but to engage with the problem first and let the eventual explanation land on prepared ground.
Test yourself and tolerate getting things wrong, as long as you check the answer. Self-testing with feedback is one of the best-supported study methods there is, and part of its power is that wrong attempts followed by correction tend to stick. Treat a wrong answer as information about what to study next, not as a verdict.
Build in feedback deliberately. Because the benefit of error depends on correction, the practical move is to make sure you find out the right answer soon after each attempt — through answer keys, worked solutions, a teacher, or a peer. Struggle without correction is the version that does not pay off.
What the research says does not help
Making mistakes with no feedback does little and can entrench the error. The benefit in the research comes from the correction that follows the attempt; repeating a wrong approach without ever finding out it is wrong tends to reinforce it rather than fix it.
Avoiding all difficulty in the name of 'efficient' study tends to backfire. Rereading, highlighting, and being walked smoothly through every step feel effective and produce fluency, but that fluency frequently does not survive a later test. The comfortable methods are often the weak ones.
Treating every mistake as a sign of inadequacy is both inaccurate and counterproductive. The research frames errors-with-feedback as a normal part of effective learning, not as evidence of low ability. A harsh self-judgment after each error tends to discourage exactly the kind of productive attempt that helps.
Do not trust the feeling of ease as a sign of learning, and always close the loop with feedback.
What this looks like in real life
Attempting the problem before you're shown how
In productive-failure studies, students who first wrestle with a problem they haven't been taught to solve — and usually get it wrong — often understand the underlying concept better once the method is explained than students taught the method up front. The early struggle seems to prepare the mind so the eventual explanation lands on ready ground.
Guessing wrong, then seeing the answer
Someone tries to recall an answer, gets it wrong, then checks the correct one. That errorful attempt followed by correction can produce better memory than simply studying the right answer from the start — as long as the loop is closed. The same wrong guess with no correction just risks reinforcing the error.
Real numbers in context
The honest version of this finding is conditional, not a slogan. Across studies of productive failure, errorful generation, and desirable difficulties, the pattern that holds up is that effortful attempts plus feedback tend to outperform smooth, error-free study on later retention and transfer. The size of the advantage varies by task and learner, and some difficulties are unproductive rather than desirable — the word 'desirable' is doing real work.
What is well established is the gap between how a method feels and how well it works. Learners and even instructors routinely rate the easier, fluent methods as more effective, while controlled comparisons favour the harder, more error-prone ones. So the practical takeaway is less 'mistakes are good' than 'do not trust the feeling of ease as a sign of learning, and always close the loop with feedback.'