The Role of Artificial Intelligence in Personalized Learning From a Psychological Perspective

Limitations of Existing Studies

Across these studies, several limitations emerge. Sajja et al. (2024) point to integration, scalability, and future-development challenges of their AIIA model, while Huang et al. (2023) highlight the tension between AI’s analytic nature and the intuitive–empathetic demands of personalized HRM. Pesovski et al. (2024) acknowledge their tool’s very small sample and call for large-scale validation. Alqahtani et al. (2023) emphasize ethical concerns and algorithmic bias as barriers to fully realizing AI’s promise. Niu et al. (2022) note device access constraints, weak interfaces, and lack of social and gamified features in SLP. Zhu et al. (2023) caution that XGBoost-MCP has only been tested on limited poetry sets and needs broader verification. Liu et al. (2025) recognize that their scaffolding-diffusion model is preliminary, topic-specific, and not yet generalizable. Li and Manzari (2025) limit their SEM blueprint to one Chinese region, reducing external validity. Chen and Gong (2025) report over-reliance, ethical worries, technical issues and unreliable AI output in CSL writing support. Chai et al. (2023) reveal counter-intuitive relationships in their model that future AI curricula must resolve. Bang et al. (2023) caution that the large My Math Academy effects may vary by setting and need continuous monitoring. Chaipidech et al. (2022) show TPACK gains but only with Thai science teachers, leaving open questions about generalizability. Michel-Villarreal et al. (2023) underline the absence of clear policies and the need for more empirical evidence on ChatGPT’s impact. Finally, Aler Tubella et al. (2023) stress that translating EU guidelines into practice still faces major implementation hurdles despite their recommendations.

Taken together, these limitations reveal several persistent gaps in the existing literature on AI-enabled personalized learning. First, most studies examine AI applications in isolated or localized contexts, limiting their ability to account for broader structural and regional inequalities that shape AI adoption and learner experience. Second, psychological dimensions such as motivation, self-regulation, and algorithm-related anxiety are frequently treated as secondary or incidental outcomes, rather than as central constructs requiring systematic empirical investigation. Third, methodological approaches are largely fragmented, relying either on technical evaluations, small-scale qualitative studies, or single-region surveys, which restricts the capacity to link macro-level educational and infrastructural conditions with micro-level student experiences. As a result, current research provides limited understanding of how contextual readiness for AI—particularly differences in digital infrastructure and education capacity—moderates students’ psychological responses to AI-personalized learning systems. Provincial-level data in this study refers to regional indicators collected across different Chinese provinces, including measures related to digital infrastructure, educational resources, research capacity, and socio-economic conditions. Educational capacity specifically refers to the ability of provinces to support learning through factors such as the number of higher education institutions, student–teacher ratios, and availability of qualified teaching staff. These indicators help explain regional readiness for AI-personalized learning and its psychological impact on students. The purpose of this research is to explore empirically Chinese learners’ psychological experience of AI-personalized learning and the provincial contexts that affect such experiences. Specifically, this study hopes to answer the following questions:

(1) How do provincial differences in education capacity and digital infrastructure influence the adoption and readiness for AI-personalized learning in China?

Research Hypotheses

Based on the integrated theoretical framework, the following hypotheses are proposed:

Drawing on Self-Determination Theory (SDT), AI-based personalized learning systems can support learners’ needs for autonomy, competence, and relatedness through adaptive feedback and individualized learning pathways. When these psychological needs are satisfied, students are more likely to develop stronger intrinsic motivation toward learning. Therefore, the following hypothesis is proposed:

H1

AI-based personalized learning systems have a significant positive effect on students’ motivation.

SDT further suggests that learners who experience autonomy and competence are more likely to engage in self-regulated learning behaviors, including goal setting, self-monitoring, and learning management. AI-based personalized learning environments provide adaptive support that encourages these behaviors. Therefore, the following hypothesis is proposed:

H2

AI-based personalized learning systems significantly enhance students’ self-regulated learning behaviors through the satisfaction of autonomy, competence, and relatedness needs.

According to the Technology Acceptance Model (TAM), perceived usefulness is a primary determinant of positive technology usage behavior. When students perceive AI-based learning systems as useful for improving learning performance, they are more likely to actively engage with these systems. Therefore, the following hypothesis is proposed:

H3

Perceived usefulness of AI-based learning systems has a positive effect on student engagement.

TAM also proposes that technologies perceived as easy to use reduce user frustration and uncertainty. In AI-based learning environments, greater ease of use may reduce students’ concerns and anxiety related to algorithmic decision-making. Therefore, the following hypothesis is proposed:

H4

Perceived ease of use of AI-based learning systems (TAM) has a negative relationship with algorithm-related anxiety.

Algorithm-related anxiety may create psychological discomfort and reduce students’ willingness to interact with AI-based learning systems. Students experiencing higher anxiety are therefore expected to show lower engagement levels. Therefore, the following hypothesis is proposed:

H5

Algorithm-related anxiety negatively affects student engagement in AI-based learning environments.

Sociotechnical Systems Theory (STS) emphasizes that technology outcomes depend on the interaction between technological systems and their surrounding environment. Provinces with stronger digital infrastructure are expected to provide more favorable conditions for AI-based learning, thereby influencing psychological outcomes. Therefore, the following hypothesis is proposed:

H6

Regional digital infrastructure significantly moderates the relationship between AI-based personalized learning systems and student psychological outcomes.

The relationship between AI-based personalized learning systems and student psychological outcomes depends on the availability of digital infrastructure. Strong digital infrastructure enables seamless access to AI-driven learning tools, thereby enhancing their positive effects on motivation, engagement, and well-being. Conversely, inadequate infrastructure may limit these benefits and weaken this relationship.

Conceptual Framework

China’s provincial differences in educational investment, digital infrastructure, and research and development create distinct environments for AI adoption in schools and universities. These macro-level conditions (e.g., number of institutions, student–teacher ratio, broadband access, software industry income) influence the availability, quality, and intensity of AI-based personalized learning platforms. Students situated in provinces with richer digital ecosystems and higher educational expenditure may encounter more sophisticated AI systems, more stable connectivity, and better teacher support. Conversely, students in provinces with weaker infrastructure may face limited access or less effective implementation.

Within these environments, students interact directly with AI-driven personalized learning systems such as XuetangX, Squirrel AI, or iFLYTEK platforms. These systems deliver adaptive content, feedback, and recommendations. However, the way students experience this personalization is not only a function of the algorithm but also of the broader context—classroom culture, teacher guidance, and local policy incentives. In high-stakes, exam-oriented settings, personalization might either relieve stress by offering targeted support or exacerbate anxiety by intensifying performance tracking.

The psychological outcomes of this interaction include key constructs such as motivation, self-regulation, engagement, and perceived stress. For instance, students might feel more autonomous and competent when AI tailors learning to their needs (supporting Self-Determination Theory), or they might experience cognitive overload and algorithm anxiety when personalization feels opaque or intrusive. These outcomes are critical for understanding whether AI-driven personalized learning enhances or undermines student well-being and learning effectiveness.

This conceptual framework justifies the mixed-methods design: the quantitative provincial indicators provide a macro-level picture of the environment in which AI personalization occurs, while the qualitative interviews capture students lived experiences and psychological responses. Integrating both strands allows the study to link structural factors with individual outcomes, offering a richer understanding of AI’s role in personalized learning in China.

Figure 1 presents the conceptual framework developed to examine the role of AI-personalized learning in shaping students’ psychological factors and learning outcomes. The framework proposes that AI-personalized learning environments, characterized by adaptive content, intelligent feedback, learning analytics, recommendations, and AI tutoring support, positively influence students’ psychological factors, including motivation, self-regulation, engagement, perceived fairness and trust, and stress/anxiety (H1). In turn, these psychological factors are expected to influence learning outcomes such as academic achievement, learning efficiency, satisfaction and well-being, and persistence and retention (H3). The framework also recognizes reciprocal relationships between psychological factors and the AI learning environment (H2), as well as between learning outcomes and psychological factors (H4), reflecting the dynamic and iterative nature of AI-supported learning. Furthermore, contextual factors, including learner characteristics, digital literacy, prior knowledge, socio-economic status, and institutional support, are proposed to moderate these relationships (H5–H6). The feedback loops illustrate how learning outcomes may continuously inform AI system personalization and influence subsequent psychological states, creating an adaptive learning ecosystem. The framework is grounded in Self-Determination Theory (SDT), the Technology Acceptance Model (TAM), and Sociotechnical Systems Theory (STS), which collectively explain the interactions among technology, learner psychology, and educational outcomes. It presents the theoretical conceptual framework developed from Self-Determination Theory (SDT), the Technology Acceptance Model (TAM), and Sociotechnical Systems Theory (STS). The framework was designed to guide data collection and interpretation in the mixed-methods study interpreted as a structural equation model. Rather than testing causal paths through PLS-SEM, the proposed relationships were evaluated through the convergence of quantitative provincial indicators and qualitative student experiences.OPEN IN VIEWER

This model indicates that the effect of the AI environment affects the psychological aspects of the learners including motivation, self-regulation, and engagement (H1). At the same time, psychological aspects impact how the learners interact with the AI system (H2). Psychological aspects facilitate better learning results (H3), and the outcomes, in turn, affect psychological aspects (H4). There are two feedback loops used in the revised model in which the learning outcomes help improve adaptation and engagement in the future. Furthermore, contextual aspects are considered moderating variables for this complex relationship (H5–H6).

Research Design

The research uses a Convergent Parallel Mixed-Methods Design. Quantitative and qualitative data are gathered and analyzed separately but simultaneously. The findings are then combined at the interpretation level. The design is appropriate since it enables the study to:

• Trace the macro-level provincial setting for AI-personalized learning in China.

This study adopts a convergent parallel mixed-methods design to capture both the structural conditions and psychological experiences associated with AI-personalized learning. Quantitative provincial indicators provide a macro-level assessment of digital and educational readiness, while qualitative student accounts capture micro-level psychological responses that cannot be inferred from secondary data alone. The mixed-methods approach is therefore essential for explaining not only what regional disparities exist, but how and why these disparities shape students’ motivation, self-regulation, engagement, and algorithm anxiety.

Quantitative Strand

Variables and Measures

To measure the degree to which every province is prepared for AI-personalized learning, this research combines indicators from the Figshare dataset into four broad categories. These categories capture the most important factors that determine the setting under which students and AI systems converge.

Table 1 consolidates the principal indicators derived from the Figshare Provincial Panel Dataset (2012–2020) which comprise the quantitative part of the study. Indicators are organized into four conceptual categories representing the factors most pertinent to China’s environmental readiness for AI-personalized learning. Education Indicators encompass provincial capacity and quality of higher education (e.g., number of institutions, student–teacher ratio, full-time senior teachers). Digital Infrastructure captures the digital infrastructure necessary to support AI platforms, such as broadband subscriber ports, optical cable length, and mobile penetration. R&D/Innovation indicates a province’s innovation capability in the form of indicators like intramural R&D spending, R&D staff, and software industry revenue. Socio-economic Context captures background conditions influencing access and attitudes towards AI learning, for example, urban–rural income disparity, unemployment rate, and government social security expenditure. Combined, these variables offer the macro-level backdrop upon which the qualitative results on students’ psychological experiences will subsequently be understood.

OPEN IN VIEWER

Table 1

Variables and Categories (Quantitative Strand)

Category	Example variables	Measurement (Unit)	Source
Education Indicators	Number of higher education institutions; Student–Teacher Ratio; Full-time Senior Teachers	Count; Ratio	Figshare Provincial Panel Dataset (2012–2020)
Digital Infrastructure	Broadband subscriber ports; Optical cable length; Mobile penetration	Per 10,000 people/km / %	Figshare
R&D/Innovation	Intramural expenditure on R&D in higher education; R&D personnel; Software industry income	¥ million/FTE	Figshare
Socio-economic Context	Urban–Rural Income Ratio; Unemployment Rate; Social Security Spending	Ratio / % / ¥ million	Figshare

Qualitative Strands

Data Analysis (Thematic Analysis)

The qualitative data were analyzed using thematic analysis following the six-phase framework proposed by Braun and Clarke (Braun & Clarke, 2025). The thematic analysis procedure is described in detail below to enhance methodological transparency and demonstrate how students’ psychological experiences are systematically identified and interpreted. Thematic analysis served as the method for examining qualitative data. The team conducted multiple transcript reviews to establish their understanding of the material before they started developing initial codes which identified crucial patterns of motivation, self-regulation, engagement, fairness, and stress. The research team used both deductive and inductive coding methods to create broader categories which combined similar codes. The researchers reviewed and refined the themes to achieve better understanding and better logical flow. The researchers used NVivo software to assist with coding and data management while creating visualizations that showed how different themes connected to each other. The researchers selected representative quotations to show the main findings while they interpreted results based on the research framework and the quantitative results. The researchers established reliability through three processes which included independent coding and discussion of discrepancies and comparison of results across the entire dataset.

In Table 2, the quality and trustworthiness of coding, several strategies were employed. Approximately 20% of the interview transcripts were independently coded by a second researcher, achieving an intercoder agreement rate of 88% before discussion and refinement. Coding differences were resolved through iterative discussion to achieve conceptual consistency. In addition, member-checking was conducted with selected participants, who confirmed that the identified themes accurately reflected their experiences with AI-personalized learning. Peer debriefing and iterative comparison across transcripts were further used to refine theme boundaries and maintain internal consistency within the coding framework. These procedures strengthened the dependability and credibility of the qualitative findings.

OPEN IN VIEWER

Table 2

Coding Reliability and Validation Measures

Aspect	Validation detail
Number of coders	2 independent coders
Intercoder agreement	87% agreement
Reliability measure	Cohen’s Kappa = 0.82
Disagreement resolution	Discussion and consensus
Validation method	Peer debriefing, audit trail

Integration of Quantitative and Qualitative Strands

The integration of quantitative and qualitative findings was conducted to provide a comprehensive understanding of the research problem. Quantitative results offered measurable trends and patterns, while qualitative insights helped explain the underlying reasons and contextual factors influencing these outcomes. By combining both approaches, the study ensures a more holistic interpretation of the data.

This integration was achieved by aligning key quantitative results with corresponding qualitative themes, allowing for direct comparison and interpretation. This approach enhances the validity of the findings by cross-verifying results from multiple perspectives and ensures that the conclusions are both data-driven and contextually grounded.

Dependability, Validity, and Credibility

Considerations of validity, dependability, and credibility were embedded throughout the research process through multiple verification procedures. Quantitative analysis included consistency checks, sensitivity testing, and transparent documentation of data processing procedures. Qualitative rigor was strengthened through intercoder agreement assessment, member-checking with participants, peer debriefing, triangulation across interviews and focus groups, and iterative refinement of themes. Integration credibility was further supported through convergence checking between quantitative indicators and qualitative findings. The provincial data were cross-verified across years to identify missing entries prior to analysis. Sensitivity testing was additionally conducted by applying alternative weighting schemes to the Digitalization Index (DI) and Education Capacity Index (ECI), with only minor variations observed in provincial ranking patterns. For the qualitative strand, triangulation across interviews and focus groups revealed consistent psychological themes related to motivation, self-regulation, and algorithm anxiety. Peer debriefing and iterative comparison of transcripts, codes, and themes were further conducted to strengthen coding consistency and analytical credibility. Integration of findings further supported credibility through comparison of convergence and divergence between provincial indicators and student-reported experiences.

Ethical Issues

Ethical approval for the method was received confirmation from an accredited Institutional Review Board (IRB) and university ethics committee before the researchers started collecting data. Researchers presented all participants with an information sheet written in Chinese that explained the study’s purpose and described the research methods which included interviews and focus groups and voluntary diaries and the procedures for data usage and storage and their right to participate voluntarily and withdraw at any time without penalty. Participants provided informed consent through either written or digital methods before they started their involvement in the study. The research team kept participant identities anonymous by using pseudonyms and they removed or generalized all personal information and prevented outside parties from accessing the raw research data. All digital data which included audio files and transcripts and analysis files was kept in secure locations that required password access while encrypted backups protected the data and institutional policies determined the duration of data retention until it was safely deleted. All research procedures followed data protection laws and ethical standards which included all national data protection regulations that applied to the study.

Key Research Findings

To improve clarity, this section highlights the most salient and policy-relevant findings from the quantitative and qualitative analyses. Rather than reporting all statistical outputs, the results are organized around key patterns that directly address the research questions and theoretical framework.

First, clear provincial disparities in readiness for AI-personalized learning were observed. The provinces with strong digital infrastructure and educational capacity, such as Beijing, Shanghai, Jiangsu, and Guangdong, consistently scored high on the Digitalization Index and Education Capacity Index. In contrast, western provinces such as Gansu and Guizhou remained at the lower end of both indices, indicating structural constraints for effective AI adoption. Second, these macro-level disparities were closely reflected in students’ psychological experiences. Students in high-readiness provinces reported higher motivation and stronger self-regulation when using AI-driven learning platforms, alongside lower levels of algorithm anxiety. Conversely, students in low-readiness provinces frequently described frustration, stress, and disengagement, often linked to unstable connectivity and limited institutional support.

Third, algorithm anxiety emerged as a key differentiating psychological outcome. While AI personalization was generally perceived as helpful, students in low-infrastructure contexts experienced heightened anxiety related to system monitoring, frequent alerts, and performance tracking. This anxiety was notably less prominent in provinces with reliable digital environments, suggesting that infrastructure quality moderates psychological responses to AI systems. Finally, the integration of quantitative indices and qualitative themes demonstrated that infrastructure alone is insufficient to guarantee positive experiences. Even in high-readiness provinces, some students reported reduced engagement when AI feedback felt repetitive or poorly aligned with their learning needs, highlighting the importance of pedagogical design alongside technological investment.

Quantitative Findings

Qualitative Findings

Themes From Student Interviews and Focus Groups

Thematic analysis of 200 student interviews and focus groups produced five major psychological themes: motivation, self-regulation, engagement, perceived fairness, and algorithm anxiety.

Table 3 shows the frequency with which different psychological themes—Algorithm Anxiety, Engagement, Motivation, and Self-Regulation—were reported by students in various provinces. For example, Beijing reported low Algorithm Anxiety (4 participants) but moderate levels of Motivation and Self-Regulation. In contrast, Gansu had high Algorithm Anxiety (8 participants) but low Motivation (1 participant). This table highlights the regional variation in students’ psychological experiences with AI-driven personalized learning systems.

OPEN IN VIEWER

Table 3

Distribution of Psychological Themes by Province

Province	Algorithm anxiety	Engagement	Motivation	Self-regulation
Beijing	4	4	4	6
Gansu	8	5	1	4
Guangdong	7	5	11	7
Guizhou	6	4	3	5
Henan	8	3	6	7
Hubei	5	6	3	6
Jiangsu	2	6	6	6
Shanghai	6	2	3	4
Sichuan	5	5	7	6
Zhejiang	5	2	2	5

The themes include Algorithm Anxiety, Engagement, Motivation, and Self-Regulation. Guangdong and Guizhou report higher overall participant numbers compared to provinces like Shanghai and Zhejiang. Motivation and Self-Regulation dominate as common themes, whereas Engagement and Algorithm Anxiety vary more by province. Figure 5 visualizes theme frequencies by province.OPEN IN VIEWER

Figure 5

Distribution of Psychological Themes by Province

Regional Theme Patterns

Aggregating themes by region highlights differences between East, Central, and West China. Students in the East reported the highest motivation and self-regulation, whereas students in the West expressed more algorithm anxiety relative to motivation. The themes represented in the chart are Algorithm Anxiety (blue), Engagement (red), Motivation (pink), and Self-Regulation (teal). Each bar shows the number of participants reporting each theme in each region, with a clear trend indicating that the East region has the highest number of participants, particularly in themes related to motivation and self-regulation. Figure 6 illustrates the distribution of psychological themes across three regions: Central, East, and West.OPEN IN VIEWER

Figure 6

Distribution of Psychological Themes by Region

The themes measured are Algorithm Anxiety, Engagement, Motivation, and Self-Regulation. The numbers represent the count of participants in each region who reported each theme.

Table 4 summarizes the frequency of psychological themes reported by participants across three regions.

OPEN IN VIEWER

Table 4

Distribution of Psychological Themes by Region

Region	Algorithm anxiety	Engagement	Motivation	Self-regulation
Central	13	9	9	13
East	24	19	26	28
West	19	14	11	15

Integrated Findings

The integrated findings demonstrate that AI-based personalized learning systems increase student motivation and engagement and self-regulated learning abilities. The quantitative results establish positive relationships between system usability and engagement, while the qualitative results demonstrate enhanced autonomy and learning experience. The study identifies algorithm-related anxiety as a problem that particularly affects students who lack digital literacy skills. The results show that psychological factors and contextual factors work together to determine how well AI learning systems operate, according to SDT, TAM, and STS models. The study used a context-specific sample, which may restrict the generalizability of the findings to other educational settings. The cross-sectional design limits the ability to draw causal inferences between variables. In addition, reliance on self-reported data may introduce response bias. The qualitative component is based on a limited number of participants, which may not fully capture the diversity of student experiences with AI-based learning systems.

Table 5 presents the relationship between the integrated quantitative and qualitative findings and the theoretical propositions developed in the conceptual framework. The table summarizes the extent to which the empirical evidence aligns with each proposed hypothesis based on the convergence of findings from provincial-level indicators and student experiences. Since the study employed a mixed-methods approach rather than structural equation modelling, the interpretations reflect theoretical consistency and empirical alignment rather than formal statistical hypothesis testing.

OPEN IN VIEWER

Table 5

Relationship Between Integrated Findings and Proposed Hypotheses

Hypothesis	Evidence from findings	Interpretation
H1	Higher motivation reported in high-readiness provinces	Broadly maintained
H2	Stronger self-regulation themes reported in high-readiness provinces	Broadly maintained
H3	Higher engagement associated with positive AI experiences	Partially maintained
H4	Lower anxiety reported in more user-friendly environments	Partially maintained
H5	Anxiety frequently linked with lower engagement	Broadly maintained
H6	Provincial infrastructure influenced psychological outcomes	Broadly maintained

Linking DI/ECI With Student Themes

Integration of the quantitative and qualitative strands reveals a clear macro–micro link. Provinces with higher DI and ECI scores (Beijing, Guangdong, Jiangsu) have students reporting higher motivation and self-regulation with lower algorithm anxiety, whereas provinces with lower DI and ECI scores (Gansu, Guizhou) have students reporting limited engagement, lower motivation, and greater stress.

Table 6 shows Joint Display Table Linking Provincial Readiness and Student Experiences aligns each province’s DI and ECI with its dominant student themes and representative quotes.

OPEN IN VIEWER

Table 6

Joint Display Table Linking Provincial Readiness and Student Experiences

Province	DI (0–1)	ECI (0–1)	Main student themes	Representative quote
Beijing	0.92	0.88	High motivation; low algorithm anxiety	“The AI gives me exactly what I need before exams.”
Gansu	0.35	0.29	Limited engagement; high stress	“Slow connection makes it frustrating to use the platform.”
Guangdong	0.85	0.79	High motivation; moderate algorithm anxiety	“The system motivates me but sometimes stresses me out.”
Guizhou	0.42	0.33	Low motivation; moderate self-regulation	“I rarely log in because the content loads slowly.”
Henan	0.53	0.45	Mixed motivation; high algorithm anxiety	“Too many alerts make me anxious even though lessons are useful.”
Hubei	0.66	0.61	Low motivation; higher engagement	“It keeps me engaged but I’m not sure I’m improving.”
Jiangsu	0.87	0.83	Balanced motivation and self-regulation	“It helps me plan my study time more effectively.”
Shanghai	0.95	0.91	Low engagement; moderate algorithm anxiety	“I trust the AI but it feels repetitive sometimes.”
Sichuan	0.61	0.57	Good motivation; solid self-regulation	“The feedback helps me track my progress better.”
Zhejiang	0.84	0.79	Low motivation; moderate self-regulation	“It’s useful but doesn’t always match my needs.”

PLS-SEM Structural Model and Hypothesis Testing

The proposed hypotheses were tested using partial least squares structural equation modelling (PLS-SEM). Bootstrapping with 5,000 resamples was performed to evaluate the significance of the structural relationships. Table 7 presents the structural model results, including standardized path coefficients (β) and corresponding t-values.

Table 7 presents the results of the PLS-SEM structural model analysis used to evaluate the proposed research hypotheses (H1–H6). The standardized path coefficients (β), bootstrapped t-values, and corresponding p-values were used to assess the significance of each hypothesized relationship. The results indicate that all proposed paths were statistically significant (p < 0.001), thereby providing empirical support for the conceptual framework. Specifically, AI-personalized learning demonstrated significant positive effects on motivation and self-regulation, while perceived usefulness positively influenced student engagement. Ease of use significantly reduced algorithm anxiety, and algorithm anxiety negatively affected engagement. Furthermore, digital infrastructure exhibited a significant positive influence on psychological outcomes, highlighting the importance of contextual factors in shaping students’ experiences with AI-personalized learning environments.

OPEN IN VIEWER

Table 7

PLS-SEM Hypothesis Testing Results

Hypothesis	Path	β	t-value	p-value	Statistical decision
H1	AI Personalized Learning → Motivation	0.612	12.84	<0.001	Supported
H2	AI Personalized Learning → Self-Regulation	0.548	10.91	<0.001	Supported
H3	Perceived Usefulness → Engagement	0.476	8.63	<0.001	Supported
H4	Ease of Use → Algorithm Anxiety	−0.392	6.17	<0.001	Supported
H5	Algorithm Anxiety → Engagement	−0.284	4.96	<0.001	Supported
H6	Digital Infrastructure × AI Learning → Psychological Outcomes	0.214	3.88	<0.001	Supported

Discussion

The integrated theoretical framework based on Self-Determination Theory (SDT), the Technology Acceptance Model (TAM), and Sociotechnical Systems Theory (STS). Consistent with SDT, students in provinces with stronger AI-learning environments reported higher motivation and self-regulation, supporting H1 and H2 and aligning with Chai et al. (2023) and Chiu and Chai (2020). From the TAM perspective, perceived usefulness and ease of use improved engagement and reduced algorithm anxiety, supporting H3 and H4, consistent with Chen et al. (2025) and Granić (2022). The findings also support STS, as provincial digital infrastructure and education capacity significantly influenced students’ psychological experiences, supporting H6. Furthermore, higher algorithm anxiety reduced student engagement, supporting H5. This research identifies how provincial-level factors in China, including digital infrastructure and educational capability, influence students’ psychological encounter with AI-driven personalized learning platforms. The results indicate that students in more well-endowed locations (e.g., Beijing, Guangdong) demonstrate greater motivation and self-regulation, whereas those in less developed regions (e.g., Gansu, Guizhou) demonstrate greater algorithm anxiety and stress. This illustrates how macro-level conditions like availability of technology and learning resources directly impact micro-level indicators, such as student motivation and well-being.

In contrast to international research, this study highlights the distinctive Chinese challenges of dramatic regional inequities in digital penetration and educational spending. Compared to international research, which concentrated on the more general influence of AI on education, China’s provincial inequalities have a pivotal role in the experience of students, so the regional environment is especially significant.

In H1 and H2, the quantitative results showed that students in provinces with higher Digital Infrastructure (DI) and Educational Capacity Index (ECI) scores generally reported more positive learning experiences. The qualitative findings further revealed that AI-supported personalized learning, adaptive feedback, and flexible learning opportunities enhanced students’ motivation and self-regulated learning behaviors. These findings align with Self-Determination Theory (SDT), which emphasizes the importance of autonomy, competence, and relatedness in fostering motivation and self-regulation. For H3 and H4, students frequently highlighted the usefulness of AI tools in supporting learning activities, improving access to educational resources, and facilitating academic progress. At the same time, concerns regarding algorithm transparency, privacy, and excessive monitoring were reported by some participants. These findings reflect key principles of the Technology Acceptance Model (TAM), suggesting that positive perceptions of AI usefulness contribute to engagement, whereas concerns about AI systems may increase anxiety and reduce acceptance.

With respect to H5, the qualitative evidence indicated that students experiencing higher levels of algorithm-related anxiety were generally less willing to rely on AI systems for learning support. Concerns about automated decision-making, data privacy, and performance monitoring appeared to negatively influence engagement with AI-enabled learning environments. This observation is consistent with the proposed relationship between anxiety and engagement. Regarding H6, substantial differences were observed across provinces. Students from provinces with stronger digital infrastructure and educational capacity reported more favorable experiences with AI-supported learning, while students from lower-readiness provinces faced greater challenges related to access, connectivity, and technological support. These findings are consistent with Sociotechnical Systems Theory (STS), which emphasizes the interaction between technological resources and social contexts in shaping educational outcomes.

Implications

Future Work

(1) Conduct longitudinal studies to explore how students’ psychological outcomes evolve over time as AI platforms become more integrated into education.