What is retrieval practice?

Definition. Retrieval practice is the finding that pulling information out of your own memory strengthens it more durably than studying the same material again. The act of recall is what builds the memory. Re-reading mostly builds the feeling of having built it.

The result that reverses depending on when you measure

Roediger and Karpicke gave students prose passages and compared restudying with being tested, then measured recall at different delays. At five minutes, restudying won: 81 per cent against 75 per cent. At one week, testing won: 56 per cent against 42 per cent. Their second experiment sharpened it. Repeated study led at five minutes, 83 per cent against 71, and at one week had collapsed to 40 per cent while the repeatedly tested group held 61. Graded entry.

Karpicke and Blunt then ran retrieval practice against a more elaborate and more respectable-looking method, concept mapping. Retrieval scored 0.67 against 0.45, roughly a 50 per cent advantage in retention a week later, and 101 of 120 students did better after retrieval than after elaborative study. Graded entry. A published comment by Mintzes and colleagues disputes the fidelity of the concept-mapping condition, which anyone citing the study should say.

Rowland's meta-analysis aggregated 159 effect sizes from 61 studies across four decades and found the effect holds at g = 0.50. With feedback it rises to 0.73; without feedback it falls to 0.39. Graded entry. The feedback split matters more than the headline, because it says retrieval alone is worth roughly half what retrieval plus correction is worth.

Why the better method feels worse

Rereading is fluent. The words go past easily and the ease gets read as knowing. Retrieval is effortful, slow, and produces errors in front of the person making them. That is why students abandon it, and also why it works. This is the same mechanism as desirable difficulty, and it produces the illusion of competence on the other side.

The practical consequence is about measurement rather than study technique. A test taken close to the learning ranks the methods in the wrong order. Any organisation evaluating a new way of working on a short horizon will systematically prefer whichever version taught people least.

The retrieval that AI removes

Asking a model is not retrieval. The answer arrives from outside, the effortful recall never happens, and the strengthening that depended on it does not occur. Everything else about the task can look identical: the work ships, the output is good, and the person feels informed.

This is the mechanism underneath capability debt. It explains how capability can fall while output stays high. Unaided work has to be preserved on purpose, because nothing in the output will report that it has stopped happening. It is also the reason cognitive offloading has a cost that offloading research on calculators and notebooks did not fully anticipate: a calculator removes an arithmetic step, and a language model removes the recall, the framing and the judgement in one move.

What this evidence does not settle

The studies are laboratory and classroom work on verbal material, mostly prose recall in undergraduates. Nobody has run Roediger and Karpicke's design on professional judgement in a workplace, and the transfer from remembering a science passage to knowing whether a deal is badly structured is an analogy rather than a measurement. The effect sizes quoted here should be treated as evidence that the mechanism is real and well replicated, and not as a predicted magnitude for anything at work.

Key sources

• Roediger, H. L. III and Karpicke, J. D. (2006). Test-Enhanced Learning: Taking Memory Tests Improves Long-Term Retention. Psychological Science, 17(3), 249-255. Graded entry.
• Karpicke, J. D. and Blunt, J. R. (2011). Retrieval Practice Produces More Learning than Elaborative Studying with Concept Mapping. Science, 331(6018), 772-775. Graded entry.
• Rowland, C. A. (2014). The Effect of Testing Versus Restudy on Retention: A Meta-Analytic Review of the Testing Effect. Psychological Bulletin, 140(6), 1432-1463. Graded entry.

Related SuperSkills research

Retrieval practice is one of the three learning mechanisms this research rests on, alongside desirable difficulty and productive struggle. Its absence is what capability debt accumulates from, and the illusion of competence is why nobody notices. See also cognitive offloading, the missed reps, how humans learn with AI and do I still need to remember things.

About this research

Rahim Hirji is the author of SuperSkills (Kogan Page, 2026), keynote speaker on AI and human capability, and founder of The SuperSkills Intelligence Company. Retrieval practice is an established finding in cognitive psychology and is not a SuperSkills term. Findings are attributed to the studies that produced them and kept separate from the interpretation.

How this research works  ·  Reviewed quarterly  ·  Found an error? Tell me and it is corrected on the page.