Digital Learning, Generative AI, and Second Language Learners’ Self-Regulation: A Meta-Synthesis

Introduction

For the first time in history, education is no longer bound to physical space. With a single tap, learners carry dictionaries, teachers, communities, and entire libraries wherever they go. Digital platforms enable immediate access to extensive textual resources, which are continually modified and restructured. This is the reality of digital learning (DL) spaces. Mobile devices and generative artificial intelligence (GenAI) tools give learners unprecedented access. Yet the real revolution is not the technology itself, but the learner’s ability to navigate this portable world with intention and control. Learners can use mobile devices to learn beyond the walls of a classroom (Kukulska Hulme & Lee, 2020) across multiple contexts, through social and content interactions (Crompton, 2013).

With the recent rise of GenAI tools such as ChatGPT and other large-language-model (LLM) platforms, the boundaries of DL have expanded even further. In this study, GenAI tools are conceptualized as an emerging extension of DL that may provide advanced scaffolding for self-regulated learning (Ge et al., 2025; Wei, 2023). While traditional DL and mobile-assisted language learning (MALL) environments support autonomous and interactive learning (Crompton, 2013; Kukulska-Hulme & Lee, 2020; Palalas & Wark, 2020), as Viberg et al. (2020) note, DL environments succeed when learners receive appropriate support to become self-regulated language learners. Building on this premise, GenAI systems have the potential to deliver adaptive, AI-mediated feedback and conversational practice, thereby supporting learners’ ability to plan, monitor, and evaluate their learning within digital environments (e.g., Li & Kim, 2024).

Self-regulated learning (SRL), a concept rooted in Zimmerman’s (2000) framework, refers to the process by which learners plan, monitor, and evaluate their own learning behaviors. In digital and AI-mediated contexts, SRL becomes a decisive factor in determining whether learners engage productively or passively with technology. Recent research (Ge et al., 2025; Lai, 2024; Shi et al., 2025) shows that GenAI tools can effectively scaffold SRL by supporting goal-setting, reflection, and adaptive feedback. However, Qian et al. (2026) caution that without rigorous evaluation, AI-generated scaffolds may contain unreliable guidance, which can undermine learners’ trust and impede the development of effective self-regulatory skills.

Despite the growing literature on MALL and SRL, few studies have synthesized evidence on how DL tools, such as GenAI tools, affect language learners’ self-regulation. Most prior work has focused on either mobile technologies or AI tools separately, leaving a gap in understanding how these digital environments collectively shape learners’ SRL behaviors.

Several reviews have examined the relationship between digital technologies and SRL, yet each addresses only part of the emerging landscape. Palalas and Wark (2020) synthesized research on mobile learning and SRL between 2007 and 2019, providing a thematic overview of definitions, devices, and measurement approaches; however, their review predates the widespread emergence of GenAI tools and focuses exclusively on mobile learning. Similarly, Lai et al. (2022) conducted a scoping review of learning strategies in mobile self-directed language learning, mapping the field but not offering an interpretive synthesis of how digital tools reshape SRL processes.

Expanding beyond language education, Junaštíková (2024) reviewed SRL in technology-enhanced higher education contexts, identifying general trends but without specific attention to second language (L2) learning or technology-mediated scaffolding. More recently, Noori (2025) examined digital technologies and SRL in language learning. Although these reviews provide valuable descriptive and thematic mappings, none provides an interpretive synthesis of empirical findings specifically at the intersection of DL, GenAI, and SRL in L2 contexts, nor do they examine how AI-generated feedback and scaffolding mechanisms influence learners’ regulatory processes.

A qualitative meta-synthesis is particularly appropriate for addressing this gap. The existing body of research spans qualitative and mixed-method designs with diverse outcome measures, contexts, and theoretical frameworks. This makes statistical aggregation through meta-analysis methodologically inappropriate. At the same time, a scoping review would primarily map the extent and characteristics of the literature without generating higher-order interpretive insights. In contrast, a qualitative meta-synthesis enables the integration and reinterpretation of findings to identify recurring patterns and contextual moderators shaping technology-based self-regulated learning (TBSRL). Therefore, this approach provides a theory-informed understanding of how learners experience and enact SRL within digital environments.

To address this gap, the present study synthesizes findings from existing research on the intersection of DL, GenAI, and self-regulated language learning. Specifically, it examines the relationship between learners’ self-regulation and the use of DL tools and GenAI in English as a foreign or second language (EFL/ESL) contexts, identifies factors that moderate this relationship, and explores the connection between TBSRL and English language learning outcomes. Although framed within L2 theory, the present synthesis focuses exclusively on English language learning contexts. Additionally, the study investigates which SRL strategies are most and least frequently used in technology-enhanced and AI-assisted environments, and it analyzes the benefits, risks, and limitations associated with integrating DL and GenAI into language education.

Literature Review

Theoretical Foundations of SRL in Technology-Mediated L2 Learning

SRL has long been recognized as a central mechanism underlying successful language acquisition (Dörnyei, 2005; Oxford, 2017). In L2 contexts, SRL enables learners to set goals, deploy strategies, monitor progress, and reflect on outcomes in ways that support sustained language development (Teng, 2024; Teng & Zhang, 2016, 2024; Zhang & Zhang, 2019, 2024). The present synthesis is primarily guided by Zimmerman’s (2000) cyclical model of SRL, which conceptualizes regulation as an iterative process involving forethought, performance, and self-reflection phases.

To capture motivational dynamics within this cycle, Pintrich’s (2004) framework is also drawn upon, particularly its emphasis on goal orientation and motivational regulation. In digital environments, these perspectives are complemented by hypermedia-based models of metacognitive regulation (Azevedo, 2010; Azevedo et al., 2010), which highlight how learners regulate cognition while interacting with complex technological systems. Together, these frameworks provide the conceptual basis for examining TBSRL in L2 contexts. Although Zimmerman’s and Pintrich’s models provide the primary theoretical grounding for this study, sociocultural and ecological perspectives emphasize that SRL in technology-mediated environments emerges through dynamic interactions among learners, digital tools, teachers, peers, and contextual affordances (Hadwin et al., 2018; van Lier, 2004; Oxford, 2017). This perspective is particularly relevant in DL and GenAI contexts, where platform design and pedagogical scaffolding may influence regulatory development.

It is important to distinguish SRL from closely related constructs such as self-directed learning (SDL) and learner autonomy. SRL typically refers to learners’ strategic regulation of cognition, motivation, and behavior within learning tasks (Pintrich, 2004; Zimmerman, 2000), whereas SDL emphasizes learner control over broader learning decisions (Garrison, 1997) and autonomy highlights independence in learning contexts (Holec, 1981). Contemporary research suggests these constructs are best viewed as overlapping but distinct dimensions of learner agency rather than hierarchical categories.

In the present study, TBSRL is used as an integrative construct referring to learners’ strategic regulation of cognition, motivation, and behavior while engaging with digital and AI-mediated language learning environments. This interpretation aligns with Dabbagh and Kitsantas’s (2012) argument that technology-mediated personal learning environments, including social media and other digital platforms, inherently require and promote self-regulated learning skills by enabling learner control, personalization, and the integration of formal and informal learning. Although related terms such as technology-supported SRL and SRL in hypermedia environments appear in prior research, TBSRL is employed here as an umbrella concept to synthesize findings across diverse technological contexts, including mobile learning, web-based platforms, and emerging GenAI systems.

DL and MALL as Environments for Self-Regulated Language Learning

DL in L2 education is grounded in multiple theoretical traditions that view technology as a mediational tool for expanding learning opportunities. GenAI can be understood as a recent development within broader DL ecosystems, encompassing advanced systems capable of producing textual, visual, and auditory outputs through machine learning techniques (Baidoo-Anu & Ansah, 2023). More broadly, DL has been defined as the application of technology to enhance teaching and learning processes (Carrier, 2017), enabling learners to transcend classroom boundaries and participate in informal and self-directed learning experiences (Jeno et al., 2018).

Historically, the role of mobile technologies in language learning has evolved substantially. Early mobile devices in the mid-1990s were primarily used for accessing electronic documents (Sophonhiranrak, 2021). By the mid-2000s, their function expanded into informal learning spaces, demonstrating educational potential beyond traditional classrooms (Unwin, 2015). More recently, mobile devices have become central tools for language learning across diverse contexts, including developing countries (Alothman et al., 2017). This trajectory reflects a broader shift toward learner-centered and ubiquitous learning environments that align closely with SRL principles.

A substantial body of research indicates that MALL environments can support learner autonomy, interaction, and engagement. For example, Kavoshian et al. (2020) reported that mobile social networks enhanced communicative skills, motivation, and feedback exchange among Iranian EFL teachers. Similarly, Chen and Tsai (2021) found that teachers conceptualized mobile learning primarily through student-centered pedagogical lenses. At the learner level, Ebadijalal and Yousofi (2021) demonstrated that mobile-assisted peer feedback reduced speaking anxiety and increased willingness to communicate. Studies focusing on writing have reported comparable benefits, including improved accuracy, performance, and attitudes toward collaborative learning (Gharehblagh & Nasri, 2020), as well as increased autonomy and motivation (Al-Shehab, 2020).

Research shows that technology-enhanced environments promote SRL behaviors when platforms include metacognitive and motivational scaffolding (Bernacki et al., 2011; Daumiller & Dresel, 2018). Similarly, Mashayekh and Dashtestani (2025) reported that a OneNote-based e-portfolio facilitated learners’ reflection, self-monitoring, and goal-setting while improving speaking self-efficacy, illustrating how digital tools can operationalize SRL principles through structured pedagogical design. Moreover, Smartphone-based learning has been associated with higher SRL engagement and improved language learning outcomes (Pérez-Álvarez et al., 2018; Singh & Miah, 2020). Navigating digital environments requires substantial regulatory effort, as learners must evaluate feedback, visualize goals, and manage online resources (Yen et al., 2018). Learners also benefit from teacher mediation in early stages of MALL (Yang, 2020), while structured self-regulatory systems have been found to enhance performance (Shih et al., 2010). More recent studies confirm a correlation between SRL and technology use, indicating that learners with stronger SRL skills engage more effectively with digital tools (Junaštíková, 2024). DL research grounded in CALL similarly emphasizes the role of SRL in facilitating meaningful engagement with digital learning environments (Jin et al., 2023; Noori, 2025).

Taken together, the reviewed studies converge on the conclusion that mobile and digital environments can facilitate aspects of self-regulated learning, particularly behavioral engagement and strategy use. Improvements in motivation, autonomy, and communicative participation are among the most consistently reported outcomes across contexts. These findings align with broader syntheses indicating that mobile learning and SRL mutually reinforce one another (Palalas & Wark, 2020).

At the same time, the literature reveals notable inconsistencies. While some studies report meaningful gains in higher-order regulatory processes, others document primarily surface-level engagement or short-term motivational effects. Variability appears linked to differences in instructional scaffolding, learner proficiency, and the extent to which explicit SRL support is embedded in technological design. These mixed findings suggest that technology access alone is insufficient to guarantee deep self-regulatory development without appropriate pedagogical and metacognitive scaffolding.

Emerging Contributions of GenAI to TBSRL

Within the broader DL landscape, GenAI has recently attracted significant attention for its potential to reshape language learning processes. Research indicates that AI applications can function as conversational partners and intelligent tutors (Liu et al., 2024a), generate practice materials and corrective feedback (Godwin-Jones, 2024), and support multiple language skills simultaneously (Limiya & Prasantha Kumar, 2025). Reviews of AI in English language teaching further highlight both pedagogical affordances and implementation challenges (Crompton et al., 2024).

Across the studies included in this meta-synthesis, approximately one quarter explicitly examined GenAI-mediated environments. Collectively, this emerging evidence suggests that GenAI may extend earlier forms of technological scaffolding in several important ways. Recent work also points to the growing role of reliable AI-driven scaffolding in supporting SRL processes while attempting to reduce reliability problems associated with hallucinated AI-generated content (Qian et al., 2026).

However, the empirical base remains uneven. Much of the existing GenAI research is concentrated in higher education EFL writing contexts (Shen & Teng, 2024; Shi et al., 2025; Wang, 2024; Wang et al., 2024). Moreover, scholars caution that AI tools should complement rather than replace human instructional guidance (Haroud & Saqri, 2025; Marín et al., 2025). Reported risks include threats to academic integrity, reduced creativity, and unintentional plagiarism (Agas, 2025). At the same time, integrity-oriented pedagogies may help learners maintain ownership of their work in AI-rich environments (Stephens et al., 2024; Stephens & Bertram Gallant, 2024), while learners themselves report tensions between improving linguistic quality and preserving authentic voice (Wang, 2024).

AI Literacy, Metacognition, And Regulatory Capacity

An emerging strand of research highlights the growing importance of AI literacy as a complementary competence to SRL. Studies of GenAI-supported writing environments indicate that learners with stronger regulatory skills are better able to evaluate AI-generated content, manage reliance on automated assistance, and maintain authorial control (Shi et al., 2025; Wang et al., 2024). Conversely, limited regulatory capacity may increase the risk of overdependence on AI support. From a metacognitive perspective, this pattern is consistent with hypermedia research showing that effective learners actively monitor and adapt their interactions with complex digital systems (Azevedo, 2010).

More broadly, AI-mediated informal digital learning studies grounded in social cognitive theory suggest that learners’ regulatory behaviors, motivation, and technology use are dynamically interconnected (Guan et al., 2025). Nevertheless, existing research remains fragmented, and the precise mechanisms linking AI literacy and SRL in L2 contexts are still under-theorized.

Sociocultural and Ecological Perspectives on TBSRL

Although much of the literature adopts cognitive–metacognitive perspectives on SRL, sociocultural viewpoints emphasize that regulation is also socially mediated and contextually situated (Vygotsky, 1978). Therefore, learners’ regulatory behaviors in DL and GenAI environments are shaped not only by individual strategy use but also by task design, platform affordances, teacher mediation, and institutional norms. Some studies note the continued importance of teacher scaffolding in early stages of mobile learning (Yang, 2020) and the role of structured self-regulatory systems in enhancing performance (Shih et al., 2010).

Synthesis and Research Gap

Despite the expanding body of work on digital and AI-mediated language learning, several gaps remain. First, prior studies tend to examine individual technologies in isolation rather than synthesizing how different forms of DL and GenAI collectively shape self-regulated language learning. Second, while many studies report positive motivational and behavioral outcomes, the specific SRL phases and dimensions most consistently supported by technology remain unclear due to fragmented reporting. Third, although interest in GenAI is rapidly increasing, its distinctive contribution to SRL processes in L2 contexts has not yet been systematically synthesized. Finally, the relationship among SRL, SDL, autonomy, and emerging AI literacy remains conceptually underdeveloped.

Purpose of the Study

Addressing these gaps requires an integrative qualitative meta-synthesis that maps the evolving landscape of TBSRL. Accordingly, the present study examines the relationship between SRL and the use of DL tools and GenAI in EFL/ESL contexts, identifies moderating factors, investigates links between TBSRL and language outcomes, and analyzes the benefits, risks, and limitations associated with integrating DL and GenAI in language education. Based on these aims, the study addresses the following research questions:

1.    What is the relationship between SRL and the use of DL tools and GenAI in L2 contexts?

2.    What factors moderate the relationship between SRL and the use of DL tools and GenAI in language learning?

3.    What is the relationship between TBSRL and English language learning outcomes?

4.    Which self-regulated learning strategies are most and least frequently used in technology-enhanced and AI-assisted language learning environments?

5.    What benefits, risks, and limitations are associated with the integration of DL and GenAI in language education?

Method

This study employed a qualitative meta-synthesis design based on the framework proposed by Sandelowski and Barroso (2006), which provides a systematic approach for integrating qualitative findings across empirical studies. To ensure transparency and replicability, the reporting process followed the PRISMA 2020 guidelines (Page et al., 2021), covering all stages of identification, screening, eligibility assessment, and inclusion. Data analysis was conducted using thematic analysis (Braun & Clarke, 2006), with NVivo facilitating the organization and coding of the extracted findings. Following Braun and Clarke’s (2006) six-phase procedure of familiarization with the data, generating initial codes, searching for themes, reviewing themes, defining and naming themes, and producing the report, the researchers moved iteratively across phases to deepen interpretation and strengthen analytic rigor. This recursive movement ensured that emerging insights remained closely connected to the original data while allowing for the development of broader interpretive thematic patterns.

A systematic search was conducted in the Web of Science Core Collection to identify journals indexed in Journal Citation Reports (JCR) and in Scopus to capture additional peer-reviewed journals. Studies published between January 2005 and December 2025 were eligible, and database searches were conducted between 2021 and 2025. Only peer-reviewed journal articles indexed in JCR and/or Scopus were included.

A wide range of keywords and keyword combinations was used to capture the breadth of digital and AI-enhanced learning contexts, including DL, generative AI, AI-enhanced learning, web-based English learning, game-based English learning, mobile learning, MALL, self-regulation, SRL, L2 learning, L2 learning with technology, mobile apps, metacognitive strategies, online language learning, self-directed learning, autonomous language learning, technology-assisted self-regulation/SRL, technology-based self-regulated English learning, and English learning apps. Boolean operators (AND, OR) were applied strategically to broaden the search and ensure that studies employing diverse digital tools and SRL-related constructs were captured. To enhance comprehensiveness, the reference lists of both included and excluded full-text articles were manually screened to identify potentially relevant studies that were not retrieved through the database search.

After retrieving initial results, titles and abstracts were screened to remove duplicates and exclude studies unrelated to SRL, DL, or technology-mediated L2 learning. Full texts of potentially relevant articles were then assessed against predetermined inclusion criteria, focusing on empirical research examining SRL or TBSRL within L2 learning contexts. Studies meeting all criteria were included in the final meta-synthesis and subsequently coded and synthesized through thematic analysis. One coder initially coded all data, and the second coder reviewed the codes, double-checked them for consistency, and provided feedback. The coders then discussed discrepancies and reached consensus on which codes to retain, remove, or merge based on the included studies. Coding decisions were systematically discussed and grounded in the included studies to ensure transparency and consistency.

Inclusion/Exclusion Criteria

Following Sandelowski and Barroso (2006), explicit inclusion and exclusion criteria were applied to ensure rigor and relevance in study selection (Table 1). Out of 461 initially identified studies, only 56 met the final inclusion criteria after full-text screening.

Table 1.

Article Selections

Data Collection Procedure

The study selection process followed Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) 2020 guidelines (Page et al., 2021). Figure 1 illustrates the number of records identified, screened, and included. We excluded studies focusing on learners with identified special educational needs because these contexts often involve specialized pedagogical supports that fall beyond the scope of the present synthesis. Additionally, case studies were excluded to maintain analytic comparability across studies, as single-case designs often yield highly context-specific findings that are difficult to synthesize systematically across diverse TBSRL settings.

Figure 1.

PRISMA 2020 Flow Diagram of the Study Selection Process

*Consider, if feasible to do so, reporting the number of records identified from each database or register searched (rather than the total number across all databases/registers).

**If automation tools were used, indicate how many records were excluded by a human and how many were excluded by automation tools.

Source: Page MJ, et al. BMJ 2021;372:n71. doi: 10.1136/bmj.n71.        

This work is licensed under CC BY 4.0. To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/

Data Analysis

In this meta-synthesis, 56 studies were included (see Appendix A). NVivo codes (hereafter referred to as nodes) were used to organize the data. After reviewing the included research papers, 94 initial nodes were created. Later, they were categorized into five core nodes and 15 sub-nodes (see Appendix B). The five core nodes include: relationship between DL and SRL (reported in 12 studies), moderating factors between SRL and DL (reported in 50 studies), SRL strategies (reported in 32 studies), relationship between TBSRL and language learning outcomes (reported in 41 studies), and benefits, risks, and limitations of DL (reported in 29 studies). Therefore, five themes were generated from the synthesis. In the first core node, four sub-nodes were identified: DL-mediated SRL, AI and autonomy, attention, and responsibility. In the second core node, the two sub-nodes of learner-related and educational factors were found. As for the third core node, two sub-nodes of the most and the least used strategies were revealed. Moreover, the fourth core node disclosed three sub-nodes of language skills and components, achievement, and self-directed technology use. Finally, for the last core node, benefits, risks, and limitations emerged.

Results and Discussions

From the 56 included studies, 94 initial NVivo nodes were created and later organized into five core nodes and 15 sub-nodes (see Appendices A and B):

1.     the relationship between DL and SRL (12 studies),

2.     moderating factors (50 studies),

3.     SRL strategies (32 studies),

4.     TBSRL and learning outcomes (41 studies), and

5.     Benefits, risks, and limitations (29 studies).

The synthesis of the 94 nodes demonstrates the complex, contextual, and task-dependent nature of SRL (Pintrich, 2004; Usher & Schunk, 2013). It also reinforces that digital tools operate as ecological affordances shaped by motivational, pedagogical, and sociocultural conditions (Lai & Gu, 2011; Reinders & Benson, 2017).

The findings can also be interpreted through Zimmerman’s (2000) cyclical model of SRL and Pintrich’s (2004) socio-cognitive framework. Across studies, digital and AI-mediated environments supported forethought processes through goal-setting, motivational beliefs, and planning; performance processes through monitoring, strategy use, and feedback engagement; and self-reflection through evaluation and adaptive revision of learning behaviors. At the same time, motivational variables such as self-efficacy, task value, and goal orientation emerged as central moderating influences, reinforcing Pintrich’s emphasis on the interaction between motivational regulation and strategic learning. The results and discussion for each core node are presented below:

Relationship between DL and SRL

The main sub-nodes were DL-mediated SRL (reported in nine studies), AI and autonomy (reported in two studies), attention (reported in one study), and responsibility (reported in 1 study). As for DL-mediated SRL, across nine studies, DL tools were reported to enhance learners’ self-regulation by supporting goal-setting, progress monitoring, and more autonomous study behaviors. Several studies attributed improved SRL to features such as reminders, feedback visualizations, or collaborative platforms. For example, learners using SRL dashboards “more actively read online” due to feedback and monitoring tools (Chen & Huang, 2014).

 Mobile-assisted learning environments similarly promoted autonomy, encouraging students to “study without teacher intervention” (Kondo et al., 2012) as well as self-directed and outside the classroom vocabulary learning (Xodabande & Atai, 2020). Collaborative writing platforms also strengthened planning and monitoring behaviors, as students could control, manage, and plan their writing activities in digital spaces (Rahimi & Fathi, 2021). AI-powered systems were also found to support SRL by enabling learners to assume greater control over their learning process (Wei, 2023).

Moderating Factors between SRL and DL

Two major categories of moderating factors emerged: learner-related factors (46 studies) and educational factors (40 studies). Learner-related factors were multi-dimensional and strongly shaped how students engaged in technology-assisted SRL. Motivation—particularly perceived usefulness, interest, value, and intrinsic enjoyment—was a consistent predictor of engagement (An et al., 2023; An et al., 2021; García Botero et al., 2018). Self-efficacy, both linguistic and technological, similarly supported strategy use, monitoring behaviors, persistence, and overall performance (Bai et al., 2022; Chen & Hsu, 2020; Chung, 2015). Learners with stronger SRL foundations also performed better in comprehension, vocabulary, and annotation tasks (Chen & Huang, 2014; Chen et al., 2014; Chen et al., 2019). Digital literacy and awareness of technological affordances influenced how efficiently learners regulated their learning (Lai & Gu, 2011; Lai & Zheng, 2018), whereas emotional variables such as enjoyment, attitudes, and interest contributed secondary yet meaningful support (An et al., 2021; Liu et al., 2014). In addition, goal orientation and motivational regulation shaped depth and sustainability of SRL behaviors, with mastery-oriented learners demonstrating more robust engagement (An et al., 2023; Aynagoz & Unal, 2024).

Educational factors—instructional, contextual, and environmental—also played a critical role. Scaffolding, including prompts, guidance, structured support, and teacher mediation, repeatedly emerged as essential for sustaining motivation and effective SRL (Hromalik & Koszalka, 2018; Kondo et al., 2012; Lai, 2015). Opportunities for interaction with peers, teachers, or technology enhanced monitoring, engagement, and strategic learning (Figura & Jarvis, 2007; Lai, 2013; Lai & Gu, 2011). Broader contextual conditions, such as resource availability and technological ecosystems, shaped the feasibility of SRL (García Botero et al., 2018; Lai & Zheng, 2018), while well-designed digital environments fostered engagement through interactive, user-friendly, and gamified features (Lai et al., 2022; Liu et al., 2014; Dehghan & Mobki, 2025). Facilitating conditions—stable access, usability, tool availability, and technical support—enabled sustained learning (Chang, 2005; Chen, 2014), and training or familiarization activities provided the necessary confidence and competence (Hromalik & Koszalka, 2018; Kondo et al., 2012). Instructional approaches that featured modeling, student-centered design, and clear task structures also reinforced self-regulation by aligning pedagogical goals with technological behaviors (Lai, 2015; Lai, 2013).

These patterns demonstrate that SRL in DL environments relies heavily on both individual dispositions and contextual guidance. The strong moderating influence of motivation and self-efficacy reflects longstanding findings that SRL is intertwined with expectancy-value judgments and performance beliefs (Dörnyei & Ushioda, 2021; Eccles & Wigfield, 2020). Technology, therefore, does not generate SRL independently; rather, it amplifies what learners bring with them, consistent with prior work on metacognition and digital literacy (Azevedo & Cromley, 2004; Lai & Hwang, 2016). Educational factors further align with socio-cognitive perspectives in which regulatory skills develop through scaffolding and co-regulation (Hadwin et al., 2018; Vygotsky, 1978). As DL tools—particularly AI-driven ones—grow increasingly complex, teachers face the challenge of making informed pedagogical decisions while managing rapidly evolving AI literacy demands (Crompton et al., 2024). The interplay between technology design, teacher mediation, and contextual support underscores the need for learning-centered design and highlights the crucial role that educators play in helping learners develop the digital literacy required for effective SRL (Laurillard, 2012).

Relationship between TBSRL and Language Learning Outcomes

Three key themes emerged regarding the relationship between TBSRL and language learning: achievement, language skills and components, and self-directed technology use.

Achievement

Across studies, TBSRL strategies were consistently associated with improved academic outcomes. Learners who actively set goals, monitored progress, and engaged with SRL-supportive technologies showed higher vocabulary, reading, writing, and grammar performance, with gains observed in both mobile-assisted and AI-supported environments (e.g., An et al., 2020; Wei, 2023).

Classroom-based applications such as flipped and blended instruction also enhanced overall language achievement by embedding SRL mechanisms into instructional design. Motivational and self-regulatory dispositions—such as mastery orientation and perceived usefulness—further contributed to achievement differences, whereas test-oriented approaches were linked to weaker engagement and poorer regulation.

Language Skills and Components

SRL-enabled digital tools positively influenced a wide range of language skills. Vocabulary learning showed the strongest and the most consistent improvements, particularly in mobile-assisted and adaptive learning environments (e.g., Chen & Chung, 2008). Reading comprehension benefited from SRL features such as goal-setting supports, annotation tools, and progress-monitoring dashboards.

Writing performance improved in SRL-scaffolded collaborative spaces and AI-supported platforms, which encouraged planning, revision, and sustained engagement (e.g., Rahimi & Fathi, 2021). Speaking and grammar gains were also reported in technology-enhanced settings that integrated SRL prompts or interactive practice opportunities. In several cases, multi-skill improvement occurred simultaneously when comprehensive SRL mechanisms were embedded into digital systems.

Self-Directed Technology Use

Learners’ willingness and ability to use technology autonomously were shaped mainly by motivational orientations and contextual conditions. Mastery-oriented learners engaged more consistently with digital tools and demonstrated more adaptive SRL behaviors (e.g., An et al., 2023). Facilitating conditions such as usability, access, and familiarity influenced technology use patterns, particularly for higher-achieving learners. Evidence also showed that students refined their goals and technology strategies over time and generally expressed positive attitudes toward ongoing, self-directed digital learning.

The positive association between SRL and language outcomes supports the longstanding claim that SRL plays a central role in successful L2 acquisition (Oxford, 2017; Tseng et al., 2006). What is notable in this synthesis is that SRL-linked gains emerged across multiple modalities—reading, writing, vocabulary, speaking, and grammar—indicating that technology may function as a cross-skill regulatory enhancer. This aligns with meta-analytic findings that digital environments can promote both strategic engagement and linguistic outcomes when designed to facilitate self-monitoring and feedback use (Rubio, 2015; Sung et al., 2017).

The presence of mixed or negative findings, especially regarding performance-goal orientations or test-driven behaviors, parallels concern in motivational research about maladaptive strategic engagement under performance pressures (Diaconu-Gherasim et al., 2024). These patterns suggest that not all SRL is equally beneficial; the quality of regulation (i.e., its adaptiveness, depth, and alignment with meaningful goals) may be a stronger determinant of L2 learning success than a mere frequency of strategy use.

SRL Strategies

Across the reviewed studies, learners frequently employed several SRL strategies in technology-supported language learning. The most common were goal-setting, planning, and evaluation (e.g., An et al., 2021; Chen & Huang, 2014; Tao et al., 2020), alongside self-monitoring, self-recording, and self-assessment (e.g., Chang, 2005; Öztürk & Çakıroğlu, 2021). Learners also reported frequent use of cognitive strategies, including reading and listening strategies, repetition, inferencing, and vocabulary learning (e.g., Figura & Jarvis, 2007; Teng & Zhang, 2024). Additionally, resource and time management strategies were commonly used in online and mobile settings (e.g., Chung, 2015; Kondo et al., 2012), and several studies noted help-seeking as a recurrent or predictive strategy in specific contexts (e.g., Karacan et al., 2022; Yazdani & Sadeghi, 2022).

In contrast, social strategies—such as peer interaction, online collaboration, and seeking feedback—were consistently identified as the least used, even when digital platforms provided opportunities for communication (e.g., Figura & Jarvis, 2007; Lai & Zheng, 2018). Some studies also reported low engagement in activities such as social networking, note-taking, or commenting after viewing instructional videos (e.g., Wang & Chen, 2019). In certain contexts, metacognitive planning, especially planning before computer-based tasks, was also relatively underused (e.g., Figura & Jarvis, 2007).

These patterns align with broader L2 strategy research. As Oxford (2017) notes, frequent use of goal-setting, monitoring, and cognitive strategies reflects typical behavior of strategic learners, suggesting that digital tools do not inherently generate SRL. Instead, learners who already employ metacognitive strategies appear better positioned to leverage the affordances of technology- or AI-assisted language learning. Thus, SRL-capable learners regulate their learning more effectively with technological support.

The consistent underuse of social and collaborative strategies mirrors Lai’s (2015) observation that learners often struggle to engage meaningfully with peers in DL contexts. These barriers are not purely technological but also motivational and socio-cognitive: learners may be unsure how to collaborate, may not view peer interaction as central to language learning, or may find digital spaces lacking the immediacy needed for sustained social engagement. From the perspective of co-regulation and socially shared regulation of learning, effective SRL development is not solely an individual process but also emerges through interaction, negotiation, and collective monitoring with others (Hadwin et al., 2018; Vygotsky, 1978). The limited use of social strategies therefore suggests that many DL environments may support individual regulation more effectively than collaborative regulation. Even though DL environments offer strong social affordances, actual adoption remains limited. Research shows that social strategies appear far less frequently than cognitive or metacognitive ones in self-directed mobile learning (Lai et al., 2022), that meaningful online engagement improves when scaffolding is provided (Kangwa et al., 2024), and that collaborative creativity requires intentional instructional design rather than emerging naturally from technology (Selfa-Sastre et al., 2022).

Benefits, Risks, and Limitations of DL

Across the reviewed studies, several recurring benefits of DL were identified. Learners consistently reported flexibility, personalization, and anytime–anywhere access, along with tailored or immediate assistance (e.g., Chen & Chung, 2008; Lai & Zheng, 2018). Digital tools and AI systems offered personalized feedback, abundant practice opportunities, and real-time suggestions, supporting various aspects of language development (e.g., Du & Alm, 2024; Wang et al., 2021). Many studies also highlighted increased autonomy, motivation, and engagement, facilitated by user-friendly or adaptive environments (e.g., An et al., 2023; Wei, 2023), as well as reduced anxiety, particularly in speaking and writing (e.g., Guan et al., 2025; Qiao & Zhao, 2023). Additional advantages included convenience, enhanced learner control, and improved performance in areas such as vocabulary and grammar (e.g., Kondo et al., 2012; Xodabande & Atai, 2020).

Fewer studies identified risks, but several recurring concerns emerged. Some reported novelty effects, where initial enthusiasm declined over time (e.g., García Botero et al., 2018; García Botero et al., 2019). Other studies warned of overreliance on AI, which could undermine learning processes or language skill development (e.g., Li & Kim, 2024; Wang et al., 2024). Reduced human interaction was also noted, with mixed experiences regarding social presence when engaging primarily with AI (e.g., Du & Alm, 2024; Guan et al., 2025). Risks further included academic integrity issues, distraction, and uncertain or misaligned AI feedback (e.g., Liu et al., 2024a; Liu et al., 2024b).

Limitations were the most frequently reported theme. Learners described time constraints, difficulties in self-monitoring, and challenges using built-in SRL features (e.g., García Botero et al., 2018; Karacan et al., 2022). Tool-related constraints such as cognitive overload, limited interactivity, and formal or time-consuming platforms were also common (e.g., Guan et al., 2025; Lai & Gu, 2011). Several studies pointed to technological constraints, including device-related issues and mismatches between tool design and learner needs or proficiency levels (e.g., Lai & Zheng, 2018). Another recurring limitation involved dependence on teacher support, with some systems requiring human intervention to clarify AI feedback or sustain engagement (e.g., Han et al., 2021; Liu et al., 2024b). Problems such as irregular usage, fluctuating motivation, and resource management difficulties—particularly when navigating multiple platforms—further constrained learners’ DL experiences (e.g., García Botero et al., 2018; Wang et al., 2024).

These findings align with broader work on AI-mediated and out-of-class learning. Benefits such as personalization, autonomy, emotional support, and engagement reflect the motivational potential of adaptive and learner-centered technologies (Banihashem et al., 2025; Khasawneh et al., 2025; Reinders & Benson, 2017; Wei, 2023). At the same time, the risks and limitations extend beyond novelty effects to include cognitive and information overload and reduced social connection. Prior research similarly notes that engagement tends to decline over time (Miguel-Alonso et al., 2024; Tsay et al., 2018), online classes can impose cognitive load (Warrick, 2021), and while AI-mediated learning may lower anxiety (Moradi & Zarei, 2025), it can also compromise meaningful interaction if insufficiently supported (Jose et al., 2025). Ultimately, these challenges emphasize that technology alone cannot ensure effective SRL; instead, learners require deliberate scaffolding, thoughtful design, and opportunities for authentic interaction.

Integrated Conceptual Model

Based on the synthesized themes, an integrated conceptual model was developed to illustrate how digital learning features, learner-related factors, and educational conditions interact through SRL to shape L2 outcomes. The diagram (Figure 2) represents the final interpretive output of the meta-synthesis and was created using \LaTeX{} (TikZ) for clarity and precision.

Figure 2.

The technology-mediated SRL mediation model with dual moderation.

Explanation of the Integrated Conceptual Model

The integrated conceptual model synthesizes how DL environments, learner-related factors, and educational conditions interact through SRL to shape L2 outcomes. DL features include affordances such as feedback systems, AI support, usability, and personalization, while learner-related factors involve motivation, self-efficacy, goal orientation, and digital literacy. Educational factors refer to scaffolding, teacher mediation, interaction opportunities, and instructional design.

At the center of the model, SRL is represented through metacognitive, cognitive, and resource-management processes. The arrows indicate that DL features influence language outcomes mainly through SRL, while learner-related and educational factors moderate the effectiveness of this relationship. The model extends Zimmerman’s (2000) SRL framework by situating self-regulation within digitally- and socially-mediated learning environments. The synthesis suggests that SRL development depends not only on individual regulation but also on technological affordances, pedagogical support, and contextual conditions.

Conclusion and Implications

Regarding the first research question, the findings indicate a generally positive relationship between SRL and the use of DL and GenAI tools in L2 contexts, particularly through support for planning, monitoring, and autonomous learning. Second, this relationship was strongly moderated by learner-related and educational factors such as motivation, self-efficacy, scaffolding, and instructional design. Third, TBSRL was consistently associated with improved language learning outcomes across multiple skills and achievement measures. Fourth, metacognitive and cognitive strategies were the most frequently reported, whereas social and collaborative strategies were comparatively underused. Finally, while DL and GenAI environments offered benefits such as flexibility, personalization, and increased engagement, they also presented risks and limitations related to overreliance on technology, reduced interaction, and uneven learner support.

This meta-synthesis shows that TBSRL is a multidimensional process shaped by the interaction of learners’ abilities, digital affordances, and instructional design. Digital tools can support planning, monitoring, and evaluation by offering feedback, reminders, and flexible learning spaces, and they often strengthen learners’ sense of control and autonomy. However, effective regulation does not arise from technology alone. Learners’ motivation, confidence, and existing regulatory habits play a central role in determining how well they engage with digital tools and sustain strategic learning over time.

Instructional Design

Instructional conditions also strongly influence TBSRL. Scaffolding, interaction, and user-friendly design help learners manage cognitive demands and engage more fully in regulatory processes, highlighting the importance of embedding SRL-oriented support within digital environments. When these supports are present, technology-mediated regulation is more likely to translate into meaningful language learning gains across skills. Structured prompts and opportunities for reflection appear particularly important for sustaining strategy use.

Learner Differences

Differences in proficiency, motivation, and SRL readiness can shape how learners benefit from digital tools, raising important considerations for equitable implementation. Learners with stronger motivation and prior regulatory habits tend to engage more productively, whereas others may underuse key strategies without explicit guidance. These patterns suggest the need for differentiated supports, including motivational scaffolds and graduated SRL guidance, to ensure that technology-enhanced learning environments remain accessible and effective for diverse learners.

Technology Considerations

Although digital environments offer clear advantages, they also present challenges, including reduced interaction, declining engagement over time, and tools that do not always match learners’ needs. These issues underscore the need for careful tool selection and realistic expectations about what technology can accomplish. Effective TBSRL depends less on the mere presence of digital tools and more on how well technological affordances align with pedagogical goals and learner characteristics.