What is metacognition?

Metacognition is the ability to monitor and regulate your own thinking. The National Academies' consensus definition puts it as monitoring and regulating one's own cognitive processes and consciously regulating behaviour, including affective behaviour. Graded entry. Both halves are load-bearing. Noticing that you do not understand something is monitoring. Doing something different about it is regulation, and knowing without the control does nothing.

The accuracy of the monitoring has its own name in that literature: calibration. It sits underneath most of the learning argument on this site and has been assumed here rather than explained.

The reason it matters is that the signal people naturally use is the wrong one. Fluency feels like learning. Familiarity feels like mastery. Both feelings are produced by ease of processing rather than by durability of memory, and the two come apart reliably enough to be measured.

The conditions that feel best teach least

Bjork and Bjork set out the finding that organises the rest: the study conditions producing the highest confidence produce the least durable learning, and conditions that feel like they are going badly produce the most. Graded entry. Learners asked to choose their own method reliably choose the one that feels best, which is reliably the weaker one.

Roediger and Karpicke put a number on the reversal. Restudying beat testing at five minutes, 81 per cent against 75. At one week, testing beat restudying, 56 against 42. Graded entry. A learner judging their own progress at the end of a study session is sampling the first measurement and acting on it.

Fisher and colleagues showed the misjudgement can be induced by a tool. Searching the internet inflated people's estimates of their own unaided knowledge, and did so even on questions the search had never touched. Graded entry. Access to an external source was being experienced as personal knowledge, which is the exact failure metacognition is supposed to prevent.

Why trying harder to self-assess does not fix it

The instinct on hearing this is to judge yourself more carefully. That does not work, because the judgement is the faculty that is broken. What works is installing checks outside the judgement.

Rowland's meta-analysis of 159 effect sizes across 61 studies puts the testing effect at g = 0.50. With feedback it rises to 0.73; without feedback it falls to 0.39. Graded entry. The feedback split is the metacognitive part. Retrieval alone strengthens the memory. Retrieval plus correction also recalibrates the estimate of what you know. The effect nearly doubles on the strength of that second thing.

Cepeda and colleagues add the timing: the spacing between practice sessions that produces the best retention widens as the target retention interval widens. Graded entry. A check taken close to the learning flatters it.

Brown, Roediger and McDaniel's Make It Stick is the accessible statement of all of this, written with two of the researchers who produced the underlying work.

What AI changes

The mechanism above is decades old and holds without any AI in the picture. One thing follows from the tools themselves rather than from new evidence about people: a model can make difficult material feel understood before the understanding exists. A fluent explanation of something you could not reconstruct produces precisely the signal that metacognition uses.

The measured version is Sankaranarayanan's. Seventy-eight participants in three conditions, manual, unrestricted AI and scaffolded AI. Both AI groups beat the control on the work and were statistically indistinguishable from each other. On the subsequent task with the tool removed, the unrestricted group failed at 77 per cent against 39 for the scaffolded group. Graded entry.

Nobody in the weaker group knew they were in it. That is the practical consequence for any organisation trying to measure AI's effect on capability by asking people: the survey instrument and the broken faculty are the same thing. This estate treats that at the illusion of competence.

What this does not settle

The studies are laboratory and classroom work on verbal and procedural material. Whether professional judgement is subject to the same calibration failure at the same magnitude has not been tested, and the one clinical measurement available, the fall in unassisted detection among experienced endoscopists, was made by researchers rather than reported by the doctors.

There is also a real counter-position. Metacognitive accuracy is not free: constant self-checking has a cost, and in domains where the environment gives fast clear feedback, Klein's tradition holds that experts should trust recognition rather than interrogate it. The case for external checks is strongest exactly where feedback is slow or absent, which is most professional work and not all of it.

What follows

Three things, and none of them is a course.

Test rather than review. The difference between reading it again and trying to produce it from memory is the entire effect, and only the second one tells you anything about what you hold.

Get the feedback, because that is the half that recalibrates. Retrieval without correction strengthens whatever you retrieved, including the wrong version.

Distrust the feeling specifically when the material was easy. Ease is evidence about the presentation, not about you. When an explanation arrives already fluent, the signal metacognition normally uses has been supplied by something other than your understanding.

Key sources

• Bjork, E. L. and Bjork, R. A. (2011). Making Things Hard on Yourself, But in a Good Way. Graded entry.
• Roediger, H. L. III and Karpicke, J. D. (2006). Test-Enhanced Learning. Psychological Science, 17(3). Graded entry.
• Rowland, C. A. (2014). The Effect of Testing Versus Restudy on Retention. Psychological Bulletin, 140(6). Graded entry.
• Fisher, M., Goddu, M. K. and Keil, F. C. (2015). Searching for Explanations. Graded entry.
• National Academies of Sciences, Engineering, and Medicine (2018). How People Learn II: Learners, Contexts, and Cultures. The National Academies Press. Graded entry.
• Brown, P. C., Roediger, H. L. III and McDaniel, M. A. (2014). Make It Stick: The Science of Successful Learning. Belknap Press. In the essential works.

Related SuperSkills research

The applied version is the illusion of competence. The mechanisms are retrieval practice, desirable difficulty and productive struggle. On the adjacent faculties, critical thinking and judgement. On measurement, assessing capability rather than output. On what accumulates, capability debt.

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. Metacognition is an established field in cognitive psychology and is not a SuperSkills coinage. This page follows the rule this research uses for foundational concepts: established work answers the human mechanism, and current studies answer what AI changes.

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