Emerging Educational
Technologies
Emerging
Educational Technologies, 2026, 1, 260101 https://doi.org/10.65553/EET.260101
Metacognition
in using GenAI for language teaching and learning
Mark
Feng Tenga
aThe Hang Seng University of Hong Kong, Hong Kong SAR, China
Submission date: May 10, 2026; Revision date:
October 10, 2026; Online: November 20, 2026
Abstract
GenAI permeates language
education. Metacognition, the awareness, monitoring, and regulation of one’s
thinking, becomes pivotal for channeling technological affordances toward
autonomous, effective learning. This article synthesizes foundational and contemporary
perspectives to clarify metacognition’s core components: knowledge of cognition
(person, task, strategy), regulation of cognition (planning, monitoring,
evaluating), and metacognitive experiences (confidence, perceived difficulty).
This article foregrounds context and culture to argue for tailored, socially
anchored pedagogy. Practical guidance details teacher modelling, structured
reflection, and routines for planning, monitoring, debugging, and evaluation,
alongside responsible integration of GenAI tools that cultivate declarative,
procedural, and conditional knowledge. This article concludes that aligning
GenAI-enabled learning with metacognitive growth can enhance calibration,
self-efficacy, and transfer, transforming AI from a crutch into a catalyst for
strategic, lifelong language learning.
Keywords: GenAI, metacognition,
language teaching, language learning, agency
Introduction
Generative
Artificial Intelligence (GenAI) shapes writing instruction (Lee et al., 2026),
Metacognition, learners’ awareness, monitoring, and regulation of their own
thinking, has never been more vital. Foundational work in educational
psychology establishes metacognition as purposeful control in the service of
goals, tightly linked to self-regulated learning, autonomy, and achievement (Flavell, 1979; Schraw, 1998; Zimmerman &
Bandura, 1994). In
language education, this translates into learners who plan, monitor, and
evaluate their strategies across reading, writing, listening, and vocabulary (Teng, 2025d), and
teachers who model and scaffold these processes (Anderson, 2008; Wenden, 1987, 1998).
GenAI
tools can amplify or erode these capacities. On the one hand, adaptive
feedback, analytics, and multimodal resources can personalize goals, surface
breakdowns, and support strategic adjustment. On the other, effortless answers
risk short-circuiting reflection, fostering overreliance and miscalibration.
The task before educators is to harness GenAI to strengthen, not supplant,
metacognitive development: cultivating declarative, procedural, and conditional
knowledge about learning (Teng, 2025abc),
explicitly addressing metacognitive experiences such as confidence and
perceived difficulty (Efklides, 2008), and
embedding assessment that captures real-time regulation (Haukås et
al., 2018). Situated within diverse contexts,
including motivation-challenging environments like Macao, this article outlines
principles and practices for aligning AI-enabled language learning with
metacognitive growth, so learners become more autonomous, calibrated, and
strategic in an GenAI-rich world.
Literature review
Understanding
metacognition
Metacognition
refers to thinking about one’s own thinking, people’s awareness of how they
learn, the strategies they use, and how they monitor and regulate those
processes to reach goals (Teng, 2025d). In educational psychology, Flavell’s
work is foundational. He defined metacognition as the active monitoring and
regulation of cognitive processes in the service of goals and distinguished
core elements: metacognitive knowledge, metacognitive experiences, goals, and
strategy activation (Flavell, 1976,
1979). Later, he described two broad
dimensions that have become widely adopted in research and practice: knowledge
of cognition (what we know about ourselves as learners, about tasks, and about
strategies) and regulation of cognition (how we plan, monitor, and evaluate
learning) (Flavell, 1985). These pillars shape how learners
approach tasks, assess progress, and adjust strategies to improve outcomes.
Metacognitive
knowledge is commonly classified into three types (Paris et al., 1984).
According to Schraw and Dennison (1994), declarative knowledge is explicit
understanding of oneself as a learner and of task demands and available
strategies (e.g., recognizing one’s strengths, preferred approaches, or typical
difficulties). Procedural knowledge concerns knowing how to execute strategies
(e.g., how to summarize a text or plan an essay). Conditional knowledge
involves knowing when and why to use particular strategies,
critical for adapting to task requirements and contexts. Complementing
knowledge, metacognitive regulation comprises the cyclical processes that
direct learning: planning (setting goals, choosing strategies, allocating
time/resources), monitoring (checking comprehension and performance in real
time), and evaluating (judging outcomes and the effectiveness of strategies).
Expanded models add information management (organizing, elaborating, and
focusing information) and debugging (identifying and correcting errors or
misunderstandings) as key subprocesses (Schraw & Dennison, 1994;
Schraw, 1998, 2001).
A
closely related line of work emphasizes metacognitive experiences, conscious
feelings and judgments that arise during cognition (e.g., feeling of knowing,
perceived difficulty, confidence). Efklides argued
that these experiences interact with metacognitive knowledge and skills through
monitoring and control loops, guiding adjustments to cognition in pursuit of
goals (Efklides, 2001,
2006,
2008).
These experiences are consequential in classrooms because affective states
(e.g., satisfaction, anxiety) shape subsequent strategy choices and the
development of metacognitive knowledge itself.
Across
disciplines, metacognition is recognized as pivotal for learning, problem
solving, and transfer, yet its conceptualization varies, producing definitional
inconsistencies (Azevedo, 2020; Veenman et al., 2006).
Norman et al. (2019) synthesized three research branches:
pre-reflective aspects (processes that may precede explicit awareness), active
regulation (how learners manage tasks), and developmental trajectories across
the lifespan. Despite these differences, a shared premise remains:
teachable metacognitive processes enable learners to set goals, monitor
progress, and flexibly adjust strategies, supporting academic success and
resilience (Schraw, 1998; Zimmerman & Bandura, 1994).
In
language education, metacognition is foundational yet historically
underutilized. Wenden (1987, 1998)
pioneered its application, mapping person, task, and strategy knowledge to
second language (L2) learning and arguing that metacognitive knowledge
energizes and guides self-regulated learning (SRL). Language learners with
stronger metacognitive awareness better anticipate challenges, believe in their
capacity to succeed, and take proactive steps to achieve goals. Subsequent work
positioned metacognition alongside language awareness and metalinguistic
awareness—the explicit sensitivity to language form, function, and use—which
enrich declarative knowledge about learning processes and enable deliberate
strategy choice (Gass & Selinker, 2008; Svalberg, 2007).
In practical terms, teachers can foster metacognition by activating prior
knowledge, modelling strategies, guiding goal setting, supporting monitoring,
and leading evaluation discussions, moving learners toward autonomy and
strategic control (Anderson, 2002, 2008).
Recent
syntheses foreground metacognition’s breadth in language teaching. Haukås (2018)
defines it as awareness and reflection about knowledge, experiences, emotions,
and learning across language learning and teaching, interlinking it with
beliefs, strategy use, multilingual/metalinguistic awareness, intercultural
awareness, and self-efficacy. This broad lens captures how learners and
teachers think about tasks, themselves, and context, and how that thinking
shapes actions. Teng and Zhang (2022) connect SRL and language learning
strategies, highlighting metacognition’s role in coordinating cognition,
behavior, environment, and motivation. Their perspective encourages examining
how learners set writing goals and regulate cognition, motivation, and behavior
during composing and revising (see also Teng, 2024).
Building on these traditions, Teng et al (2022)
describe metacognition as dual-level (observational monitoring and managerial
control) operating across acquisition, retention, and retrieval, with
metacognitive knowledge, experiences, and skills integrated through reflection.
Teng (2023)
further emphasizes the personal, developmental nature of metacognition:
knowledge, experience, and skill mutually reinforce through practice, making
metacognition a dynamic, improvable resource.
Crucially,
metacognition is distinct from cognition. Whereas cognition encompasses
processes like memory, comprehension, and problem solving, metacognition is the
reflective regulation of those processes—planning, monitoring, and evaluating
learning and performance (Flavell, 1979; Schraw, 2001).
This distinction matters for instruction because metacognition is explicitly
teachable. Techniques such as think-alouds, modelling
error correction, guided self-questioning, checklists, concept mapping,
reciprocal teaching, and reflective journals cultivate planning, monitoring,
and evaluation, thereby strengthening SRL and learner autonomy (Dimmitt &
McCormick, 2012; Schraw, 1998).
Evidence also indicates that teachers’ metacognitive understanding of their own
practice enhances their ability to design, scaffold, and assess learning that
builds students’ metacognition (Fairbanks et al., 2010).
Developmentally,
metacognition emerges early and evolves, with debates about age-related decline
complicated by findings that older adults can outperform younger adults on some
metacognitive tasks and learn new metacognitive skills (Brinck &
Liljenfors, 2013; Hertzog, 2016;
Pennequin et al.,
2010). A useful practical distinction is that
metacognition has trait-like and state-like properties (Sato, 2022):
some individuals habitually engage in reflection across contexts (trait), while
momentary conditions (e.g., task structure, stress) can amplify or dampen
metacognitive engagement (state) (see also Teng, 2025d).
Pedagogy can leverage this by designing tasks and environments that cue and
support metacognitive states—structured planning prompts, mid-task monitoring
checks, and post-task evaluations—while nurturing more enduring habits.
Metacognition’s
relevance in language domains is broad. In reading, awareness and regulation
enable learners to set purposes, select strategies, monitor comprehension, and
repair misunderstandings—capabilities that vary with exposure and literacy
traditions and benefit from explicit instruction (Lehtonen, 2000;
McLeod &
McLaughlin, 1986). In writing, metacognition underpins
planning, revising, and the use of feedback—activities where learners must
align goals, assess text against criteria, and adapt strategies. More
generally, successful language learners and teachers exhibit awareness of task
complexity, their beliefs, and context; they plan, monitor, and evaluate
systematically; and they align strategies with goals and conditions.
Several
cross-cutting issues deserve attention. First, measurement remains challenging:
strategy questionnaires have faced criticism for reliability and construct
clarity, motivating more robust, mixed-method assessment of strategic and
self-regulatory processes (Rose, 2012). Second, context matters. Metacognition
is socially and culturally situated; learners’ beliefs, motivations, and
community norms shape how metacognitive knowledge is formed and enacted, which
argues for tailored, context-sensitive pedagogy (Flavell, 1979).
Third, teacher education is pivotal. Building metacognitive competence in
pre-service and in-service teachers supports classroom practices that make
thinking visible and cultivates students’ reflective agency (Haukås et al., 2018).
Metacognition
thus comprises knowledge about learners, tasks, and strategies; conscious
experiences that signal progress or difficulty; and regulatory skills that
plan, monitor, and evaluate learning (Efklides, 2008; Flavell, 1979;
Paris et al., 1984;
Schraw &
Dennison, 1994). It is teachable, context-sensitive,
and both trait- and state-like (Sato, 2022; Schraw, 1998).
In language education, it anchors learner autonomy, strategic behavior, and
self-efficacy; it guides teachers’ design and reflection; and it strengthens
SRL across reading, writing, listening, and vocabulary learning (Anderson, 2008;
Haukås, 2018;
Teng & Zhang,
2022; Teng et al., 2022;
Wenden, 1998).
Centering metacognition, through explicit modelling, structured reflection, and
validated assessment, helps learners plan effectively, track understanding, and
make informed adjustments, thereby improving achievement and cultivating
lifelong, adaptive language learning.
Critical issues in
understanding metacognition
Metacognition
holds strong promise for improving language learning and teaching, yet several
critical issues complicate its effective use. At the conceptual level, the
field faces enduring ambiguity about what precisely counts as metacognition.
Foundationally, metacognition refers to the active monitoring and regulation of
cognition toward goals (Flavell, 1976, 1979).
However, researchers emphasize different facets—some attend to pre-reflective
processes that may precede awareness, others to active regulation during tasks,
and still others to developmental changes across the lifespan (Norman et al., 2019).
Calls for more cohesive theory underscore the need to specify the construct and
its mechanisms (Azevedo, 2020). This ambiguity is especially
problematic in language education, where metacognition overlaps with
self-regulated learning (SRL), language learning strategies, motivation,
self-efficacy, and language awareness. Without clear boundaries, metacognition
can be conflated with general strategy use or affective variables, diluting its
instructional usefulness. A pragmatic resolution is to adopt explicit
operational definitions that fit L2 outcomes, e.g., knowledge of cognition
(person, task, strategy) and regulation of cognition (planning, monitoring,
evaluating), while mapping connections, not equivalences, to SRL, motivation,
and language awareness (Flavell, 1985; Paris et al., 1984; Schraw, 1998).
Model
integration poses a second set of issues. The widely used knowledge–regulation
framework (Flavell, 1985) and the
declarative–procedural–conditional typology (Paris et al., 1984)
provide structure but can be taught as static lists rather than enacted
competencies. Extensions that include information management and debugging (Schraw &
Dennison, 1994) enrich regulation but complicate
sequencing: should educators build mastery of planning, monitoring, and
evaluating first, or integrate debugging and information management from the
outset? Likewise, Efklides’ work on metacognitive
experiences—feelings of knowing, perceived difficulty, and confidence that
inform control—highlights a dimension often marginalized in L2 interventions (Efklides, 2001, 2006,
2008).
If classroom designs ignore these experiences, learners may remain miscalibrated, over- or under-regulating their efforts.
More recent models that distinguish observational monitoring from managerial
control and spread metacognitive activity across acquisition, retention, and
retrieval add needed granularity, but they also raise design questions about
how tasks should cue monitoring versus control at each stage and how reflection
integrates knowledge, experience, and skill over time (Teng et al., 2022).
A coherent instructional approach must therefore link explicit knowledge,
structured regulatory practice, and the deliberate naming and calibration of
metacognitive experiences.
Measurement
and validity form a third, persistent bottleneck. Rose (2012)
criticized many language strategy measures for weak reliability and construct
clarity, a critique that applies equally to metacognition assessments reliant
on self-report. Inventories often conflate strategy frequency with
metacognitive quality and are vulnerable to social desirability and recall
biases. They infer regulation retrospectively rather than capturing in-situ
monitoring and control. A stronger approach triangulates methods: process
evidence from think-alouds, stimulated recalls, and
observations to record planning, monitoring, and debugging in real time;
product-linked indicators such as the alignment between stated goals and
outcomes or the quality of revisions following monitoring; experience sampling
to track confidence and perceived difficulty during tasks, which
operationalizes metacognitive experiences (Efklides, 2008); and performance tasks that require
justifying strategy choices, thereby tapping conditional knowledge. Crucially,
assessments should be tailored to modality, genre, and proficiency—listening,
for instance, demands rapid online monitoring that differs from the slower
planning and evaluation cycles of writing (Goh & Vandergrift, 2021).
Validity arguments should specify construct maps (knowledge, skills,
experiences) and test convergent and discriminant validity against SRL,
motivation, and proficiency measures, ideally in longitudinal designs to
examine whether metacognition predicts growth beyond initial ability (Schraw, 1998;
Zimmerman &
Bandura, 1994).
Context
and culture constitute a fourth critical domain. While metacognition is often
presented as an individual, conscious control system, language learning is socioculturally situated, and local ecologies shape how
metacognition is learned and enacted. The Macao context, with weaker extrinsic
academic pressure due to economic stability and fewer high-stakes incentives,
exemplifies a motivational ecology in which learners might not spontaneously
invest in reflective planning, monitoring, and evaluating unless tasks are
meaningful and socially anchored (see Teng, 2025d).
In such settings, metacognitive pedagogy should leverage authentic,
collaborative tasks aligned with students’ interests and lived experiences,
possibly integrating technology and peer learning to spark curiosity and
sustain engagement. By contrast, in high-stakes testing cultures, metacognition
can be instrumentalized as test-taking tactics, emphasizing performance
monitoring at the expense of deeper, transferable regulation. Cultural
traditions around language awareness also matter:
where explicit grammar instruction dominates, learners may have strong
declarative knowledge yet underdeveloped regulatory fluency; in communicative
traditions, the inverse may occur (Gass & Selinker, 2008; Svalberg, 2007).
Teacher beliefs and institutional norms are equally decisive. As Haukås et al.
(2018)
argue, teachers’ metacognitive awareness and collaboration support classroom
cultures that value reflective discussion and strategy orchestration. Without
supportive norms and professional development, classrooms may default to
transmission and rote practice that undercut metacognitive development.
Pedagogy
and teacher education, a fifth set of issues, require moving beyond “strategy
lists” toward reflective systems embedded in tasks. Anderson’s
framework—preparing and planning, conscious strategy decisions, monitoring,
orchestrating multiple strategies, and evaluating—implies a cyclical curriculum
where metacognition is not a one-off lesson but a recurring structure across
units and modalities (Anderson, 2002, 2008).
Effective instruction activates prior knowledge, models expert thinking through
think-alouds, provides guided goal setting, inserts
mid-task monitoring checkpoints, and closes with evaluation and transfer
discussions. It also makes errors productive by modeling debugging and
normalizing uncertainty. In listening, for example, teachers can prompt
pre-listening predictions, cue monitoring at multiple levels (lexical,
propositional, discourse), teach recovery strategies for breakdowns, and
debrief what worked and why (Goh & Vandergrift, 2021). In writing,
instructors can scaffold goal articulation, plan content and organization,
monitor alignment while drafting, debug local and
global issues, and evaluate drafts against genre criteria. Teacher
metacognition is pivotal here: instructors must surface their own planning,
monitoring, and evaluation processes, which implies sustained professional
development and collaborative reflection (Fairbanks et al., 2010;
Haukås et al., 2018).
Because measurement is difficult, assessment should be integrated into
instruction through planning sheets, monitoring logs, confidence and difficulty
ratings, revision rationales, and reflective journals that function as
formative feedback loops rather than compliance artifacts. Bridging SRL and
strategy research, as advocated by Teng and Zhang (2022),
suggests coupling metacognitive regulation with motivational and social
regulation—managing effort and interest, seeking help, and coordinating group
work—particularly in L2 writing where cognition, motivation, and behavior
co-evolve (Teng,
2024).
Development,
individual differences, and state–trait dynamics represent a final,
cross-cutting set of issues. Metacognition exhibits both trait-like and
state-like qualities. Some learners habitually engage in reflection across
contexts; others require task cues or supportive conditions to activate
metacognitive engagement (Sato, 2022). Instructionally, this means designing
environments that reliably cue metacognitive states—through prompts,
checkpoints, and reflective routines—while cultivating durable habits over
time. Calibration is central. Efklides (2008)
emphasizes that metacognitive experiences, such as confidence and perceived
difficulty, guide control; miscalibration can derail strategy selection,
leading to premature abandonment or unproductive persistence.
Prediction–performance–reflection cycles help align confidence with outcomes,
improving future planning and monitoring. Differences by proficiency and
modality also matter. Novice learners often need heavier scaffolding and
explicit declarative knowledge about strategies; advanced learners benefit from
deepening conditional knowledge (when and why) and practicing the orchestration
of multiple strategies in complex tasks. Listening demands rapid, online
monitoring and repair; writing allows more deliberate planning and evaluation (Goh &
Vandergrift, 2021). Equity considerations are essential as
well: without inclusive scaffolds, metacognitive instruction can amplify
advantages for already reflective learners. Differentiated supports, explicit
modeling, and accessible tools can help distribute metacognitive opportunities
more equitably.
These
critical issues carry implications for research and practice in language
education. Conceptually, scholars should state clearly whether they are
targeting metacognitive knowledge, regulation, experiences, or their
integration, and justify the theoretical lens (Efklides, 2008; Flavell, 1979;
Paris et al., 1984;
Schraw, 1998).
Methodologically, studies should prioritize mixed-method, task-embedded
assessments, aligning measures to modalities and proficiency levels and
building validity arguments that separate metacognition from adjacent
constructs such as motivation and general strategy use (Rose, 2012; Zimmerman &
Bandura, 1994). Contextually, interventions should be
adapted to local motivational ecologies and cultural traditions, leveraging
authentic tasks and collaborative structures to make reflective work meaningful
and sustainable, as highlighted in the Macao context. Pedagogically, teacher
education must foreground metacognition, equipping teachers to model their own
processes and to orchestrate cycles of planning, monitoring, debugging, and
evaluating in reading, writing, listening, and vocabulary instruction (Anderson, 2008;
Goh &
Vandergrift, 2021; Haukås et al., 2018).
Developmentally, programs should scaffold calibration and strategy
orchestration over time, tuning supports to learners’ proficiency and modality
demands and attending to motivational and social regulation alongside cognitive
control (Teng & Zhang, 2022; Teng, 2024).
Metacognition in the age of GenAI
Metacognition is
essential to the use of GenAI. Teng (2025a)
tackles a timely question at the intersection of EFL writing, feedback, and
GenAI: how metacognitive awareness (MA) shapes students’ perceptions and
experiences when using ChatGPT for writing feedback. Its mixed-methods design
is a core strength, combining group comparisons and regression analyses with
thematic interview insights to triangulate relationships between MA and
outcomes such as motivation, engagement, self-efficacy, and collaborative
tendencies. The finding of significant group differences and positive
associations between MA and responsible, effective use of ChatGPT aligns with
longstanding theory that MA underpins self-regulation, strategic control, and
self-efficacy in learning (Brown, 1987; Flavell, 1979;
Schraw &
Dennison, 1994), and with recent applied evidence that
learners with higher MA better calibrate their engagement with GenAI tools (Abdelhalim, 2024).
Methodologically, Teng (2025a) advances the field by relating MA
levels to concrete writing-process outcomes and by evidencing that higher MA
correlates with more critical and effortful use of ChatGPT. The inclusion of
semi-structured interviews is a welcome remedy, but richer process data—prompt
logs, screen-capture traces of revision cycles, or think-alouds—would
more directly capture monitoring, debugging strategies, and conditional
knowledge in action (Schraw & Dennison, 1994). Given that
ChatGPT’s affordances are highly prompt-contingent, performance-based tasks
that evaluate when and why students choose specific AI functions (e.g., idea
generation vs. line editing vs. argument organization) would help differentiate
responsible use from overreliance and clarify how conditional knowledge
operates under authentic constraints (Paris et al., 1984).
Teng (2025b)
addresses a timely and consequential gap: how EFL learners mobilize MA when
using ChatGPT for writing feedback. Its mixed-methods design is a strength,
advancing beyond the field’s tendency to rely on self-report scales by pairing
a theoretically grounded instrument with interviews that surface actual
practices and experiences, thereby responding to longstanding concerns about
overreliance on questionnaires in strategy/self-regulated learning research (Rose, 2012).
The findings, ranging from uncritical copying to sophisticated leveraging of
GenAI feedback, echo broader claims that metacognition is teachable, dynamic,
and context-sensitive, with meaningful implications for pedagogy, measurement,
and ethics (Efklides, 2008;
Flavell, 1979;
Schraw &
Dennison, 1994; Teng et al., 2022;
Teng, 2022).
Conceptually, the framing of MA aligns well with classic distinctions between
knowledge of cognition and regulation of cognition and with the
declarative–procedural–conditional typology (Flavell, 1979;
Paris et al., 1984;
Schraw &
Dennison, 1994). Mapping these subprocesses into
planning, monitoring, and evaluating for the Metacognitive Awareness in
Utilizing ChatGPT for Writing (MAUCW) scale is defensible and practical for
instructional design. However, the dynamic, experience-based dimension of
metacognition, feelings of knowing, perceived difficulty, and confidence,
receives less operational attention despite its centrality to calibration and
control (Efklides, 2008).
Given the observed spectrum of student behaviors, from copying outputs to
critically engaging with feedback, future versions of the instrument could
integrate measures of metacognitive experiences to detect miscalibration
(over-trust or under-trust of AI) and to connect those experiences with
subsequent control decisions.
On measurement, the
validation of a context-specific MA scale (MAUCW) tailored to ChatGPT use in
Teng (2025b)
is a notable contribution as GenAI reshapes task affordances. Anchoring item
development in established instruments and SRL theory supports construct
coverage (Schraw
& Dennison, 1994; Teng & Zhang, 2016;
Teng, 2020;
Teng & Zhang,
2024; Zimmerman & Bandura, 1994).
Still, field-wide concerns about self-report—construct clarity, social
desirability, and retrospective bias—remain salient (Rose, 2012).
The addition of interviews is a step toward multimethod validation; subsequent
work could incorporate process evidence (think-alouds,
prompt/revision trace data) and product-linked indicators (alignment of
revisions with goals and criteria) to capture in-situ regulation, including
monitoring and debugging (Schraw & Dennison, 1994). Because GenAI
use is prompt-contingent, conditional knowledge—knowing when and why to use
specific GenAI affordances—deserves targeted assessment tasks to distinguish
responsible, strategic use from overreliance (Paris et al., 1984;
Teng, 2025a;
Yao et al., 2024).
Pedagogically,
metacognitive scaffolds are necessary to move learners beyond surface tactics
to reflective orchestration (Anderson, 2008; Teng & Ma, 2024).
Instruction should explicitly structure planning (goal articulation, criteria
setting, and ethical prompt design), monitoring (mid-draft checks comparing
GenAI suggestions against task goals and source integrity), and
evaluation/debugging (justifying accepted or rejected edits through rationale
logs and revision memos), thereby cultivating conditional knowledge and
calibration (Flavell, 1979; Schraw & Dennison, 1994;
Schraw, 2001;
Yao et al., 2024).
Evidence that ChatGPT can bolster engagement, self-efficacy, and performance
when well guided supports embedding human-in-the-loop protocols that harness
immediate GenAI feedback while preserving writer agency and critical judgment (Allen &
Mizumoto, 2024; Ghafouri et al., 2024;
Teng, 2024ab). Sequencing
supports is also key: novices benefit from stronger declarative and procedural
scaffolds (what strategies exist, how to prompt and verify), whereas advanced
writers need opportunities to deepen conditional knowledge and orchestrate
strategies across drafting and revision (Paris et al., 1984;
Teng et al., 2022).
Ethically, the
tensions highlighted in the literature, copyright/source opacity, bias,
privacy, and the displacement of teacher/peer feedback, are integral to
metacognitive regulation in GenAI contexts (Akgun & Greenhow, 2022;
García-Peñalvo,
2023; Dizon et al., 2025;
Teng, 2025ab). High-MA
learners’ selective verification and critical acceptance of GenAI suggestions
observed in related studies suggest that ethics can be operationalized as
metacognitive routines: source checking, attribution decisions, transparency
about GenAI assistance, and reflection on the division of labor between human
and GenAI. Incorporating explicit ethical prompts such as “What did ChatGPT
contribute? What did I contribute? How did I verify?” can convert integrity
policies into teachable metacognitive practices rather than mere compliance.
Conclusion
Looking ahead, the
promise of GenAI will be realized not by replacing human reflection but by
structuring human–GenAI collaboration that strengthens it (Liu & Wang,
2026; Watson Todd, 2026). Research should specify how distinct
GenAI roles (tutor, evaluator, reviser, resource) cue different regulatory
subprocesses and experiences, and how to sequence supports developmentally
across acquisition, retention, and retrieval. Practice should embed routine
prompts for goal setting, mid-task monitoring, debugging, and post-task
evaluation, alongside the explicit naming and calibration of confidence and
perceived difficulty. With coherent theory, validated assessment,
context-sensitive pedagogy, and deliberate teacher learning, metacognition can
serve as the compass that guides learners toward autonomy, resilience, and
strategic sophistication in an GenAI-enabled future.
Acknowledgement
We gratefully acknowledge the anonymous reviewers for their diligent
efforts in reviewing this study.
Declaration
of Interest Statement
There is no potential conflict
of interest.
References
Brinck, I., & Liljenfors,
R. (2013). The developmental origin of metacognition. Infant and Child
Development, 22(1), 85-101. https://doi.org/10.1002/icd.1749
Efklides,
A. (2001). The systemic nature of metacognitive experiences. In P. Chambers, M.
Izaute, & P. -J. Marescaux
(Eds.), Metacognition: Process, function and use (pp. 19–34). Dordrecht,
Netherlands: Kluwer.
Efklides,
A. (2006). Metacognition and affect: What can metacognitive experiences tell us
about the learning process? Educational Research Review, 1, 3–14. https://doi.org/10.1016/j.edurev.2005.11.001
Efklides,
A. (2008). Metacognition: Defining its facets and levels of functioning in
relation to self-regulation and co-regulation. European Psychologist, 13, 277–287. https://doi.org/10.1027/1016-9040.13.4.277
García-Peñalvo,
F.J. (2023). The perception of Artificial Intelligence in educational contexts
after the launch of ChatGPT: Disruption or panic? Education in the Knowledge
Society, 24, 19. https://doi.org/10.14201/eks.31279
Gass, S. M., & Selinker, L. (2008). Second
language acquisition: An introductory course (3rd ed.). New York, NY:
Routledge.
Ghafouri, M., Hassaskhah,
J., & Mahdavi-Zafarghandi, A. (2024). From
virtual assistant to writing mentor: Exploring the impact of a ChatGPT-based
writing instruction protocol on EFL teachers’ self-efficacy and learners’
writing skill. Language Teaching Research. https://doi.org/10.1177/1362168824123976
Hertzog, C. (2016). Aging
and metacognitive control. In J. Dunlosky
& S. K. Tauber (Eds.), Oxford library of psychology. The Oxford handbook
of metamemory (pp. 537-558). New York, NY, US: Oxford University Press.
Lee, Y.-J., Parrila, R., & Colenbrander,
D. (2026). Adopt, adapt, or reject? Analysing
student-AI interaction in university curriculum-aligned writing tasks. Applied
Language Sciences, 2, 260102. https://doi.org/10.65553/ALS.260102
Lehtonen,
T. (2000). Awareness of strategies is not enough: How learners can give each
other confidence to use them. Language Awareness, 9(2), 64–77. https://doi.org/10.1080/09658410008667138
Lockl, K., & Schneider, W. (2006).
Precursors of metamemory in young children: The. role of theory of mind and
metacognitive vocabulary. Metacognition and Learning, 1(1), 15–31. https://doi.org/10.1007/s11409-006-6585-9
Pennequin,
V., Sorel, O., & Mainguy, M. (2010). Metacognition, executive functions and
aging: The effect of training in the use of metacognitive skills to solve
mathematical word problems. Journal of Adult Development, 17(3), 168-176.
https://doi.org/10.1007/s10804-010-9098-3
Teng, M. F. (2023). Metacognition. In
Z. Wen, B. Adriana, B., Mailce,
& F. Teng (eds.,). Cognitive individual differences in second language
acquisition: Theory, assessment & pedagogy (pp.175-199). Berlin: De
Gruyter Mounton.
Wenden,
A. L. (1987). Metacognition: An expanded view on the cognitive abilities of L2
learners. Language Learning, 37(4), 573–597. https://doi.org/10.1111/j.1467-1770.1987.tb00585.x
Author Bios
Mark Feng Teng, Ph.D., worked in Macao
Polytechnic University, and currently in English Department at the Hang Seng University of Hong
Kong. His research portfolio mainly focuses
on computer-assisted L2 vocabulary acquisition, reading and writing. He recently focuses on AI
research. His publications have appeared in international journals, including
Applied Linguistics, TESOL Quarterly, Language Teaching Research, System,
Applied Linguistics Review, Computer Assisted Language Learning, Computers
& Education, Foreign Language Annals,
and IRAL, among others. His recent monographs were published by Cambridge, Routledge, Springer, and Bloomsbury.
ORCID: 0000-0002-5134-8504