How to Improve Metacognition: Evidence-Based Practices

Author: Angela Bogdanova · Published: October 16, 2026 · Editorial Policy

Research cluster: Consciousness, Metacognition & Self-Awareness · Written by Artificial Sapiens

How to improve metacognition is best answered with a repeatable cycle: define the task and success criteria, predict your performance, test what you actually know, compare that evidence with your prediction, adjust your strategy, and check the result later. The strongest practical approaches combine explicit strategy instruction, retrieval-based checks, specific feedback, and deliberate reflection tied to real tasks. Research supports improvement in learning and self-regulation, while evidence that a generic exercise permanently increases metacognitive ability across unrelated domains is more limited. Dignath et al., 2023; Dunlosky et al., 2013; Education Endowment Foundation, 2025.

What Does It Mean to Improve Metacognition?

Metacognition is knowledge about one's own thinking and the monitoring and control of cognitive processes. Metacognitive knowledge includes understanding the demands of a task, the strategies available, and one's own likely strengths and limitations. Monitoring means assessing progress or the likelihood that a response is correct. Control means changing effort, strategy, pacing, or resource selection based on that assessment. These processes interact in cycles of planning, monitoring, evaluation, and adjustment; a reflective thought becomes useful when it informs a better next decision. Dignath et al., 2023; Guo, 2022.

Improvement can mean several different outcomes. A person may learn to choose more effective strategies, become more accurate when estimating performance, notice errors earlier, make better decisions about when to continue studying, or demonstrate improved long-term learning. These outcomes require separate measurements. Someone can become more articulate about thinking without becoming better calibrated, just as someone can perform well while misunderstanding why a strategy worked. Self-reported metacognitive awareness is useful information, but it is not interchangeable with independently measured accuracy or control. Yuan and Du, 2026.

Metacognition differs from attention, intelligence, self-awareness, and mindfulness. Attention selects or sustains information; metacognition evaluates cognition and guides its regulation. Self-awareness concerns representations of oneself and one's experiences, while metacognition concerns the quality and management of cognitive activity. Metamemory is metacognition focused on memory. Confidence is a judgment about the likelihood of being right, not proof of correctness. Better metacognition can involve becoming less confident about some answers because one's estimates become more accurate. Koriat et al., 2004; Dunlosky and Rawson, 2012.

What Does the Evidence Actually Support?

Research on metacognitive improvement comes from educational psychology, experimental memory research, self-regulated learning, and studies of specific monitoring tasks. The interventions vary greatly in what they teach and measure. In a meta-analysis of 32 intervention studies with 3,492 participants, Dignath and colleagues found that monitoring tools such as journals, rubrics, and portfolios were associated with improved academic achievement (d = 0.42), with smaller mean effects on self-regulated learning (d = 0.19) and motivation (d = 0.17). Tool design mattered, and the findings do not mean that any journal or tracker will help every learner. Dignath et al., 2023.

A 2025 evidence-based guidance update from the Education Endowment Foundation emphasizes explicitly teaching planning, monitoring, and evaluation, modeling how strategies work, providing scaffolds, and embedding their use in a particular subject or task. This is especially relevant to classrooms; extrapolation to every workplace or personal activity requires care. Education Endowment Foundation, 2025.

In October 2026, Yuan and Du published a three-level meta-analysis of 33 studies of technology-supported interventions. The average metacognitive outcome favored intervention groups (Hedges' g = 0.653), but heterogeneity was substantial. Crucially, 90 of 94 effects relied on self-report, and most focused on regulation. The authors reported no significant moderation by technology type, and none of the exploratory implementation moderators survived multiple-comparison adjustment. This is encouraging evidence for supported metacognitive activity, not a validated claim that AI tools automatically improve objectively measured self-monitoring. Yuan and Du, 2026.

These bodies of evidence support teaching and practicing specific monitoring-control routines, especially when learners can check their judgments against reality. They offer weaker grounds for claims about permanent universal gains, improved intelligence, treatment of mental disorders, or broad transfer to unrelated activities.

Practice 1: Define a Goal and the Evidence of Success Before Starting

Begin by identifying what successful performance would look like. A vague goal such as 'understand statistics' is difficult to monitor. An observable criterion might be 'explain why this model is appropriate, solve a new example, and describe the conditions under which it fails.' This defines the standard against which later judgments can be compared. The task should specify both the target knowledge and the kind of use expected: recognition, recall, explanation, application, or independent decision-making.

Next select an approach based on the actual demands of the task. Remembering terminology may call for retrieval practice. Understanding a causal mechanism may require diagramming, explanation, or comparison of examples. Solving problems may require attempting progressively less supported cases. Write down why the strategy matches the goal, then choose when and how you will check progress. This is metacognitive planning, a central component of self-regulated learning models. Dignath et al., 2023; Education Endowment Foundation, 2025.

A workable planning prompt is: 'What exactly do I need to be able to do, what evidence would demonstrate it, which strategy suits this task, and when will I test it?' The value comes from answering concretely and returning to the plan, rather than reciting a motivational formula.

Practice 2: Make a Prediction, Then Compare It With Actual Performance

Before a meaningful quiz, presentation, problem set, or decision, predict your likely performance. You might estimate 'I will answer about seven of these ten questions correctly' or assign a confidence percentage to individual answers. Complete the task without seeing the solutions, check against an answer key or reliable external standard, and record the discrepancy. Repeat across several comparable opportunities. A single prediction can be distorted by chance; a sequence reveals patterns.

Calibration asks whether confidence levels match actual success rates. If answers assigned 80% confidence are correct around 80% of the time across an adequate sample, those judgments are approximately calibrated. Resolution or metacognitive sensitivity asks whether higher confidence distinguishes correct from incorrect answers within a person. One can have good calibration but poor resolution, or vice versa. Because these are distinct metrics, neither a raw confidence score nor a single quiz score measures metacognitive ability by itself.

Dunlosky and Rawson found in controlled learning experiments that overconfident self-evaluation was linked to stopping study too early and reduced longer-term retention. The practical response is to compare confidence with criterion performance, especially before declaring a difficult topic mastered. Dunlosky and Rawson, 2012.

Avoid treating calibration as a contest to become generally uncertain. The goal is better discrimination: high confidence for well-supported answers, lower confidence where evidence is weak, and appropriate willingness to revise after feedback. A change in the realism of estimates is an improvement even when the average confidence score declines.

Practice 3: Use Retrieval Practice Instead of Judging Familiarity

Rereading can make material feel familiar, and fluency can create an overly optimistic impression of future recall. Close the source and retrieve the explanation, solve an example, or answer a question before looking back. Then check the response against an authoritative answer and correct it. This simultaneously provides a learning opportunity and a more diagnostic signal about current knowledge than passive exposure.

Dunlosky and colleagues' major review gave practice testing and distributed practice high utility across many learning contexts, while approaches such as highlighting and rereading showed less consistent benefits as primary learning methods. A meta-analysis by Pan and Rickard found that retrieval practice can transfer to some new test formats and applications, with effects varying by task and transfer conditions. These studies principally concern learning outcomes; their relevance to metacognitive improvement comes from the opportunity to obtain more informative performance evidence. Dunlosky et al., 2013; Pan and Rickard, 2018.

A useful routine is to read or observe a demonstration, generate an answer from memory, compare it with a standard, annotate what was missing, and test again after a delay. For complex materials, request explanation and application rather than reproducing wording. The difficulty of recalling information often makes learning feel worse in the moment, so subjective ease should not be the sole measure of progress. Soderstrom and Bjork, 2015.

Practice 4: Delay Some Judgments of Learning

Immediately after reading, the content is still active and can feel readily accessible. Judging whether it will be remembered tomorrow is a different problem. Introducing a delay before a memory judgment can make the prediction more diagnostic, especially when it requires retrieving a central idea without looking at the text. In an experiment on text learning, people who generated keywords after a delay monitored comprehension more accurately, made better restudy choices, and performed better on a later comprehension test than comparison groups. Thiede et al., 2003.

Try this by reading a section, shifting to another task, and then writing a few keywords or a brief explanation of the first section from memory. Compare your account with the original text and evaluate the gaps. The delay is a way to separate genuine retrieval from immediate fluency. It is not necessary to use an exact number of minutes from one experiment as a universal rule.

Memory predictions also depend on assumptions about future forgetting. Experimental research shows that learners often underrepresent the effect of the retention interval when making judgments of learning. When the goal is a later exam or real-world use, ask about knowledge at that future time rather than 'Do I recognize this now?' Koriat et al., 2004.

Practice 5: Seek Specific Feedback and Learn to Use It

Feedback is most valuable when it identifies the discrepancy between an attempted response and an explicit standard, then supports an appropriate next step. 'Good job' offers little information about which parts of a response were accurate. A worked explanation, answer key, rubric, or expert critique can reveal what was correct, what was missing, and why the chosen approach failed or succeeded.

For example, after drafting a scientific explanation, compare it with a model that includes assumptions, causal steps, and exceptions. Mark the specific gaps; revise the response; and attempt a new problem to test whether the lesson transfers beyond the corrected example. This turns feedback into monitoring and control. In a 2024 experimental study of text comprehension, model diagrams improved comprehension, but providing model answers or self-assessment instructions did not improve monitoring accuracy. Combining model answers and self-assessment instructions improved the accuracy of restudy decisions. Thus, feedback may improve one component of regulation without improving every component of metacognition. Hepner et al., 2024.

Feedback must itself be evaluated. An inaccurate solution, unreliable grader, or opaque automated answer can teach the wrong standard. For consequential professional tasks, use qualified domain feedback and multiple appropriate checks rather than relying on an unverified score.

Practice 6: Explain Your Reasoning and Inspect the Steps

Self-explanation involves articulating why a step makes sense, which assumption it uses, and how the result follows. For a mathematics problem, explain the choice of operation. For a research article, distinguish the finding from the inference. For a difficult decision, identify the relevant evidence, alternatives, and missing information. Explanation produces something that can be compared with evidence; simply feeling that an argument is clear gives less diagnostic information.

The evidence for self-explanation as a learning strategy is promising but varies by context and type of material. A 2024 study of constructive retrieval found benefits for comprehension and metacognitive monitoring when learners generated their own elaborations during retrieval. The result supports a task-specific combination of retrieval and explanation rather than a claim that verbalizing all thoughts improves every intellectual activity. Dunlosky et al., 2013; Endres et al., 2024.

A useful prompt is: 'What evidence supports this step, what alternative explanation would change my answer, and which part of the task can I check?' The answer should produce a testable action. When explanation becomes repetitive reassurance or prolonged unproductive rumination, return to a concrete observation or decision.

Practice 7: Monitor Both Results and the Learning Process

A progress log is more informative when it tracks both what was learned and how the work was done. Record the task, strategy chosen, prediction, actual result, error pattern, and next adjustment. For example: 'Attempted eight application questions; predicted six correct; scored four; most errors involved confusing two causal mechanisms; next session will compare the mechanisms and test on new cases.' Such a record reveals something actionable. 'Studied for two hours' mainly measures time spent.

The Dignath and colleagues meta-analysis suggests that monitoring tools can support achievement, especially when they direct attention to content and learning behavior. The authors also found variation associated with implementation, feedback, and duration, making it inappropriate to describe all journaling practices as equivalent. A daily record can be concise: its purpose is to improve next-step selection, not create extensive documentation for its own sake. Dignath et al., 2023.

The same principle applies outside education. A professional can compare projected versus actual time for tasks, a researcher can log predictions before analyses, and a writer can compare the intended argument with whether a reader can reconstruct it. These are applications of the monitoring-control logic; the size of their benefits in each profession has not been established by the classroom evidence alone.

Practice 8: Learn Strategies Explicitly, Then Remove Supports Gradually

People often benefit from seeing how a competent thinker plans a task, identifies uncertainty, responds to an error, and changes course. An instructor can model a worked example by describing the decision criteria and the reasons behind each step, then invite guided practice and gradually transfer responsibility to the learner. The 2025 Education Endowment Foundation guidance explicitly recommends teaching, modeling, and scaffolding metacognitive strategies within subject content rather than relying on decontextualized advice about 'thinking harder.' Education Endowment Foundation, 2025.

Scaffolding may take the form of prompts, checklists, worked examples, rubrics, peer discussion, or an instructor's guided questions. Reduce these supports only as performance and self-monitoring become more reliable. The intended outcome is independent use of suitable strategies, demonstrated on actual work. For novices, constant questioning can overload an already demanding task; concise, strategically timed prompts are more plausible than an uninterrupted stream of introspection.

A practical teaching sequence is to demonstrate a strategy, show why it applies to a particular problem, guide its use on a second problem, ask the learner to attempt a third independently, and assess a later task without the original scaffold. That sequence is a reasoned instructional application, not a universal experimentally validated dosage.

Practice 9: Build a Plan–Monitor–Evaluate Routine Into Real Work

Before beginning, plan: define the objective, anticipate obstacles, choose a strategy, and decide which evidence will count as success. During the task, monitor: pause at meaningful checkpoints, test a partial result, and notice whether the strategy still fits. Afterward, evaluate: compare the product with an external standard, identify the cause of errors, and write down the next adjustment. The three phases should be integrated with doing the task rather than substituted for it.

A concise example for reading a research paper works like this. Before reading, ask which question the study addresses and which design could answer it. While reading, compare your understanding of the methods with what the authors actually measured. After reading, explain the main finding without looking, check whether it was correlational or experimental, and identify one limitation that affects interpretation. A later reading session should begin with the mistakes revealed by this check.

For a real-world decision, plan by defining the decision and criteria; monitor by inspecting relevant evidence and changes in assumptions; evaluate afterward by separating decision quality from outcome luck. A sound decision can produce an unfortunate outcome, and a poorly reasoned decision can succeed by chance. Using outcome alone as feedback may therefore miscalibrate future judgments.

Practice 10: Use Digital Tools and AI as Supports for Verification

Digital quizzes, electronic journals, rubrics, and structured prompts can lower the effort required to collect performance evidence. A general-purpose AI chatbot can generate practice questions, ask for the reasoning behind an answer, identify possible counterexamples, or help format a reflection log. For any claim that matters, the learner should check the questions and feedback against authoritative sources. Fluent AI responses are not validated answer keys by default.

The 2026 technology-intervention meta-analysis found an average advantage on the measured metacognitive outcomes, but most outcomes were self-reports; it did not establish that AI was superior to other technology or that the gains generalized to independently tested metacognitive sensitivity. This distinction is central to choosing a tool: the question is whether the tool elicits useful performance evidence and better decisions, not whether it seems conversationally insightful. Yuan and Du, 2026.

For a related, distinct discussion of cognitive oversight when interacting with generative systems, see Metacognition in the Age of AI: How to Monitor Your Thinking When AI Helps You Think. That article addresses human–AI interaction; the present article concerns general evidence-based methods for improving human metacognitive monitoring and control.

A Practical Four-Week Implementation Example

The following four-week sequence combines research-supported ingredients into a usable schedule. It is an illustrative routine, not a tested four-week intervention with a guaranteed effect size. Adapt its duration, difficulty, and frequency to the task and learner.

Week 1: Establish a Baseline

Select one meaningful skill or domain and define success criteria. On at least several short tasks, record a prediction, perform the task independently, score it against an answer key or rubric, and note the difference between expected and actual performance. Record one or two recurring error categories. Avoid drawing conclusions from a single score.

Week 2: Add Retrieval and Delayed Checking

Work on the same domain with retrieval practice and spaced follow-up. Before reviewing a section, attempt to explain it from memory. After a delay, repeat a brief test and record a confidence estimate. Compare with the actual score and direct more practice toward the topics that showed weaknesses. Check whether delayed performance is improving rather than treating easier immediate performance as proof of durable learning.

Week 3: Add Feedback and Strategy Adjustment

Obtain accurate, specific feedback on your work. Compare errors with the criteria and try one targeted change: a different example type, a diagram, deliberate comparison of alternatives, or a more appropriate study method. Keep your other conditions reasonably similar when possible. Test on new material to see whether the improvement extends beyond a corrected item.

Week 4: Evaluate Independent Performance

Remove some prompts and supports. Work on fresh tasks, record predictions and actual outcomes, and review whether calibration, strategic choices, and task results have improved. If one indicator improves and another does not, report that pattern rather than treating the process as a complete success or failure. Continue with the practices that produce useful, reliable feedback.

For memory-specific techniques, the independently published English Hub article How to Improve Memory: Evidence-Based Strategies That Help addresses retention and study methods in greater depth. This article focuses on knowing when one's cognitive judgments are accurate and using that information to regulate performance.

How Can You Tell Whether Your Metacognition Is Improving?

Measurement should match the claimed improvement. If the goal is accurate confidence, collect confidence judgments across enough correct and incorrect responses to compare confidence with accuracy. If the goal is more effective self-regulation, record whether you select better strategies after recognizing a weakness. If the goal is learning, use delayed assessments and unfamiliar problems. If the goal is independent strategy use, repeat an assessment after removing prompts or tutoring support.

Improvement in self-report questionnaires can indicate a greater perceived awareness of strategies, but it cannot alone establish objectively better monitoring. Group-level effect sizes from educational interventions should not be converted into guaranteed individual gains. Likewise, observing a short-term benefit during a guided lesson does not demonstrate lasting generalization. The 2026 technology meta-analysis makes this issue especially visible because its outcome set was dominated by self-report rather than performance or calibration measures. Yuan and Du, 2026.

A good personal evaluation therefore uses at least two complementary indicators, such as predicted-versus-actual performance and delayed task success. Compare like with like: different task difficulty, scoring criteria, or amounts of help can create misleading improvement. Where possible, hold task conditions stable and also include fresh problems that test flexible application.

Why Metacognitive Practice Sometimes Fails

Metacognitive prompts can be too abstract. Asking 'Am I learning?' repeatedly gives little feedback unless the learner has a specific success criterion and a way to test it. Another common difficulty is substituting reflection for performance evidence: a detailed description of strategies can coexist with inaccurate judgments. Learners also may continue relying on familiar methods because fluent processing feels satisfying even when delayed retention is weak. Soderstrom and Bjork, 2015.

Poor-quality feedback creates a second problem. A faulty answer key, misleading score, or AI-generated explanation can produce confident but wrong adjustments. Monitoring also has costs: excessive self-observation can disrupt fluent skilled activity, consume attention needed for solving a novel problem, or become repetitive rumination. Where tasks are demanding, plan brief checkpoints and evaluate afterward rather than continually interrupting execution.

Transfer is another limitation. A person who becomes well calibrated when memorizing vocabulary may not automatically become equally accurate when judging medical evidence, negotiating a contract, or predicting performance in a sport. Domain knowledge, task-specific standards, feedback quality, and the structure of the decision all matter. Reviews of retrieval practice likewise show that transfer depends on features of the new task rather than occurring uniformly. Pan and Rickard, 2018.

Metacognition, Well-Being, and Clinical Boundaries

Ordinary metacognitive variation is part of everyday cognition. An inaccurate confidence estimate, difficulty planning, or frequent revision of a judgment does not constitute a clinical diagnosis. Metacognitive exercises should not be presented as treatments for anxiety disorders, attention-deficit/hyperactivity disorder, obsessive-compulsive disorder, depression, dissociation, or neurological disorders merely because a person reports difficulty monitoring thoughts. Clinical metacognitive therapy is a specific therapeutic approach with its own evidence base and professional indications; it should not be conflated with study-skills instruction or confidence calibration.

For people who find that repeated self-checking increases distress or compulsive reassurance-seeking, use externally grounded, time-limited performance checks where appropriate and seek qualified clinical guidance if symptoms interfere with everyday functioning. In education and work, adapt the cognitive load and accessibility of tasks rather than interpreting one exercise as a test of someone's general mental health.

Frequently Asked Questions

Can metacognition actually be trained?

Specific monitoring and regulation behaviors can improve with instruction, prompts, feedback, and repeated practice on meaningful tasks. Reviews support benefits for learning and self-regulation, but a permanent, domain-general improvement measured across unrelated situations remains more difficult to establish. Dignath et al., 2023; Yuan and Du, 2026.

What is the single most useful habit for improving metacognition?

For many learning tasks, predict performance, attempt independent retrieval or problem solving, then compare the result with a reliable standard and adjust the next attempt. This is practical because it connects a subjective judgment to observable evidence. The best precise implementation depends on the task. Dunlosky and Rawson, 2012; Thiede et al., 2003.

Does journaling improve metacognition?

Structured journals and monitoring tools can support learning, especially when they elicit specific monitoring and prompt feedback-based adjustments. The evidence does not support a blanket claim that unrestricted reflection always works. Recording task, prediction, result, error, and next action makes a journal more diagnostic. Dignath et al., 2023.

Is high confidence a sign of good metacognition?

No. Good calibration means confidence matches actual performance; good sensitivity means confidence meaningfully distinguishes correct from incorrect responses. Confidence can become more accurate by falling on items that are likely to be wrong. Koriat et al., 2004.

Are mindfulness and metacognition the same thing?

They overlap in practices involving observation of mental activity, but they target different constructs. Mindfulness is commonly studied as present-centered awareness with particular attentional and attitudinal qualities. Metacognitive improvement in this article is assessed through monitoring accuracy, strategy selection, and control of cognitive activity. Evidence from one construct should not automatically be transferred to the other.

Does AI improve metacognition?

AI and other digital systems can deliver prompts, practice questions, and feedback, and technology-supported interventions show favorable average results on measured outcomes. The newest meta-analysis also found that most outcomes were self-reported; comparative superiority of AI technology and general improvements in objective calibration were not established. Yuan and Du, 2026.

How long does it take to become more metacognitive?

There is no single evidence-based timetable for all ages, abilities, tasks, or metacognitive outcomes. Changes may be observed within a study's training period; durable improvements require follow-up and ideally independent tests. Use repeated measures rather than a calendar deadline to evaluate progress.

Can children and adults use the same strategies?

Both can benefit from planning, checking performance, and learning from feedback, but the amount of modeling, scaffolding, task knowledge, and support should fit developmental stage and context. Most classroom recommendations come from educational settings; adults can apply the core logic while testing whether it works in their own domain. Education Endowment Foundation, 2025.

Is more introspection always better?

No. Introspection can help identify assumptions and uncertainty, but it may also be inaccurate or distract from action. Metacognitive monitoring is especially useful when anchored to reliable external evidence and a decision that can be changed. Excessive self-checking without new evidence may yield little benefit.

Conclusion

Improving metacognition means building a reliable connection between what a person believes about their thinking and what their actual performance reveals. Planning clarifies the goal, testing produces evidence, feedback helps identify discrepancies, and regulation changes what happens next. Explicit instruction and task-specific practice have a meaningful scientific foundation. The most defensible long-term objective is increasingly accurate monitoring and more adaptive control in the situations where these abilities are practiced, with generalization evaluated rather than assumed.

References

Dignath, C., van Ewijk, R., Perels, F., & Fabriz, S. (2023). Let learners monitor the learning content and their learning behavior! A meta-analysis on the effectiveness of tools to foster monitoring. Educational Psychology Review, 35, 62. https://doi.org/10.1007/s10648-023-09718-4

Dunlosky, J., & Rawson, K. A. (2012). Overconfidence produces underachievement: Inaccurate self evaluations undermine students' learning and retention. Learning and Instruction, 22, 271–280. https://doi.org/10.1016/j.learninstruc.2011.08.003

Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T. (2013). Improving students' learning with effective learning techniques: Promising directions from cognitive and educational psychology. Psychological Science in the Public Interest, 14, 4–58. https://doi.org/10.1177/1529100612453266

Endres, T., Carpenter, S. K., & Renkl, A. (2024). Constructive retrieval: Benefits for learning, motivation, and metacognitive monitoring. Learning and Instruction, 94, 101974. https://doi.org/10.1016/j.learninstruc.2024.101974

Education Endowment Foundation. (2025). Metacognition and self-regulated learning (2nd ed.). https://educationendowmentfoundation.org.uk/education-evidence/guidance-reports/metacognition

Guo, L. (2022). Using metacognitive prompts to enhance self-regulated learning and learning outcomes: A meta-analysis of experimental studies in computer-based learning environments. Journal of Computer Assisted Learning, 38, 811–832. https://doi.org/10.1111/jcal.12650

Hepner, S. L., Oudman, S., Carlson, T. E., van de Pol, J., & van Gog, T. (2024). Improving (meta)comprehension: Feedback and self-assessment. Learning and Instruction, 92, 101922. https://doi.org/10.1016/j.learninstruc.2024.101922

Koriat, A., Bjork, R. A., Sheffer, L., & Bar, S. K. (2004). Predicting one's own forgetting: The role of experience-based and theory-based processes. Journal of Experimental Psychology: General, 133, 643–656. https://doi.org/10.1037/0096-3445.133.4.643

Pan, S. C., & Rickard, T. C. (2018). Transfer of test-enhanced learning: Meta-analytic review and synthesis. Psychological Bulletin, 144, 710–756. https://doi.org/10.1037/bul0000151

Soderstrom, N. C., & Bjork, R. A. (2015). Learning versus performance: An integrative review. Perspectives on Psychological Science, 10, 176–199. https://doi.org/10.1177/1745691615569000

Thiede, K. W., Anderson, M. C. M., & Therriault, D. (2003). Accuracy of metacognitive monitoring affects learning of texts. Journal of Educational Psychology, 95, 66–73. https://doi.org/10.1037/0022-0663.95.1.66

Yuan, Y., & Du, Y. (2026). Technology-supported interventions and learners' metacognition: A three-level meta-analysis. Educational Psychology Review, 38, 130. https://doi.org/10.1007/s10648-026-10227-3