What Is Metacognition? A Simple Definition
Metacognition is your awareness of your own thinking and your ability to steer it. People often shorten it to "thinking about thinking." In practice, it means knowing what you know, noticing what you don't, and changing how you study when the current approach isn't working.
The developmental psychologist John Flavell is usually credited with coining the term in the 1970s. His 1979 paper in American Psychologist, which announced metacognition as "a new area of cognitive-developmental inquiry," described it as "knowledge and cognition about cognitive phenomena."
That shorthand is catchy, but it leaves out the part that matters most for students. Flavell's paper was just as much about monitoring: noticing how well your memory and comprehension are working, and acting on what you notice. Knowing that you learn better by testing yourself is metacognitive knowledge. Actually stopping mid-chapter because you realize you can't explain the last page, and switching strategies, is metacognitive regulation.
A useful way to picture it comes from Thomas Nelson and Louis Narens, who in 1990 described two levels of thought. The object level does the work: reading, solving, remembering. The meta level sits above it, watching. Information flows up (monitoring: "I don't understand this paragraph") and decisions flow down (control: "reread it slowly, then try to explain it"). Good learners aren't necessarily faster at the object level. They're better at the traffic between the two.
Types of Metacognition: Knowledge and Regulation
Most researchers divide metacognition into two types: knowledge of cognition and regulation of cognition. Gregory Schraw and Rayne Dennison built one of the most widely used questionnaires on this split, the 52-item Metacognitive Awareness Inventory (1994), and their categories are a practical map of what to work on.
| Component | Subskill | What it means | Student example |
|---|---|---|---|
| Knowledge of cognition | Declarative knowledge | Knowing about yourself as a learner and about strategies | "I lose focus after about 25 minutes of reading." |
| Knowledge of cognition | Procedural knowledge | Knowing how to use a strategy | "I know how to turn a heading into a question." |
| Knowledge of cognition | Conditional knowledge | Knowing when and why a strategy fits | "Flashcards work for vocabulary, not for proofs." |
| Regulation of cognition | Planning | Setting goals and choosing an approach before you start | "Tonight: chapters 4 and 5, then a 10-question self-test." |
| Regulation of cognition | Monitoring | Checking comprehension and progress while you work | "I've read this paragraph twice and can't summarize it." |
| Regulation of cognition | Evaluating | Judging the result and the process afterward | "I missed every question on ratios. I need worked examples." |
Schraw and Dennison also score information management and debugging, but for a student the three stages of plan, monitor, and evaluate are the part worth memorizing. The Education Endowment Foundation's guidance for teachers is built on the same cycle.
Conditional knowledge is the subskill most people skip. Knowing that retrieval practice works is declarative. Knowing how to do it is procedural. Knowing that it's the right tool for a biology exam in three weeks, but not for untangling a confusing argument in a philosophy paper you're reading for the first time, is conditional. Most strategy advice skips this part and hands you tools without saying when to use each one.
Metacognition Examples: What It Looks Like When You Study
Examples of metacognition include predicting your score before a practice test, stopping to summarize a section from memory, and sorting test mistakes by cause. The table shows the same situations handled with weak and strong metacognitive habits.
| Situation | Weak metacognition | Strong metacognition |
|---|---|---|
| Reading a textbook chapter | Reads straight through, highlights a lot, feels done | Stops after each section to summarize from memory, marks what didn't come back |
| Preparing for an exam | Rereads notes until they feel familiar | Takes a practice test, predicts the score first, studies the gap |
| Watching a lecture video | Plays it at 2x and nods along | Pauses before each new topic to guess what comes next, notes confusion with a timestamp |
| Using ChatGPT for homework | Asks for the answer, copies the reasoning | Attempts the problem first, then asks the AI to find the flaw in the attempt |
| Getting a test back | Looks at the grade, files it away | Sorts every mistake by cause (careless, conceptual, never studied) and changes the plan |
| Studying in a group | Lets the strongest member explain everything | Explains a topic aloud and notices where the explanation breaks down |
None of the strong versions takes more talent, only a minute or two and a pause to check whether the task is working. That pause is metacognition made visible.
Then there's the example everyone knows. You finish a page and realize you have no idea what it said. Your eyes moved; your mind was elsewhere. Catching that is metacognitive monitoring, and it's a skill: the sooner you notice, the less you have to reread.
Why Metacognition Matters: What the Evidence Shows
Metacognition matters because it predicts achievement beyond intelligence and, in classroom trials, responds to teaching.
It predicts achievement on its own. In 2018, Ohtani and Hisasaka pooled 149 samples from 118 articles on metacognition, intelligence, and academic performance, and metacognition still predicted performance after intelligence was controlled for. Marcel Veenman and colleagues had put rough numbers on the overlap in 2006: averaged across their own studies of learners of different ages, tasks, and subjects, metacognitive skill uniquely explained about 17% of the variance in learning outcomes, intelligence uniquely explained about 10%, and the two shared another 20%. Being smart and managing your own learning well are related but separate skills, and the second adds something the first doesn't.
Teaching it raises results. The Education Endowment Foundation (EEF), which reviews education research for schools in England, sums up "metacognition and self-regulation" as "high impact for very low cost based on extensive evidence," worth an average of about eight months of additional progress over a year. Meta-analyses of classroom interventions point the same way. Charlotte Dignath and Gerhard Büttner (2008) found average effects on academic performance of 0.61 in primary school and 0.54 in secondary school from self-regulated learning training. A 2018 meta-analysis of 48 interventions by Hester de Boer and colleagues found that the benefits of metacognitive strategy instruction held up and grew slightly at follow-up (g = 0.50 right after training, 0.63 an average of about five months later). The interventions themselves are ordinary: plan before a task, check understanding during it, review what worked afterward.
One caution: these are averages across many classrooms. They show that metacognitive training works in general, not that one worksheet will add eight months to your grade.
Why Your Self-Assessment Is Often Wrong
Metacognition only helps if your monitoring is accurate. Unfortunately, the research on calibration (how well your confidence matches your performance) is humbling.
The best-known finding is the Dunning-Kruger effect. In their 1999 paper, Justin Kruger and David Dunning reported that participants scoring in the bottom quartile on tests of humor, grammar, and logic "grossly overestimated" themselves: their scores put them in the 12th percentile, but they estimated they were in the 62nd. Researchers still debate how much of that pattern is a statistical artifact. A similar pattern shows up in real college courses.
In a study published in 2000, Douglas Hacker and colleagues followed students through three exams in a real college course, asking them to predict their scores before each test and estimate them afterward. The strongest students were accurate and got more accurate over the semester. The lowest scorers were grossly overconfident, both before and after taking the exams.
Confidence matters because it often decides when you stop studying. In 2012, John Dunlosky and Katherine Rawson showed what that costs. Students practiced recalling key-term definitions and graded their own answers, and a computer dropped each definition from practice once they'd judged it correct three times. Students who gave themselves credit for wrong answers retired definitions before they'd learned them, and they remembered less on a test two days later. The title of their paper sums it up: "Overconfidence produces underachievement."
Monitoring does get more accurate under the right conditions. In 1991, Thomas Nelson and John Dunlosky had students rate how likely they were to recall each word pair they studied, either right away or after a short delay. Immediate ratings sorted remembered pairs from forgotten ones only modestly (a gamma correlation of +.38), while delayed ratings were far more accurate (+.90), even though recall itself was nearly identical (about 46% in each condition). A 2011 meta-analysis of 112 effect sizes by Rhodes and Tauber confirmed that this delayed-JOL effect is large and reliable. Nelson and Dunlosky's explanation: right after study, the answer is still in short-term memory, so it feels known; after a delay, only what you actually learned is left to judge.
The other big source of miscalibration is fluency, which is why rereading and heavy highlighting feel so productive. We cover that trap in depth in the illusion of competence, including the survey in which most students at one university named rereading as their main strategy.
Nine Metacognitive Strategies for Students
Metacognitive strategies are techniques for planning, monitoring, and evaluating your own learning. The nine below follow that cycle. You don't need all of them: pick one from each stage and run them for two weeks.
Planning
1. Write a 3-line study plan before you open the material. What's the goal of this session? Which resources will you use? How will you know you've learned it? Patricia Chen and colleagues tested a version of this in a 2017 study in Psychological Science. Before exams in an introductory statistics course, students in the intervention group completed a brief self-administered exercise about which resources they'd use to prepare, why each would help, and how they'd use them. Across two studies, their final grades came out 3.45 and 4.65 percentage points higher than the control group's, about a third of a letter grade.
2. Predict your score, then test. Before any self-test or practice exam, write down the score you expect. Afterward, compare. The gap between prediction and result is your calibration error, and tracking it over a few weeks shows you whether your confidence can be trusted. Many people find, at first, that it can't.
3. Activate what you already know. Before reading a chapter, spend two minutes writing what you already know about the topic and one question you hope it answers. This gives your monitoring a baseline. When the chapter contradicts or extends your list, you notice it.
Monitoring
4. Judge your learning after a delay, from the cue alone. Don't rate how well you know something while it's still on the screen. Close the material, wait at least a few minutes, then look only at the heading or question and try to produce the answer before you rate yourself. This is the delayed-JOL effect turned into a habit.
5. Use self-testing as a measuring tool. Retrieval practice is usually sold as a memory technique, and it is one. It's also the best diagnostic you have. A blank-page recall (the blurting method is a structured version) shows you exactly what you can produce without cues, which is what exams and real work demand.
6. Explain it to yourself, out loud or in writing. Self-explanation means stopping to explain why each step or claim makes sense. Michelene Chi's 1989 research found that the students who learned most from worked physics examples generated more explanations linking each step to the underlying principles. In a small 1994 follow-up with 24 eighth graders, those prompted to explain a passage on the circulatory system to themselves gained more understanding than peers who read it twice. A 2018 meta-analysis of 64 research reports by Bisra and colleagues found a moderate overall benefit (g = 0.55). Check your explanations against the source, though: about a quarter of the explanations in Chi's 1994 study were wrong. When your explanation stalls, you've found a gap.
Evaluating
7. Use an exam wrapper. When a graded test comes back, answer three questions before you look at anything else: How did I prepare? What kinds of mistakes did I make? What will I do differently next time? Marsha Lovett at Carnegie Mellon popularized these short reflection sheets. Controlled studies of their effect on grades are mixed (one controlled classroom study found no significant benefit), so treat a wrapper as the evaluation step that feeds your next plan.
8. Keep an error log. Write down every mistake from practice problems along with its cause: careless slip, missing concept, misread question, never studied. After a few weeks, patterns appear. If half your errors are misread questions, the fix isn't more content review.
9. Teach it to someone else. Preparing to teach forces you to organize the material and exposes the parts you can't explain. The research on the protégé effect finds the benefit real but modest, far below the viral "90% retention" claim, and in one experiment it vanished when people taught with their notes open. Explain from memory, and treat every question you can't answer as a gap.
All Nine at a Glance
| Strategy | Stage | What it trains | Evidence strength | Time cost |
|---|---|---|---|---|
| 3-line study plan | Planning | Goal setting, strategy choice | Promising (Chen et al., 2017) | 5 to 15 min |
| Predict, then test | Planning and evaluating | Calibration | Indirect; in one course the lowest scorers were grossly overconfident (Hacker et al., 2000) | 1 min |
| Prior-knowledge dump | Planning | Baseline for monitoring | Limited direct evidence | 2 min |
| Delayed judgments of learning | Monitoring | Telling known from unknown items | Strong for relative accuracy (g = 0.93) | None |
| Self-testing as diagnosis | Monitoring | Knowing what you can produce | Very strong for memory (testing effect); indirect as a diagnostic | 10 to 20 min |
| Self-explanation | Monitoring | Finding comprehension gaps | Moderate for learning (g = 0.55) | Slows reading noticeably |
| Exam wrapper | Evaluating | Linking results to process | Mixed for grades | 10 min per exam |
| Error log | Evaluating | Spotting mistake patterns | Practitioner evidence | 5 min per session |
| Teaching someone | Monitoring and evaluating | Exposing gaps | Modest (protégé effect) | 20+ min |
Metacognitive Questions to Ask Before, During, and After Studying
Metacognitive questions are prompts you ask yourself to plan, monitor, and evaluate your learning. Kimberly Tanner's 2012 guide for biology instructors organized them around that cycle, and the list below follows her structure.
| Stage | Questions to ask yourself |
|---|---|
| Before (planning) | What's my goal for this session? What do I already know about this? Which strategy fits this material? How will I know I've learned it? How much time do I really need? |
| During (monitoring) | Can I explain what I just read without looking? Which part is confusing, specifically? Is this strategy working, or am I just going through the motions? Should I slow down, switch strategies, or ask for help? |
| After (evaluating) | What can I now do that I couldn't before? Which predictions were wrong? What would I change next time? What's still unclear, and when will I come back to it? |
Write the answers down, because a question answered silently is easy to wave away, and keep the routine short. Three questions before, one check during, and three after is plenty; a 20-minute ritual won't survive a week.
Metacognition and AI: Better Answers, No Better Self-Knowledge
Generative AI raises a new metacognitive problem. It can make your output better without making you any better at judging what you've learned, and it can hide what it costs you. Three recent studies point the same way.
Metacognitive laziness. In a randomized study published in the British Journal of Educational Technology (2025), Yizhou Fan and colleagues had 117 university students work on an essay with help from ChatGPT, a human expert, a writing-analytics toolkit, or no extra tool. The ChatGPT group improved their essay scores the most, but they showed no advantage in knowledge gain or transfer. The authors' name for the risk is "metacognitive laziness": learners hand the planning and monitoring work to the tool.
Better scores, no better self-knowledge. In a study published in Computers in Human Behavior (2026), Fernandes and colleagues had people answer 20 logical reasoning questions from the LSAT, with or without ChatGPT's help. The AI group scored about three and a half questions higher, yet both groups overestimated their own scores by about four questions. The AI raised performance without making people any better at judging it, and people with higher AI-literacy scores were less accurate about their own performance, not more.
Practice gains that don't transfer. In Hamsa Bastani and colleagues' trial with about a thousand high school math students in Turkey (PNAS, 2025), those who practiced with a plain GPT-4 chatbot did 48% better than students without AI on practice problems, then 17% worse on a later exam taken without it, and they didn't perceive the harm. We unpack that study, and the hint-giving tutor that avoided the harm, in desirable difficulties.
The fix is to change the order: your own attempt first, the AI second. A few rules help:
- Attempt first, then ask. Write your answer, outline, or explanation before you open the chatbot. Then ask it to critique your attempt rather than produce its own.
- Ask for questions, not answers. "Quiz me on this chapter, one question at a time, and don't tell me the answer until I try" keeps the retrieval work with you.
- Re-test without the tool. If you used AI to understand something, check a day later whether you can explain it with the AI closed.
A Metacognitive Study Loop Built on Your Highlights
Highlighting on its own is a weak study strategy, because marking a sentence takes almost no thought. But highlights are also a record of your attention, which makes them good raw material for monitoring. Here's a loop that works with Glasp's web highlighter or any tool that keeps highlights and notes together.
Step 1: Highlight with a confidence code. Glasp gives you four highlight colors; let them record how well you understand each passage. For example: pink, the default, for key ideas; green for ideas you could already explain; yellow for things you don't understand yet; blue for links to something else you've read.
Step 2: Write one note per highlight, in your own words. Say why the passage matters (self-explanation in miniature), then turn it into a question the passage answers. If you can't do either, recolor the highlight yellow. That's a monitoring signal you'd otherwise miss.
Step 3: Wait, predict, then test. Two days later, read only the questions in your notes. Rate your confidence in each answer from 1 to 5, then answer it and check the highlight. Confident misses are your calibration errors, and anything you can't reproduce goes back to yellow.
Step 4: Quiz yourself across everything you've saved. Glasp's AI chat answers from your own highlights and notes and shows the saved passages it pulled up. Ask it to quiz you on a topic one question at a time, and answer each one before you open the sources it lists. You do the recalling; the chat only finds the material.
Step 5: Calibrate against other readers. On a page other Glasp users have read, the Community tab in the extension's panel lists the passages two or more of them highlighted. A passage several people marked and you skipped is worth a second look, an outside check your own monitoring can't provide. To see what other readers are highlighting beyond this page, browse Glasp's community.
The same loop works for video and books. YouTube Summary lets you highlight lines of a lecture transcript with the timestamp attached, so you can pause, predict what the speaker will say next, and mark what didn't land. On a Kindle, import your Kindle highlights and run the same delayed self-test on a book you finished last month. To turn highlights into questions for spaced review, see our guide to active recall.
Frequently Asked Questions
What is metacognition in simple terms?
Metacognition is knowing what you know and what you don't, and using that knowledge to manage how you learn. It includes noticing when you're confused, choosing a better strategy, and checking whether your studying actually worked. The psychologist John Flavell is usually credited with coining the term in the 1970s, and "thinking about thinking" is the usual shorthand.
What's the difference between cognition and metacognition?
Cognition is the thinking itself: reading, solving a problem, remembering a fact. Metacognition watches and steers that thinking. Summarizing a chapter is cognition; noticing that your summary skipped the hardest section, and going back for it, is metacognition.
What are the three stages of metacognition?
The most common framework has three stages: planning (setting a goal and choosing a strategy before you start), monitoring (checking your understanding while you work), and evaluating (judging the result and the process afterward). Researchers also separate metacognitive knowledge into declarative, procedural, and conditional knowledge, which describe what you know about yourself as a learner and about strategies, how to use them, and when each one fits.
What are some examples of metacognitive strategies?
Common examples include writing a short study plan, predicting your test score before checking it, summarizing a section from memory before moving on, explaining a concept aloud, rating your confidence after a delay, keeping an error log, and filling in a reflection sheet after an exam.
Can metacognition be taught or improved?
Yes. Metacognitive skills respond well to instruction. Meta-analyses of classroom interventions show gains in achievement, and the Education Endowment Foundation estimates about eight months of additional progress from metacognition and self-regulation approaches. Adults can practice the same habits: predicting their scores, judging their learning after a delay, and reflecting afterward.
Is metacognition the same as critical thinking?
No, though they overlap. Critical thinking is about evaluating claims and arguments, usually someone else's. Metacognition is about evaluating your own thinking: whether you understand, whether your strategy is working, and how confident you should be. Good critical thinkers usually have strong metacognition, because spotting the weak point in your own reasoning is a metacognitive act.
What's the difference between metacognition and self-regulated learning?
Self-regulated learning is the broader idea. It includes metacognition plus motivation, emotion, time management, and choosing your study environment. Metacognition is the monitoring and control engine inside it. You can think of self-regulated learning as the whole car and metacognition as the steering.
Does using AI hurt metacognition?
It can. In recent studies, students who used ChatGPT on an essay improved their scores without learning more than other groups, people solving reasoning problems with AI scored higher but judged their own performance no better than people without it, and high school students who practiced math with a plain chatbot did worse on a later exam without it and didn't notice the harm. The risk is letting the tool do your planning and checking. Attempt the task before you open the AI, and re-test yourself later with it closed.
The Skill Behind Every Other Study Skill
Every study technique you've heard of depends on metacognition to work. Spaced repetition needs you to know what you've forgotten. Active recall needs you to notice what didn't come back. Even choosing between techniques is a metacognitive decision. Without accurate self-monitoring, you'll keep doing whatever feels productive, and feeling productive is not the same as learning.
Try it on the next thing you read. Highlight it with Glasp, write down how much you expect to remember, and test yourself two days later with the page closed. The gap between the two numbers is your calibration error, and shrinking it is what metacognition is for.
