Self-Regulated Learning

Self-regulated learning (SRL) describes learners as active participants who can shape and develop their cognitive and behavioral actions in a successful way. AI tools can either scaffold SRL development or inadvertently short-circuit it by removing the regulatory demands that build expertise.(Designing a mobile chatbot-based learning journaling system for intrinsic motivation and engagement)(The Evidence Base on AI in K-12: A November 2026 Review) Longitudinal evidence sharpens the distinction: reflective use around an AI tool predicted critical thinking (β = 0.43) but not knowledge gain, and tool access on its own changed neither.(Generative AI and Learning Dynamics in Higher Education: A Longitudinal Empirical Study)

Questions to Consider

• SRL describes learners actively managing their learning through three phases: forethought (goal setting, planning, self-efficacy), performance (strategy, self-observation), and self-reflection (evaluation, adaptation). Before you read, which phase do you actually do well — and which do you skip even though you know better?
• The page's core tension: AI can scaffold self-regulation or short-circuit it by removing the regulatory demands that build expertise. How can a tool that makes a task easier also make you a weaker regulator of your own learning — and can you feel the difference in your own use?
• Students often show a 'production deficit': they possess self-regulation knowledge but fail to deploy it spontaneously — asking a chatbot to 'extract the main ideas' and skipping planning and monitoring entirely. Have you caught yourself doing the cognitive equivalent of this, even while knowing the better strategy?
• When support is present, your own judgment can drop out of the picture: in one trace-data study the available support, not the learner's metacognitive accuracy, was what determined the revision strategy chosen. If external help overrides your read on your own learning, which judgment would you deliberately keep in your own hands?
• Research found a 'miscalibration gap': students can perceive more learning with GenAI while retaining less — preferring AI over note-taking despite weaker retention. If you feel productive while using a tool, how would you ever discover that you're not actually learning more?
• Whether GenAI functions as a scaffold, shortcut, or partner depends more on the learner's regulatory capacity than on the tool itself. But the page also shows self-regulation buffers — yet does not cancel — the harm of deep cognitive offloading. What does that 'buffers but doesn't cancel' caveat mean for designing better AI tools?
• Set a goal before reading: pick one task you regularly use AI for, and decide in advance which of the three SRL phases (forethought, performance, reflection) you'll deliberately protect from being automated. What result will tell you it worked?

Introduction

SRL is the process whereby learners actively manage their own learning through three interrelated phases:

1. Forethought: Goal setting, strategic planning, Self-Efficacy beliefs
2. Performance: Strategy deployment, self-observation, attention focusing
3. Self-reflection: Self-Assessment, causal attribution, adaptation

Proficient self-regulated learners employ cognitive strategies to improve success and utilize Metacognition to refine their learning processes continuously.(Designing a mobile chatbot-based learning journaling system for intrinsic motivation and engagement)

Empirical work in a technology-mediated course confirms the phase links: in an eight-week adaptive stochastics study (194 students), task value, self-efficacy, goal orientation and positive emotions in the pre-actional phase predicted actional regulatory behavior while negative emotions hindered it, and post-actional satisfaction fed forward into the next pre-actional phase (Mejeh & Fromm (2026)).

A systematic mapping of 84 AI–SRL studies found the research concentrated on higher-education students and on the metacognitive and cognitive aspects of self-regulation, with the motivational dimension underexplored and over a third of studies specifying no SRL theory at all (Banihashem et al. (2025)).

Crucially, SRL around AI is shaped by perception as well as behavior: Yilmaz et al. demonstrate that whether students perceive feedback as coming from AI or a human significantly affects their self-regulated learning and revision behavior — a reminder that the social framing of AI, not just its content, changes how learners regulate around it.

One failure mode sits before regulation begins: GenAI-assisted learning procrastination — unnecessary delay in starting GenAI-supported work despite a genuine intention to use it. Among 1,243 Chinese undergraduates, more frequent GenAI use went with less of it (β = −0.195), and learning GenAI anxiety strengthened the intention-delay link (Li et al. (2026)).

Where GenAI enters the cycle decides its effect: mapping Zimmerman's forethought, performance, and reflection phases onto AI-mediated environments, a co-agency framework argues the point of entry determines whether the tool amplifies or erodes the learner's sense of control, and that offloading supports transformative learning only when the decision is intentional rather than routine (Poudyal (2026)).

A cross-sectional survey of 434 Special Education undergraduates measures one such chain: GenAI literacy was positively associated with GenAI-assisted self-regulated learning behaviors (total B = 0.701), running mainly through learning Learner Agency (B = 0.345) rather than challenge emotions (B = 0.061) (Yang et al. (2026)).

Metacognitive self-regulation is a measured outcome, not an assumption, in Shao, Tang and Ju (2026), who found that an AI-VR flipped classroom raised it alongside achievement, and that learners with higher technological literacy benefited more.

Digital Support for SRL

Learning Journals

Learning journals are a promising SRL intervention: by reflecting on their learning processes, students increase awareness of cognition and strengthen regulatory capacity. Key design considerations:

• Structure matters: Open-ended journals often produce shallow entries; guided prompts and example models improve depth
• Motivation decay: Mobile journaling apps commonly see rapid engagement decline after a few days
• Scaffolding trade-off: AI assistance that writes reflections for students undermines the SRL practice; assistance that structures prompts without authoring content preserves it

Dashboards communicating SRL profiles to teachers

Mejia-Domenzain et al. (2026) extend digital SRL support to the teacher side: their Learning Analytics dashboard (DashED) communicates ML-derived self-regulated learning profiles to teachers in blended classrooms, and how teachers act on those profiles is context-dependent. In use, flipped-classroom (university) teachers followed a sequential exploration and favored course-level adaptation and showing dashboards in class, whereas vocational teachers revisited summary pages and used the tool mainly for individual coaching sessions. The actions teachers proposed were shaped by the content represented and their teaching level rather than the plot type — university teachers favored weekly tests and course adaptation, vocational teachers direct, individualized coaching. This positions the dashboard as a scaffold for teachers' regulation of instruction, with design needs that vary by context rather than a single optimal interface.

The student-facing counterpart is sparser: a dashboard that showed 46 secondary students their own GenAI prompts and their text overlap with the model's replies was opened by only about a third of the class, making voluntary exposure rather than the visualization itself the binding constraint (Fong et al. (2026)).

Scheu et al.'s 2×2 Experiment (2026)

In a randomized field experiment with 179 students over 22 days, two design principles were compared:

Key insight: The course improved SRL skills and intrinsic motivation through skill transfer, while the assistant improved engagement without affecting motivation.(Designing a mobile chatbot-based learning journaling system for intrinsic motivation and engagement)

AI Tools and the SRL–Motivation Reciprocal Loop

A foundational principle of SRL theory is that self-AI Regulation in Education skills and Motivation form a reciprocal relationship:

• Better SRL → more successful learning → higher self-efficacy → stronger motivation
• Higher motivation → more effortful engagement → better SRL practice

AI tools can enter this loop at different points:

• SRL-first design (e.g., structured courses, graduated hints, reflection prompts): Strengthens the loop by building genuine skill
• Engagement-first design (e.g., autocomplete, content generation): May boost behavioral engagement without entering the motivation loop, risking tool dependence

Strategic Regulation of GenAI as SRL

Kim (2026) reframes effective GenAI use in academic writing as strategic regulation — an enacted SRL practice of verifying, revising, selectively adopting, or rejecting AI output. In a mixed-methods study of 107 students, higher AI anxiety was positively associated with verification and revision (β=.24), while evaluative capacity predicted active revision and selective integration (β=.46). Students clustered into four regulatory types — Uncritical Reliance (18.7%), Selective Integration (34.6%), Evaluative Transformation (31.8%), and Strategic Rejection (14.9%) — showing that AI Literacy in Higher Education functions less as acceptance than as regulatory competence grounded in Evaluative Judgment and ethical responsibility. This positions SRL as the core mechanism distinguishing critical from uncritical AI use.

Governance of the tool is itself a measurable facet: a two-wave validation with Chinese EFL learners (EFA N = 305; CFA N = 342) separates six regulatory dimensions and names environmental regulation — checking generated output for accuracy, filtering suggested resources, and setting boundaries on reliance — as one in its own right (Wang, Zhang and Zhang (2026)).

Classify before delegating: Tsim and Gutoreva (2025) turn self-regulation into a per-task protocol — label each subtask Substitute, Complement, Aid or Non-negotiable, justify it in a short metacognitive note, and keep an audit trail of prompts, drafts and human revisions — with the cycle repeating so one task's assignment informs the next. Every prompt as a regulatory trace. Reading 429 prompts from 23 students as self-regulated learning events showed the structure mix shifting with task design while 53 of 59 submissions stayed close to the model's output, locating regulation in the exchange rather than the artifact (Kontorovich & Wallach (2026)). The interaction itself as an object of regulation. Brunnström and Palmqvist (2026) document the same regulation demand from the other direction: in an eight-round demonstration using a chatbot to prepare a take-home examination answer, the AI's default output stayed at the quantitative, multistructural end of the SOLO taxonomy — polished, submission-ready and pedagogically thin — and reached a usable three-step learning loop only after repeated meta-level interventions ("this is overwhelming, can you condense it?"). Their conclusion is that productive use required "the very self-regulatory skills the tool was expected to support": the learner must set incremental goals, request difficulty adjustments, and reflect on what is not yet understood, on top of the disciplinary content itself. They name this capacity AI-interaction literacy and treat disengaging from the tool as a legitimate regulatory decision rather than a failure of persistence (Metacognition).

• Satisfaction is not self-regulation. Liang et al. (2026) surveyed 689 undergraduates in industry-education programs and tested a serial mediation model in which the perceived affordances of AI-generated content raise AIGC Self-Efficacy (beta = 0.583) and, through it, learning Motivation (beta = 0.565) and self-regulated learning (beta = 0.250), with motivation the heaviest single predictor of SRL (beta = 0.527). The load-bearing negative results sit alongside those paths: the quality of AI assessment feedback predicted satisfaction strongly (beta = 0.712) but not self-efficacy (beta = 0.131), and satisfaction had no significant effect on self-regulated learning (beta = 0.032). A well-liked, well-functioning assistant is therefore not evidence that regulation improved — the mechanism runs through confidence and motivation, not through the learner's experience of the tool.

Dependency works through effort regulation, not use volume. In a survey of 982 undergraduates, GenAI dependency predicted lower effort regulation (β = −0.257), which predicted perceived learning outcomes (β = 0.497), giving a negative indirect path (β = −0.128) while frequency of use carried no significant total association (Li, Zhang and Yang (2026)).

GenAI-aware reflection as SRL

The 5P Reflection Model (Kadel et al. 2026) re-centers structured reflection in the GenAI era as an enacted SRL practice. SRL owns reflection only as one phase of its cycle; the practice itself — prompts, journals, portfolios and scored reflective text — is carried on the Reflection page, which also sets out where it differs from Metacognition and Self-Assessment. Learners increasingly co-create meaning with AI, and traditional reflection models struggle to authenticate what a student has done. The 5P model (Purpose, Process, Product, Pitfalls, Plan) fuses forethought-driven goal setting, reflection-in-action (documenting prompts and iterations), reflection-on-action (validating the probabilistic output against external sources), an explicit pitfalls stage for hallucination, plagiarism, and over-reliance, and a forward-looking plan — embedding emotional monitoring throughout. Its "process over product" philosophy treats structured documentation of the human-AI interaction as the regulatory demand that preserves authenticity, Learner Agency, and metacognitive depth, positioning GenAI-aware reflection as a scaffold for self-regulation rather than a substitute for it.

Relationship to Tutoring-Specific Design

Tutoring-specific AI aligns with SRL-first design: it provides graduated scaffolds that preserve Learner Agency and require strategic self-regulation. General-purpose AI often removes the regulatory demands entirely.(The Evidence Base on AI in K-12: A November 2026 Review)

For example:

• Bastani et al.'s tutoring-specific chatbot preserved step-by-step reasoning (SRL demand)
• The general-purpose GPT variant simply provided answers (SRL bypass)

Evidence Across Contexts

A meta-analysis of 34 studies (91 effect sizes) estimates that AI-based interventions raised self-regulated learning by g = 0.76, with resource management (g = 0.93) larger than metacognitive (0.73) or cognitive strategies (0.48). Tool type and pedagogical role moderated the effect; generative-AI status, discipline, intervention length, and SRL measure did not (Zhang et al., 2026).

• Mixed evidence and the miscalibration gap. A rapid review of PreK-12 GenAI research finds metacognitive gains during supported tasks often do not persist when support is removed, and that GenAI can increase perceived learning even when durable learning is absent (the miscalibration gap — students preferred GenAI over note-taking despite weaker retention). Students need explicit, stage-appropriate training to decide what to delegate and when independent effort matters.(Young people, learning, and generative AI: A rapid literature review and implications for PreK-12 education)

• SRL is rarely the active target. Across 394 GenAI higher-education studies, self-regulated learning and reflection were among the least frequently enacted uses and among the outcomes targeted least (Lin et al. (2026)).

• Agentic initiative vs. self-regulation tension. Woollaston et al. (2026) note that as agents automate more of a task, the less self-regulated cognitive work the learner performs — so designs should give learners control over agent initiation (dynamic, fading scaffolding) to preserve self-regulatory capacity rather than outsourcing it.

• Self-regulation shapes AI coding-assistant use. A study of AI coding assistants found high-Computational Thinking students showed stronger self-regulatory coherence (planning-execution-self-reflection) and used AICA for code understanding, while low-CT students used it for immediate answer retrieval.

• SRL co-occurs with lower digital distraction in online learning. Shi et al. (2026), using unsupervised data mining on 530 college students, found that SRL strategies — goal setting, environment structuring, and time management — co-occurred most consistently with lower digital distraction in online learning, alongside learner-instructor and learner-content engagement. The finding positions concrete SRL training as a high-leverage intervention for focused online study.

• Metacognitive awareness, not tool access, is the strongest driver of adaptive STEM performance. Alatoai and Alshahri (2026) built and validated the AI-STEM-MLCS with 649 secondary students in Saudi Arabia (CFI = 0.983, RMSEA = 0.019; McDonald's ω = 0.888 to 0.905) and found its four dimensions explained 68% of the variance in self-regulated learning performance (R² = 0.68). AI-based metacognitive awareness was the strongest predictor (β = 0.38, p < 0.001), followed by cognitive transfer and adaptability (β = 0.29, p = 0.008) and AI-enhanced self-regulated learning (β = 0.21, p = 0.040), while creative and critical AI-STEM reasoning did not predict the criterion (β = 0.14, p = 0.135). The authors read the pattern as evidence that adaptive feedback raises STEM performance mainly by prompting learners to examine errors, recalibrate, and reuse strategies, so the instrument works better as a diagnostic that locates meta-learning gaps than as a single global score. It is a self-report measure validated in one national system, so the coefficients are provisional.

• External support can bypass a learner's own metacognitive judgment. Iqbal et al. (2026) traced 87 EFL students revising an essay under GenAI (ChatGPT 4.0), human-expert, or no support. The support condition was by far the strongest correlate of the revision strategy a student chose (Cramér's V = 0.668), well ahead of carry-over from the preceding writing task (V = 0.333), while metacognitive judgment accuracy (p = 0.172), writing skill (p = 0.261) and Motivation (p = 0.683) were unassociated. Metacognitive judgment tracked strategy only when no support was available (permutation test p = 0.0460). Students with GenAI support gained the most (about 4 points on average under the decreasing-help-seeking strategy, against −0.5 points for the same strategy under human-expert support), yet no revision strategy was associated with score change (H(3) = 3.895, p = 0.273), so the gains came from the tool rather than from better self-regulation. The authors' orientation is that GenAI should prompt reflection on one's revision strategy instead of supplying direct help.

• Reflective use tracks critical thinking, not knowledge gain. Melanou et al. (2026) followed three parallel classes of business informatics students (N = 87) across a nine-week course, with knowledge rising in every group (F(1, 50) = 29.87, p < 0.001, η²p = 0.374) and no AI advantage, no Time × Group interaction, and no Matthew effect (F(1, 48) = 2.46, p = 0.124; BF01 = 8.70). Reflective use (checking sources and verifying AI output before adopting it) was markedly higher in the AI condition (3.72 vs 2.82; F(1, 40) = 20.21, p < 0.001) and predicted critical thinking (β = 0.43, p < 0.001) but not knowledge gain, which germane cognitive load did predict (β = 0.51, p = 0.001). The practical reading is that AI access is not itself an intervention, and that its metacognitive payoff surfaces in reasoning quality before it surfaces in test scores.

• Structure, not AI alone, drives SRL gains. Mapping Zimmerman's phases onto project-based hybrid instruction with AI as metacognitive scaffolding produced an N-Gain of 0.72 versus 0.28 for conventional instruction (t = 8.47, p < 0.001, Priyambudi et al. (2026)), who argue AI strengthens self-regulation only inside task structure — unstructured use risks metacognitive laziness.

• AI speed can displace the peer dialogue that carries co-regulation. Coding 28,123 forum sentences across four graduate courses, one human-only and two with heavy AI TA use, Davis et al. (2026) found the AI TA was first responder in 97% of student-started assignment threads, with a median 0.92 minutes to its reply against 490 when a peer replied first. Thread starts were unchanged (p = .73) but reply posts fell from 18.2 to 10.9 per student (p < .001), and peer repliers' share of follow-up sentences dropped from 91% to 45% between semester halves.

• The resulting SRL decline is compositional. Because peer repliers supply about 70% SRL-coded sentences against 20 to 27% for thread starters, aggregate regulatory language fell from 49% to 36% (p < .001). Each role's own rate held steady, so the drop reflects who was writing rather than a change in how individuals write. Self-regulation still predicted grade (β = 2.63, p = .017), and peer replies moderated that relationship (β = 1.45, p = .013). The grade penalty attached to AI-dominated threads (β ≈ −1.38, p = .038) then vanished once 17 high self-regulators were removed (β ≈ −1.29, p = .454), which the authors read as the cost running through lost co-regulation rather than through the AI directly. It is observational, so the mechanism stays unproven.

• Emotional scaffolding moves trace-measured SRL phases that cognitive scaffolding alone does not, and self-report barely sees it. A preregistered experiment with 144 high school students compared plain ChatGPT, an SRL-scaffolding agent and an emotional version that asked about feelings at three points and adapted tone and challenge along control-value lines. Trace coding found large condition effects in all three phases (ε2 = 0.44 forethought, 0.34 performance, 0.34 self-reflection). SRL prompts alone lifted goal setting and planning to the emotional version's level, while the emotional layer added more performance-phase activity (r = 0.22) and self-reflection (r = 0.31). Self-reported SRL separated the conditions only in the performance phase; trace-self-report correlations were 0.313 there, 0.22 for forethought and 0.04 for self-reflection. The authors read the trace differences as co-regulated enactment rather than independent regulatory capacity (Diordiev, Stenberdt & Makransky, 2026).

• In groups, regulation is triggered by the AI's own prompt, and preconception diversity changes how that prompt lands. Wang et al. (2026) compared 8 different-preconception and 6 similar-preconception pre-service teacher groups working with an on-demand metacognitive chatbot and a proactive voice agent. Prompt function converged across conditions at 39 prompts, 18 against 21, while sequential structure did not: diverse groups sustained positive or neutral evaluation of the output and similar groups sustained negative evaluation alongside recurrent task agreement. Raw-data inspection traced part of the negative loop to the proactive speaking agent, so the authors read it as a timing and modality mismatch rather than a rejection of support. Their design implication is that shared-regulation support in groups that have already converged should stay optional.

LLM-Mediated SRL: Scaffold, Shortcut, or Partner?

A cluster of Learning Letters studies (2026) converges on a central tension: GenAI can scaffold, short-circuit, or partner with self-regulation depending on design and how learners regulate its use. The evidence points to SRL itself — not the tool — as the decisive variable.

• Viberg et al. find that LLMs are woven into a layered Help-Seeking ecosystem rather than replacing human support: students try tasks independently first, then consult ChatGPT as a low-barrier first step, peers for conceptual negotiation, and instructors for high-stakes issues. They favor instrumental help-seeking (hints, step-by-step guidance) over executive help-seeking (direct solutions), exercising selective Trust and verifying outputs against course materials — a four-stage process (deciding whether help is needed, choosing whom to ask, determining the type of help, judging the help received) that can be measured and taught.
• Atif & Dickson-Deane frame GenAI use as Cognitive Offloading that can be either scaffolded (learners critique and adapt AI outputs, keeping Learner Agency and sense-making) or substitutional (learners accept outputs with minimal verification, shifting control to the tool). In a study of 267 postgraduate IT students, the same tool could scaffold or shortcut SRL depending on learner strategy — confident users showed agency in goal setting and monitoring; less confident users saw GenAI as a shortcut or misconduct.
• Lim & Bannert show the risk concretely: students voluntarily used a genAI chatbot (73%) and scored higher on essays, but they offloaded comprehension and synthesis (asking the chatbot to "extract only the main ideas") and engaged in almost no planning or monitoring — outsourcing key regulatory decisions. This reflects a production deficit: students possess SRL knowledge but fail to deploy it spontaneously, so genAI tools should prompt reflection (a monitoring scaffold) when queries indicate offloading.
• Song et al. demonstrate that SRL is both a stable aptitude and a dynamic state: individual baselines are consistent, but metacognitive knowledge and Well-Being decline systemically over a semester, driven by assessment deadlines. They show GenAI can act as a context-aware learning partner when it is given personal, temporal, and contextual data — supporting students without replacing their effort. This argues against "one-time-fits-all" personalization based on baseline aptitude alone.
• de Barba extends SRL theoretically, arguing the field has narrowed to task-focused regulation and to optimizable behavioral proxies in educational technology. The paper proposes a cross-scale account of learner agency — regulation (within tasks), integration (across time and contexts), and positioning (critically in relation to the conditions framing learning) — as a design orientation for algorithmically mediated environments.
• Self-regulation buffers but does not cancel offloading harm. Chen (2026) shows that self-regulated writing attenuates — but does not eliminate — the negative association between deep Cognitive Offloading and independent no-AI outcomes in GenAI-assisted writing: the offloading-by-SRL interaction was positive (B = 0.22), flattening the harm from a slope of −0.54 (low SRL) to −0.33 (high SRL) but not canceling it. A bounded-support condition pairing delegation limits with compulsory reflection produced the strongest independent performance, evidence that metacognitive regulation partially protects learners yet cannot fully compensate for delegating the cognitive work itself.

Where rental sits inside Winne's models. de Barba (2026) maps the mechanism onto SRL as operations the system performs in the learner's place: noticing, monitoring and evaluative judgment are exercised by the tool, an impasse is resolved by search rather than by the learner's own If-Then, Else choice, and the standards by which the learner judges are borrowed rather than built. A capability that a learner believes they hold while the tool holds it — hidden rental — suppresses the error signal that would otherwise prompt regulation, and ten propositions specify what to test, including larger costs in domain acclimation and standards drifting toward the model's own reference points.

The collective lesson: SRL is the core mechanism distinguishing critical from uncritical AI use. Whether GenAI functions as a scaffold, shortcut, or partner depends on learners' regulatory capacity and on whether tools are designed to preserve (rather than remove) the regulatory demands that build expertise.

Which support is in play decides which mediator carries the association: among 3,003 Chinese pre-service teachers, perceived AI tool support ran to innovative competence mainly through AI Self-Efficacy (60.88% of that path's total effect), while perceived school-based intelligent environment support ran mainly through self-regulated learning (29.64% versus 18.60%) (Liu et al. (2026)).

Students reported more procrastination (β = 0.525) alongside perceived performance gains, which separates delaying study work because AI is available from delaying it with AI's help — a distinction a single self-regulation score cannot carry (Gouroubera et al., 2026).

Implications

• For journaling/chatbot tools: Combine SRL instruction (course-based) with optional writing support to get both motivation and engagement gains
• For tool designers: Make evaluation a required step, not an optional one. Iqbal et al. found that the available support, rather than the learner's metacognitive judgment, decided the revision strategy students used and that GenAI score gains did not come from better regulation, while Melanou et al. found reflective use predicted critical thinking where mere tool access did not. Interactions that ask learners to verify, compare, and re-strategize are the ones carrying the learning value.
• For instructors: Treat AI-based metacognitive awareness as the first lever and check it explicitly. Alatoai and Alshahri found it the strongest predictor of adaptive STEM performance (β = 0.38) and that transfer and adaptability came second (β = 0.29), so activities that have students examine errors and carry a strategy into a new problem context do more than general prompting for critical or creative reasoning.
• For AI policy: Procurement criteria should ask whether a tool develops or displaces self-regulation
• For researchers: Long-term studies measuring SRL outcomes (not just immediate performance) are essential; Melanou et al. show the payoff of that patience, since a semester's reflective use moved critical thinking (β = 0.43) without moving knowledge gain.

Conversational Agents and SRL in Simulation Games

• Conversational agents supporting self-regulated learning in games. Wenzel, Geiger, and Liening (2026) show that an AI conversational agent (Lara) in a business Simulation game can support self-regulated learning through metric-based formative feedback, on-demand guidance, and structured reflection — addressing the common limitation that simulation games provide limited formative feedback and reflection prompts. Evaluations with student teachers and BSG participants reported positive perceptions of the agent's cognitive and social presence and its support for self-regulation.

Automated SRL Detection: What Transfers and What Doesn't

• Embedding-based SRL classifiers cross languages within a domain, but not across domains. A study of 26 students (chemistry and formal logic; think-aloud protocols coded with the Winne-and-Hadwin model) found embedding-based models (text-embedding-3-small) trained on English classify German SRL at ~0.8 AUC with ~0.1 degradation — except enacting (~0.7). Cross-domain transfer fails: chemistry-to-logic drops to ~0.6, enacting to chance (~0.5) — the domain wall, not language, binds.
• Three failure modes explain where transfer breaks. (1) Concept resolution at everyday rather than domain meaning — "cancel" vs. chemistry "kürzen" — breaks enact classification. (2) Anthropomorphization — German students unfamiliar with a heavily scaffolded tutor addressed it as a person, producing out-of-distribution utterances; transfer was better on less-scaffolded platforms. (3) Hint-reading vs. plan-formulation ambiguity — models can't tell reading a metacognitive hint from planning without log data. Realizing errors is robust across every condition — noticing you're wrong is context-independent. Fixes: expert dictionaries or fine-tuning for domain terms; log data for hints. (Large language models generalize SRL prediction to new languages within but not between domains)
• Shallow cues over-attribute regulation in curriculum mapping. Given assessment descriptions, GPT-o3-mini read mentions of deadlines or scheduling as evidence of a self-regulatory skill; Self-Management was over-identified this way, its agreement with expert weights ranging from 0.53 to -0.03 (Jayalath et al., 2026).
• Structured question chains turn prompting into regulation. In a quasi-experiment with 83 undergraduates, both question-chain conditions produced more learning-strategy and monitoring behavior than free-form questioning (χ² = 6.551, p = .038) and scored higher on metacognitive awareness and achievement (F = 3.406, p = .038; F = 3.807, p = .035), with no difference between the chain forms (Lin et al., 2026). Sequencing is what does the work.
• Group-level detection follows the same split, and talk carries more than logs. In dyads modeling a truck's motion in C2STEM, 394 task-context segments from 36 high schoolers were labeled with seven socially shared regulation codes and predicted from LLM summaries of each segment: text-only embeddings reached mean AUC 0.650 while log-derived features alone reached 0.526, near chance for a construct that is largely social. Planning was invisible within a single segment (0.5036) and became detectable once the two segments on either side were concatenated (0.6711), whereas reflecting was best once the logs were added (0.5862), the same hint-reading-versus-plan ambiguity this section records for individual learners (Zhang et al., 2026).
• Automated SRL labels hold for enacting and fail for planning. In a fourth-grade mixed-reality science class, a pipeline fusing gaze, gesture, speech and game logs labeled 75 human-checked segments at 74.67% overall, from 93.3% for enacting and interacting down to 46.7% for planning (Fonteles et al., 2026).
• Persistence coded from dialogue with a pedagogical agent runs negative against performance, in LLM and human coding alike. In a conversation-based assessment of 107 English-learner middle school students, persistence detected from chat-log turns by eight ChatGPT4o configurations correlated negatively with task performance (r = -0.19, p < .001), and one human coder's persistence counts matched that direction at r = -0.32 (Ober et al. (2026)). Perceived challenge (r = -0.25) carried the only stronger negative association, and the other constructs ran negative too (self-efficacy r = -0.15, prior KSAs r = -0.14, interest r = -0.08). The authors read persistence expressions as possible unproductive struggle and argue for distinguishing adaptive from maladaptive persistence — strategic re-attempting versus repetitive trial-and-error — before any automated system treats persistence talk as engagement.
• GPT-4o can build a theory-grounded SRL codebook from its knowledge base alone. Given only references to Zimmerman (2000) and Winne and Hadwin (1998), GPT-4o proposed constructs covering all three Zimmerman phases and three of the four Winne-Hadwin stages, with only the task-definition phase missing, and human refinement settled the codebook at 11 constructs (Zambrano et al. (2026)). Adding the 955 think-aloud utterances from 15 students to the prompt closed the gap, covering all phases of both models and introducing data-grounded codes such as Error Identification. Human refinement remained essential: two coders merged overlapping GPT proposals (Planning versus Goal Setting, Monitoring versus Evaluation, Outcome versus Process Evaluation), and no hallucinations appeared.

Prompting for self-regulation is not the same as teaching it. In a four-session randomized trial with 371 students in Grades 7–9, neither utility-value prompts nor cognitive-strategy prompts improved strategy use, interest, effort or domain knowledge relative to standard ChatGPT (Fütterer et al., 2026). The null matters because the intervention added reflection prompts to an unchanged learning task: whatever regulated the work sat in the task and the classroom, not in the prompt.

Common Misconceptions

• "Self-regulation is a learner trait." Regulated learning decomposes into self-reflection, causal attribution and adaptation, and proficient learners employ cognitive strategies to improve success. These are trainable processes that a design can support or undermine.
• "Any adaptive design supports regulation equally." Scheu et al.'s 2x2 randomized field experiment with 179 students over 22 days compared two design principles and found them different in mechanism, so "adaptive" is not a single treatment.
• "Anxiety undermines regulation." In a mixed-methods study of 107 students, higher AI anxiety was positively associated with verification and re-checking. Anxiety functioned as a prompt to verify, which is regulation rather than its absence.
• Metacognitive SRL predicted learning-relevant load where the AI feedback did not. In a five-week field study with 234 students and 1,147 observations, metacognitive self-regulation was a substantial positive predictor of germane cognitive load (β = .363, p < .001 intention-to-treat; .346, p < .001 usage-based) and did not significantly predict extraneous load — while mere access to conversational AI feedback predicted neither. Only sustained engagement with the feedback separated students on germane load, and irregular use did not (β = −.219, p = .116), which locates the active ingredient in the learner's regulation of processing rather than in the feedback supply (Wenzel, Geiger and Liening, 2026).

Connected Concepts

• Learners — Learners: the umbrella for the learner-side concepts
• Metacognition — the cognitive monitoring SRL relies on
• Self-Assessment
• Self-Efficacy — a forethought-phase belief driving effort
• Scaffolding — graduated support that preserves regulatory demand
• Feedback — input learners regulate around
• Feedback Literacy — the capacity to act on feedback
• Help-Seeking — a strategic SRL behavior
• Motivation — the reciprocal partner of self-regulation
• Cognitive Offloading — the risk when AI removes regulatory work
• Generative AI — the technology that can scaffold or short-circuit SRL
• AI Literacy — regulatory competence in AI use
• Self-Directed Learning — the broader autonomy construct
• Learner Agency — the learner's capacity to act with intention, central to regulation, integration, and positioning
• Adaptive Learning — personalization that can support regulation
• Formative Assessment — continuous feedback for regulation
• Learning by Teaching — a strategy building self-regulation
• Intelligent Tutoring — systems that scaffold SRL
• Large Language Models (LLMs) — the underlying model of AI tools
• Retrieval, Spacing and Interleaving — scheduling, self-testing and study-strategy choices learners make
• Cognitive Surrender
• Reflection

Connected Articles

• Enhancing School Students’ Self-Regulated Learning through Generative AI Support: A Randomized Controlled Trial — 371 German students: reflection and strategy prompts added nothing over standard ChatGPT (Fütterer et al., 2026)

• A Meta-Analysis of AI-Assisted Self-Regulated Learning — Meta-analysis of AI-assisted self-regulated learning: g = 0.76 across 34 studies, with tool type and pedagogical role moderating the effect (Zhang et al. 2026)

• Generative AI dependency and perceived learning outcomes among university students: an indirect association involving effort regulation — Generative AI dependency and perceived learning outcomes: the effort-regulation pathway (Li, Zhang & Yang 2026)

• Analyzing embodied learning in classroom settings: A human-in-the-loop AI approach for multimodal learning analytics — Per-category accuracy of LLM-segmented SRL in an embodied classroom (Fonteles et al. 2026)

• What the submission does not show: Complexifying cognitive offloading in students' mathematical work with LLMs — What the submission does not show: complexifying cognitive offloading in students’ mathematical work with LLMs (Kontorovich & Wallach 2026)

• Leveraging GenAI models for Curriculum Analytics: A Human-Centered Approach for Curriculum Mapping in Higher Education — Model over-attribution of self-regulation from shallow cues in curriculum mapping (Jayalath et al. 2026)

• AIGC affordance and student self-regulation in private undergraduate education: a serial mediation model of self-efficacy and learning motivation — AIGC Affordance and Student Self-Regulation

• AI-interaction literacy: reflections on how generative AI might be used to support self-regulated learning in higher education — AI-interaction literacy: steering a chatbot demanded the SRL it was meant to support (Brunnström & Palmqvist 2026)

• The 5P Reflection Model for Education in the Generative Artificial Intelligence (GenAI) Era — The 5P reflection model for the GenAI era (Kadel et al. 2026)

• Layer-sensitive cognitive offloading in generative AI-assisted writing: Supported performance and independent no-AI outcomes — Layer-sensitive cognitive offloading in GenAI-assisted writing (Chen 2026)

• Reclaiming epistemic agency: A critical framework for human-generative AI co-agency in education

• Learner Agency Across Scales: An Integrative Perspective on Self-Regulated Learning in Algorithmically Mediated Environments — Learner agency across scales: regulation, integration, positioning

• The rented self: How AI-enabled tools decouple performance from becoming — The rented self: rental mapped onto Winne's SRL models, with ten propositions and within-learner indices

• GenAI as a Learning Partner: Supporting Self-Regulated Learning Over Time Without Replacing Effort — GenAI as a context-aware learning partner over time

• How Do Students Regulate Their Learning With a GenAI Chatbot? — How students regulate learning with a genAI chatbot

• Scaffold or Shortcut? Postgraduate IT Students' Use of Generative AI and Self-Regulated Learning — Scaffold or shortcut? GenAI dual role in SRL

• Efficiency vs. Effectiveness: Self-Regulated Learning with LLM-Mediated Help-Seeking — LLM-mediated help-seeking in STEM: layered, instrumental, and verified

• Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task

• Fostering self-regulated learning through adaptive learning technology: A differentiated perspective on the role of feedback

• A systematic mapping review at the intersection of artificial intelligence and self-regulated learning

• Supporting self-regulated learning through generative AI feedback in online higher education: the importance of student perceptions of the source of feedback — GenAI feedback and SRL: perceived source matters

• From AI anxiety to strategic regulation: How university students transform generative AI into a strategic learning resource — From AI anxiety to strategic regulation

• Young people, learning, and generative AI: A rapid literature review and implications for PreK-12 education — Mixed evidence on metacognition/self-regulation with GenAI

• Agentic AI and Pedagogical Best Practice: The Tension Between Automation and Learning — Agentic AI and the self-regulation tension

• Decreasing Digital Distraction in College Students: Associated Online Learning Strategies Identified by Unsupervised Data Mining Approaches — SRL and lower digital distraction in online learning (Shi et al. 2026)

• Making machine learning findings accessible to teachers in blended classrooms — Making ML findings accessible to teachers in blended classrooms

• SCAN: A Decision-Making Framework for Task Assignment with Generative AI — SCAN: a learner-facing loop of task identification, justification and post-task reflection

• Using Learning Analytics to Support Secondary School Students' Writing with Generative AI — Using Learning Analytics to Support Secondary School Students' Writing with Generative AI

• Exploring the Impact of AI-Based Learning Environments on Student Self-Regulation and Adaptive STEM Learning — A validated instrument for AI-supported self-regulation in adaptive STEM learning, with metacognitive awareness as the strongest predictor

• Human or GenAI Support? Conditions Impacting Students' Strategy Choices in an Essay Revision Task — Support condition, not metacognitive judgment, drove revision strategy choice in an essay-revision experiment

• Generative AI and Learning Dynamics in Higher Education: A Longitudinal Empirical Study — Longitudinal: reflective AI use predicted critical thinking but not knowledge gain, with no Matthew effect

• GenAI literacy and GenAI-assisted self-regulated learning behaviors: the roles of learning agency and challenge emotions — GenAI literacy related to self-regulated learning behaviors mainly through learning agency

• Pre-Service Teachers' Perceptions of AI Technological Support and School-Based Intelligent Environment Support: Model-Based Indirect Associations with Innovative Competence via AI Self-Efficacy and Self-Regulated Learning — Support type flips the mediator: AI self-efficacy vs self-regulated learning

• GenAI Use and GenAI-Assisted Learning Procrastination in University Students: The Roles of Planned Behavior Constructs and Learning GenAI Anxiety — GenAI-assisted learning procrastination: an intention-behavior gap that anxiety strengthens

• Generative artificial intelligence-supported self-regulated learning (GenAI-SRL) in L2 writing: scale development, validation, and short-form construction — A validated six-dimension GenAI-SRL scale for L2 writing, with environmental regulation as its own dimension and a 12-item ant-colony-optimized short form

• Large language models generalize SRL prediction to new languages within but not between domains — Large Language Models Generalize SRL Prediction to New Languages Within But Not Between Domains (Borchers et al. 2025)

• A Project-Based Hybrid Learning Model Using Artificial Intelligence to Enhance Students' Self-Regulated Learning — A Project-Based Hybrid Learning Model Using Artificial Intelligence to Enhance Students' Self-Regulated Learning

• AI teaching assistants in online discussion forums and peer-supported self-regulated learning — An AI TA as first responder displaced peer replies, and with them the co-regulation that amplified the grade value of self-regulation (Davis et al. 2026)

• GenAI-based metacognitive prompting in collaborative inquiry: A comparison across diverse preconception groups — Preconception diversity changed how a metacognitive prompt landed: same prompt functions, opposite sustained evaluation of the AI output (Wang et al. 2026)

• Drivers and academic outcomes of generative AI use among university students in Benin — Procrastination rose with use (β = 0.525) beside perceived gain, separating delay enabled by AI from delay merely about AI (Gouroubera et al. 2026)