A learner checks a box beside a sentence about planning.
Another asks whether they know when a strategy is useful.
A third asks about monitoring understanding while working.
By the end, the page contains a portrait of how the person describes their own metacognitive habits.
That description can be valuable.
It is not the same thing as watching those habits operate in a task.
The Metacognitive Awareness Inventory, commonly abbreviated MAI, was developed by Gregory Schraw and Rayne Sperling Dennison and published in 1994. The original article describes a 52-item inventory designed to assess adults' metacognitive awareness. Items were organised into eight subcomponents under two broad categories: knowledge of cognition and regulation of cognition.
The instrument has since been used and adapted across educational and professional populations, translated into multiple languages and modified into different response formats.
Its longevity makes it important. Its popularity makes its limitations equally important.
What the MAI was designed to assess
The MAI belongs to a family of self-report measures.
It asks respondents to report beliefs about their own metacognitive knowledge and regulatory behaviour. This differs fundamentally from task-based measures that compare confidence with accuracy, track judgments of learning against later recall, or estimate metacognitive sensitivity from trial-level performance.
The original framework has two broad dimensions.
Knowledge of cognition concerns what a person reports knowing about their cognitive processes and strategies.
Regulation of cognition concerns what a person reports doing to plan, monitor, manage information, correct errors and evaluate learning.
Those broad dimensions contain eight commonly described subcomponents: declarative knowledge, procedural knowledge, conditional knowledge, planning, information-management strategies, comprehension monitoring, debugging strategies and evaluation.
These categories make the inventory attractive for educational reflection because they give language to several parts of metacognitive activity rather than reducing awareness to one question.
A 52-item inventory is not a 52-item microscope
The MAI's items can reveal patterns in self-description.
They cannot directly observe cognition.
A respondent may endorse a statement because it is aspirational, because it sounds like what a good learner should do, because they remember a few salient examples, or because they genuinely and accurately recognise a stable habit.
The score alone cannot distinguish those possibilities.
This is not a defect unique to the MAI. It is a general property of self-report measurement.
The methodological question is therefore not “Is self-report subjective?” It obviously is. The question is whether the self-report has sufficient validity for the particular inference being made.
The original validation established a useful starting structure
Schraw and Dennison's 1994 paper reports two experiments supporting a two-factor model corresponding to knowledge of cognition and regulation of cognition. The factors showed high internal consistency in the original work and were moderately intercorrelated.
The study also found relationships between aspects of the inventory and other measures, including reading comprehension and pretest judgments of monitoring ability.
Importantly, the original article did not establish that the MAI directly measures online monitoring accuracy. The knowledge-of-cognition factor was reported as unrelated to monitoring accuracy in one experiment.
That finding is conceptually useful.
People's beliefs about their metacognitive habits and their actual calibration on a specific task are not guaranteed to coincide.
Internal consistency is not validity
One of the most common errors in questionnaire use is to report Cronbach's alpha and conclude that the instrument is valid.
Internal consistency addresses how coherently items behave together under certain assumptions. It does not establish that the scale measures the intended construct, that its factor structure generalises, that respondents interpret items as intended, or that scores predict meaningful outcomes.
A questionnaire can be highly internally consistent and still be weak for a particular purpose.
This distinction became especially clear in a 2025 psychometric review and meta-analysis of the MAI in health professions education by Andrew Cale and colleagues.
Across 1,966 records screened, 23 studies met inclusion criteria for the validity review. The authors found substantial internal-consistency evidence, with aggregated coefficients roughly in the 0.805 to 0.844 range for the original 52-item, five-point Likert version in the studies suitable for meta-analysis.
At the same time, important forms of validity evidence were missing.
The 2025 review found support and gaps at once
Cale and colleagues evaluated evidence across several validity domains.
The five-point Likert version had the strongest support among the response formats examined for test content, internal structure and relations to other variables. Yet the review found no evidence in the included literature for response processes or consequences of testing.
That is a meaningful gap.
Response-process evidence asks whether people interpret items and use the response scale in the way the instrument assumes. Consequences-of-testing evidence asks what follows from using scores in practice.
These questions become increasingly important when an inventory moves from research description toward educational decision-making.
A scale can be reliable enough for group research while still being poorly justified for individual diagnosis, selection or high-stakes intervention.
Factor structure has not been perfectly stable
The MAI is often described as though its two broad factors and eight subcomponents are psychometrically settled everywhere.
Later research is more complicated.
Systematic psychometric reviews have found mixed evidence regarding detailed factor structure, and different studies have used shortened versions, altered response scales and translated forms. Some analyses support broad knowledge and regulation factors more readily than the exact eight-component architecture. Others report alternative structures.
This does not make the MAI unusable.
It means that researchers should not assume a published subscale structure has automatically transferred to a new population.
If a study intends to interpret specific subscale scores, it needs evidence that those scores are meaningful in that sample and version.
Response format changes the instrument
The MAI has appeared in dichotomous, five-point Likert and sliding analogue forms, among others.
These are not cosmetically identical.
A dichotomous response forces a categorical judgment. A Likert scale captures degree. A continuous slider may invite still finer distinctions but can also change how people use endpoints and middle values.
Cale and colleagues' 2025 review found the five-point Likert version most supported in the health-professions literature they examined, while also emphasising remaining validity gaps.
That conclusion is context-specific.
It does not prove that one response format is universally best in every language, age group or purpose.
Researchers should identify the exact version used rather than citing the MAI as though all variants were the same test.
Translation requires more than replacing words
The MAI has been translated and adapted in many countries.
Cross-language use creates at least three problems.
First, metacognitive vocabulary may not map cleanly across languages. Words for monitoring, awareness, strategy and evaluation carry educational and cultural assumptions.
Second, response styles differ across populations. Some groups use extreme response options more readily than others.
Third, educational practices change what an item means. A statement about planning study may function differently where curricula, assessment and learner autonomy differ substantially.
Translation should therefore include psychometric validation rather than relying only on linguistic equivalence.
For Swedish use, the Institute's terminology-control rules require review before indexing or publication.
Self-report awareness is not metacognitive sensitivity
This boundary is essential.
Task-based metacognitive measures ask how confidence relates to correctness on specific trials. Meta-d-prime estimates second-order sensitivity relative to first-order task sensitivity. Calibration compares confidence with observed outcomes.
The MAI asks respondents to describe their own knowledge and regulation tendencies.
A person can score highly on the MAI while showing poor calibration in a laboratory task. Another may show strong task-based metacognitive sensitivity while endorsing fewer general self-regulation statements.
Neither result automatically invalidates the other.
They may be measuring different levels of the metacognitive system.
The error is to call all of them “metacognition” and expect convergence without specifying construct and method.
The inventory can be useful for reflection without becoming diagnosis
In education, the MAI can help structure conversations.
Items can prompt learners to consider whether they plan, monitor comprehension, change strategies when confused or evaluate outcomes after study. Aggregated responses can support research on perceived metacognitive awareness.
Used formatively, the instrument may provide language for reflection.
But formative usefulness is not the same as psychometric diagnosis.
A student should not be labelled “low metacognition” solely because of one questionnaire score. Nor should a programme claim that an intervention improved actual metacognitive accuracy simply because self-reported MAI scores increased.
If the desired outcome is monitoring accuracy, measure monitoring accuracy.
If the desired outcome is self-reported strategy awareness, the MAI may be closer to the construct.
Pre-post change requires special caution
Questionnaires are often used before and after a course.
A higher post-course score can have several meanings. Learners may actually regulate better. They may have learned the vocabulary of regulation and become more willing to endorse items. They may understand the questions differently. They may have become more self-aware and, paradoxically, rate themselves lower because they now recognise deficits they previously missed.
This phenomenon is not unique to the MAI. It is a general challenge in self-assessment.
Therefore, intervention studies should ideally combine self-report with behavioural or performance-based evidence when the claim concerns actual regulatory change.
A score change is evidence of a score change first.
Its psychological interpretation requires a validity argument.
Licensing and access should be checked before operational use
The MAI is widely reproduced online, often without clear provenance.
Researchers and institutions should consult the original publication and authorised or properly validated versions before reproducing, modifying or administering the instrument. Any licensing, copyright or permission conditions should be checked for the intended context.
The Institute should not host or reproduce a full instrument merely because copies circulate publicly.
This is both an intellectual-property boundary and a methodological one: uncontrolled versions can silently change wording, response formats or scoring.
An instrument with altered items is not automatically the same instrument.
The MAI can be misused in organisational settings
Organisations may be tempted to treat metacognitive awareness as a workforce competency and use a questionnaire for selection or performance management.
That would require much stronger evidence than most educational uses.
High-stakes assessment needs demonstrated validity for the target population and consequence, evidence of fairness, appropriate reliability, and clarity about what decisions the score supports.
The 2025 review's absence of evidence about consequences of testing in health-professions studies is a warning against jumping from research use to consequential classification.
A reflective inventory should not become an employee ranking instrument by administrative enthusiasm alone.
AI can improve or distort self-report in new ways
AI assistants can now coach people through self-reflection, explain metacognitive concepts and even help complete questionnaires.
This changes the response environment.
If a participant asks an assistant what an item “really means,” they may answer a clarified construct rather than the original prompt. If a system suggests desirable study strategies before the questionnaire is completed, responses may partly reflect instruction delivered during measurement.
The Institute's synthesis principle is simple: the administration context is part of the instrument.
If AI assistance is permitted, record it. If independent self-report is required, design the administration accordingly.
The MAI literature itself does not yet provide a validated protocol for AI-mediated administration.
Convergent evidence should be expected to be imperfect
If the MAI measures reported metacognitive knowledge and regulation tendencies, perfect agreement with trial-level confidence measures would actually be conceptually surprising. The methods differ in timescale, response mode and target. A questionnaire asks for a generalised self-description; a laboratory measure samples behavior under a constrained task.
The useful question is therefore not whether two measures correlate perfectly, but whether the pattern of convergence and divergence fits a defensible construct model. Weak correlation may reveal measurement error, but it may also reveal that the instruments capture different components.
Researchers should avoid using “convergent validity” as a ritual requirement divorced from theory. The comparison measure must itself be appropriate to the aspect of metacognition being claimed.
What the evidence does not show
The MAI does not directly measure moment-to-moment metacognitive accuracy.
A high score does not prove that a person is well calibrated on a specific task.
High internal consistency does not establish complete construct validity.
The original two-factor framework does not guarantee identical factor structure across every population, translation or modified version.
A pre-post score increase does not by itself establish improved learning outcomes or improved metacognitive sensitivity.
Validity evidence from health professions education does not automatically generalise to schools, workplaces or clinical diagnosis.
And widespread use is not evidence that every use is justified.
Interpretation checklist
Identify the exact version, language and response scale.
State whether the intended inference concerns knowledge of cognition, regulation of cognition, specific subscales or an overall score.
Report internal consistency, but do not stop there.
Examine factor structure where subscale interpretation matters.
Seek evidence on relations to other variables and, where relevant, response processes.
Avoid using MAI scores as a proxy for task-based calibration without direct evidence.
For intervention studies, combine self-report with behavioral or outcome measures when possible.
For high-stakes use, require a separate validity argument addressing consequences and fairness.
A questionnaire can open a door without becoming a mirror
The value of the MAI is not that it reveals a hidden true level of metacognition.
Its value is that it gives structured access to how people report their own cognitive knowledge and regulatory habits.
That is useful evidence.
It is evidence of a particular kind.
A mature science of metacognition gets stronger when it stops demanding that every instrument tell the whole story.
Research references
Schraw G, Dennison RS. Assessing Metacognitive Awareness. Contemporary Educational Psychology. 1994;19(4):460–475. https://doi.org/10.1006/ceps.1994.1033
Cale AS, Agosto ER, Anak Ganeng BK, Kruskie ME, McNulty MA, Robertson KA, et al. Meta2: A meta-analysis and psychometric evaluation of the Metacognitive Awareness Inventory (MAI) in the context of health professions education. Anatomical Sciences Education. 2025;18(8):786–801. https://doi.org/10.1002/ase.70085
Craig K, Hale D, Grainger C, Stewart ME. Evaluating metacognitive self-reports: systematic reviews of the value of self-report in metacognitive research. Metacognition and Learning. 2020;15:155–213. https://doi.org/10.1007/s11409-020-09222-y
Rahnev D. A comprehensive assessment of current methods for measuring metacognition. Nature Communications. 2025;16:701. https://doi.org/10.1038/s41467-025-56117-0