Self-regulated learning as a protective factor for adolescent psychological adjustment in AI-enhanced learning environments: an empirical study

Introduction

The rapid integration of artificial intelligence (AI) technologies into educational practice has fundamentally transformed the learning landscape for adolescents worldwide [1, p. 23]. AI-powered adaptive learning platforms, intelligent tutoring systems, and automated feedback mechanisms are increasingly prevalent in secondary education, promising personalized learning trajectories and enhanced academic outcomes [2, p. 145]. However, alongside these benefits, concerns have emerged regarding the psychological impact of intensive technology-mediated learning on developing adolescents [3, p. 67].

Adolescence represents a critical developmental period characterized by significant cognitive, emotional, and social transformations [4, p. 89]. During this stage, individuals develop essential capacities for self-regulation, identity formation, and autonomous functioning that shape their subsequent psychological trajectories. The introduction of AI-driven learning environments into this developmental context raises important questions about how such technologies interact with adolescents’ psychological adaptation processes.

Self-regulated learning (SRL) has been identified as a key metacognitive competency that enables learners to actively manage their cognitive, motivational, and behavioral processes during learning activities [5, p. 34]. Zimmerman’s cyclical model of self-regulation posits that effective learners engage in forethought (goal setting, strategic planning), performance control (self-monitoring, strategy implementation), and self-reflection (self-evaluation, adaptive adjustment) [5, p. 56]. Research has consistently demonstrated that higher levels of SRL are associated with improved academic performance and greater psychological well-being across diverse educational contexts [6, p. 201].

Despite growing research on both AI in education and adolescent self-regulation, relatively few studies have examined how SRL functions as a potential protective mechanism within AI-enhanced learning environments specifically [7, p. 78]. This gap is particularly significant given that AI learning tools can simultaneously support and undermine autonomous self-regulatory processes depending on their design and implementation [8, p. 156]. On one hand, AI systems that provide scaffolded feedback and adaptive challenges may reinforce SRL development; on the other, systems that automate learning decisions may diminish adolescents’ opportunities to practice self-regulation skills.

The present study addresses this gap by examining self-regulated learning as a protective factor for psychological adjustment among adolescents learning in AI-enhanced environments. Specifically, we investigate: (1) the relationship between AI learning environment characteristics and adolescent psychological adjustment; (2) the mediating role of SRL in this relationship; and (3) whether SRL moderates the association between AI tool usage intensity and psychological distress indicators.

Literature review

AI learning environments and adolescent well-being

The proliferation of AI-based educational technologies has generated a growing body of research examining their effects on learner outcomes. Holmes, Bialik, and Fadel [1, p. 45] provided a comprehensive framework for understanding how AI technologies reshape educational experiences, identifying both opportunities for personalization and risks of excessive technological dependence. Luckin and colleagues [9, p. 89] demonstrated that AI tutoring systems can significantly improve subject-specific performance, yet noted that sustained engagement with such systems was associated with elevated levels of academic anxiety in a subset of adolescent learners.

Psychological adjustment in educational contexts encompasses multiple dimensions, including subjective well-being, academic stress perception, emotional regulation, and social connectedness [10, p. 112]. Research has shown that technology-intensive learning environments can both enhance and compromise these dimensions. Selwyn [11, p. 167] argued that uncritical adoption of educational technologies frequently overlooks their potential to exacerbate existing inequalities and psychological vulnerabilities. More recently, longitudinal data from the OECD Programme for International Student Assessment [12, p. 45] indicated that moderate use of digital learning tools was associated with optimal well-being outcomes, while both low and high usage patterns predicted reduced psychological adjustment.

Self-regulated learning: theoretical foundations

Self-regulated learning theory provides a robust framework for understanding how learners actively manage their engagement with learning tasks [5, p. 78]. Pintrich’s [13, p. 453] model identifies four phases of SRL–forethought, monitoring, control, and reflection–operating across cognitive, motivational, behavioral, and contextual domains. This multidimensional conceptualization is particularly relevant to AI-enhanced learning environments, where learners must navigate complex technological interfaces while simultaneously managing their cognitive and emotional resources.

Recent research has extended SRL theory to technology-mediated contexts. Azevedo and colleagues [14, p. 234] developed the concept of “metacognitive regulation in hypermedia” to describe how learners deploy self-regulatory strategies when interacting with complex digital learning environments. Their work demonstrated that learners who engaged in more sophisticated metacognitive monitoring and evaluation strategies achieved deeper conceptual understanding and reported lower levels of cognitive overload.

SRL as a protective factor

The conceptualization of SRL as a protective factor draws on resilience theory, which identifies internal resources that buffer individuals against adverse outcomes in challenging circumstances [15, p. 89]. In educational contexts, Duckworth and Seligman [16, p. 112] demonstrated that self-regulatory capacities predicted academic success and psychological well-being more reliably than cognitive ability alone. Subsequent meta-analytic reviews confirmed that SRL skills are significantly associated with reduced academic stress, lower rates of burnout, and enhanced psychological adjustment across developmental stages [6, p. 345].

Within technology-mediated learning specifically, Broadbent and Poon [17, p. 67] found that time management, metacognitive strategies, and effort regulation–core components of SRL–significantly predicted academic achievement and emotional well-being in online learning contexts. However, their review also noted that the protective function of SRL may vary depending on the specific technological features of the learning environment. This finding underscores the need for research examining SRL within the distinctive context of AI-enhanced learning, where the technology itself may actively shape the conditions under which self-regulation operates.

Research methodology

Participants and procedure

The sample comprised 387 adolescents (207 female, 180 male) aged 13–17 years (M = 15.2, SD = 1.34) recruited from six secondary schools in Almaty, Kazakhstan. All participants had been using AI-enhanced learning platforms (such as adaptive mathematics software or AI-powered language learning applications) for a minimum of three months as part of their regular educational program. Informed consent was obtained from both participants and their parents or legal guardians. The study received ethical approval from the Institutional Review Board of Abai Kazakh National Pedagogical University.

Data were collected through an online survey administered during scheduled class periods under the supervision of trained research assistants. The survey was available in both Kazakh and Russian languages, and all instruments were subjected to rigorous forward-backward translation procedures with subsequent piloting to ensure cross-cultural validity.

Instruments

Self-Regulated Learning. SRL was assessed using the Motivated Strategies for Learning Questionnaire (MSLQ) [18, p. 801], adapted for adolescent populations and technology-mediated learning contexts. The adapted instrument comprised 44 items across six subscales: intrinsic goal orientation (α = 0.82), extrinsic goal orientation (α = 0.78), task value (α = 0.85), self-efficacy for learning (α = 0.89), metacognitive self-regulation (α = 0.84), and effort regulation (α = 0.76). Items were rated on a 7-point Likert scale ranging from 1 (not at all true of me) to 7 (very true of me).

Psychological Adjustment. Psychological adjustment was measured using the Strengths and Difficulties Questionnaire (SDQ) [19, p. 3], which assesses five dimensions: emotional symptoms, conduct problems, hyperactivity/inattention, peer relationship problems, and prosocial behavior. The total difficulties score (excluding prosocial behavior) was used as the primary indicator of psychological maladjustment (α = 0.81), while the prosocial behavior subscale and an additional well-being supplement served as indicators of positive adjustment.

Academic Stress. The Academic Stress Questionnaire (ASQ) [20, p. 156] was employed to measure perceived academic pressure across four domains: workload stress, examination anxiety, self-expectation stress, and despondency (α = 0.88). Respondents rated 20 items on a 5-point scale from 1 (never) to 5 (always).

AI Learning Environment Engagement. A purpose-designed questionnaire assessed the frequency, duration, and perceived quality of interaction with AI-based learning tools. This instrument included 15 items measuring three dimensions: usage intensity (hours per week spent using AI tools), perceived usefulness of AI feedback, and perceived autonomy within AI learning environments (α = 0.83).

Data analysis

Data were analyzed using structural equation modeling (SEM) with maximum likelihood estimation in Mplus version 8.6. The hypothesized model specified AI learning environment characteristics as exogenous variables, SRL as a mediating variable, and psychological adjustment indicators as endogenous variables. Model fit was evaluated using standard criteria: chi-square test, Comparative Fit Index (CFI > 0.95), Tucker-Lewis Index (TLI > 0.95), Root Mean Square Error of Approximation (RMSEA < 0.06), and Standardized Root Mean Square Residual (SRMR < 0.08) [21, p. 234]. Indirect effects were tested using bootstrapping with 5,000 resamples to generate 95% bias-corrected confidence intervals. Additionally, multi-group analysis was conducted to examine potential moderation by gender and age group.

Results

Descriptive statistics and correlations

Table 1 presents the descriptive statistics and bivariate correlations among study variables. SRL total scores were normally distributed (M = 4.52, SD = 0.87, skewness = –0.23, kurtosis = –0.18). Psychological adjustment total difficulties scores ranged from 2 to 31 (M = 12.45, SD = 5.67), consistent with community norms for adolescent populations. AI tool usage intensity averaged 6.3 hours per week (SD = 3.21).

Table

Descriptive statistics and bivariate correlations (N = 387)

Note. * p < .05, ** p < .01. Psych. Adj. = Psychological Adjustment (total difficulties); AI Usage = AI tool usage intensity (hours/week)

Bivariate correlations revealed significant negative associations between SRL and psychological difficulties (r = –0.41, p < .01) and between SRL and academic stress (r = –0.38, p < .01). SRL was positively correlated with subjective well-being (r = 0.47, p < .01). AI tool usage intensity showed a small positive correlation with SRL (r = 0.12, p < .05) but was not significantly correlated with psychological adjustment (r = 0.09, p > .05).

Structural equation model results

The hypothesized structural model demonstrated good fit to the data: χ²(142) = 267.34, p < .001; CFI = 0.96; TLI = 0.95; RMSEA = 0.048 (90% CI [0.038, 0.057]); SRMR = 0.042. All factor loadings for the latent constructs exceeded 0.60, confirming adequate measurement quality.

The structural paths revealed several important findings. First, perceived autonomy within AI learning environments was significantly and positively associated with SRL (β = 0.34, p < .001), indicating that AI systems perceived as supporting learner autonomy facilitated higher levels of self-regulation. Second, perceived usefulness of AI feedback was positively associated with SRL (β = 0.22, p < .01), while AI usage intensity showed no significant direct relationship with SRL (β = 0.07, p = .18).

Regarding psychological adjustment outcomes, SRL demonstrated a significant negative direct effect on total psychological difficulties (β = –0.39, p < .001) and a significant positive direct effect on subjective well-being (β = 0.42, p < .001). SRL also showed a significant negative association with academic stress (β = –0.35, p < .001). These findings indicate that higher levels of self-regulation were consistently associated with more favorable psychological adjustment outcomes.

Mediation analysis

Bootstrap analysis of indirect effects confirmed that SRL significantly mediated the relationship between AI learning environment characteristics and psychological adjustment. The indirect effect of perceived autonomy on psychological difficulties through SRL was significant (indirect β = –0.13, 95% CI [–0.19, –0.08]), as was the indirect effect on well-being (indirect β = 0.14, 95% CI [0.09, 0.21]). Similarly, perceived feedback usefulness exerted significant indirect effects on psychological difficulties (indirect β = –0.09, 95% CI [–0.14, –0.04]) and well-being (indirect β = 0.09, 95% CI [0.04, 0.15]) through SRL.

These results support the hypothesized role of SRL as a mediating protective mechanism: features of AI learning environments that support autonomy and provide useful feedback enhance adolescents’ self-regulatory capacities, which in turn promote more favorable psychological adjustment outcomes.

Moderation and multi-group analyses

Multi-group SEM analysis by gender revealed measurement invariance (metric and scalar) across male and female subgroups (ΔCFI < .01), indicating that the measurement model functioned equivalently across genders. However, structural path comparison revealed that the protective effect of SRL on psychological difficulties was significantly stronger for female adolescents (β = –0.46, p < .001) than for male adolescents (β = –0.31, p < .001), Δχ²(1) = 4.87, p = .027.

Age-based moderation analysis, comparing younger (13–14 years) and older (15–17 years) subgroups, revealed that the mediating role of SRL was more pronounced among older adolescents. Specifically, the indirect effect of perceived autonomy through SRL on well-being was stronger for the older group (indirect β = 0.18, 95% CI [0.11, 0.27]) compared to the younger group (indirect β = 0.09, 95% CI [0.02, 0.17]), consistent with developmental expectations regarding the maturation of self-regulatory capacities.

Discussion

The present study provides empirical evidence for the role of self-regulated learning as a protective factor for adolescent psychological adjustment in AI-enhanced learning environments. Three principal findings merit discussion.

First, the results confirm that features of AI learning environments are differentially associated with adolescent self-regulation. Perceived autonomy emerged as the strongest predictor of SRL, aligning with self-determination theory [22, p. 68], which posits that autonomy support is fundamental to intrinsic motivation and self-regulatory engagement. This finding has important implications for the design of AI educational technologies: systems that preserve and enhance learner agency–rather than automating learning decisions–are more likely to foster the self-regulatory competencies that underpin healthy psychological adjustment.

Second, SRL functioned as a significant mediator between AI learning environment characteristics and psychological adjustment outcomes, supporting its conceptualization as a protective factor. Adolescents who reported higher levels of metacognitive self-regulation and effort regulation experienced lower psychological difficulties and higher well-being, even when controlling for the intensity of their AI tool usage. This finding extends prior research on SRL and well-being [6, p. 345] by demonstrating that the protective function of self-regulation operates specifically within technology-mediated learning contexts featuring AI components.

Third, the moderation analyses revealed meaningful developmental and gender differences. The stronger protective effect of SRL among female adolescents may reflect gender differences in the deployment of metacognitive strategies documented in prior research [23, p. 189]. The age-related pattern–with stronger SRL mediation among older adolescents–is consistent with the developmental trajectory of executive functions and metacognitive abilities during adolescence.

These findings carry several practical implications. Educational practitioners and developers of AI learning tools should prioritize the design of systems that scaffold self-regulatory skills rather than replacing them. Explicit SRL training programs integrated into AI-enhanced curricula may serve as a preventive strategy for maintaining adolescent psychological well-being. Additionally, the differential effects by gender and age suggest that tailored interventions may be more effective than uniform approaches.

Several limitations should be acknowledged. The cross-sectional design precludes causal inference; longitudinal research is needed to establish the temporal precedence of SRL relative to psychological adjustment outcomes. The sample was drawn from a single national context (Kazakhstan), and generalizability to other cultural and educational systems requires further investigation. Self-report measures may be subject to social desirability and common method biases. Future research should incorporate behavioral indicators of SRL (such as log data from AI platforms) alongside self-report measures to provide more objective assessments of self-regulatory behavior.

Conclusion

This study demonstrates that self-regulated learning serves as a significant protective factor for adolescent psychological adjustment within AI-enhanced learning environments. SRL mediates the relationship between AI learning environment features–particularly perceived autonomy and feedback usefulness–and indicators of psychological well-being and distress. The findings underscore the importance of designing AI educational tools that support, rather than supplant, adolescent self-regulation. As AI technologies continue to permeate educational settings globally, ensuring that these tools foster self-regulatory competencies represents a crucial pathway for protecting adolescent mental health and promoting adaptive development.