Frontiers in Psychology 系统综述：AI 驱动的工作系统与员工倦怠，提出 AIMS 元系统框架

Abstract

Background:

The rapid integration of artificial intelligence (AI) into workplace systems is transforming job design and employee experiences. While AI promises efficiency gains, it also creates a paradox by simultaneously reducing workload and increasing psychological strain, leading to divergent effects on burnout.

Objective:

This study systematically reviews evidence on the relationship between AI exposure and employee burnout and develops the AI as Meta-System (AIMS) framework.

Methods:

Following PRISMA 2020, four bibliographic databases and Google Scholar were searched from inception to 31 March 2026. Forty-three peer-reviewed primary empirical studies were included: 20 measured burnout or an established burnout dimension, and 23 examined mechanisms, moderators, proxy outcomes, measurement, or implementation conditions. Methodological quality was appraised using the Mixed Methods Appraisal Tool at the criterion level.

Results:

Findings varied according to the functional role of AI and the study design. Assistive AI was generally associated with lower burnout or exhaustion in randomized and pre–post healthcare studies, whereas monitoring and algorithmic-control exposures were associated with greater burnout or related psychosocial strain in observational studies. Perceptual AI exposure produced direct, indirect, and null associations through pathways involving job stress, job insecurity, work–family interference, perceived organizational support, and organizational commitment. Three randomized studies provided the strongest evidence for assistive interventions. Evidence for nonlinear effects was limited to supporting proxy outcomes and did not directly establish an inverted U-shaped AI–burnout relationship.

Conclusion:

Artificial intelligence is not inherently harmful or beneficial; its associations with burnout depend on its functional role, implementation, and employee appraisal. The AIMS framework integrates these pathways and identifies propositions requiring longitudinal and experimental testing.

1 Introduction

The rapid diffusion of artificial intelligence (AI)–driven work systems is fundamentally transforming how work is organized, executed, and evaluated across industries. Organizations increasingly rely on algorithmic management, real-time monitoring, and AI-enabled decision support to enhance productivity, efficiency, and innovation. While these systems promise substantial operational gains, a growing body of empirical research suggests that they may also generate unintended negative consequences for employee psychological health (Stamate et al., 2021). Empirical studies indicate that AI awareness, adoption, and algorithmic control may be associated with job stress, job insecurity, poorer well-being, or burnout (Kong et al., 2021; Kim and Lee, 2024; Dong et al., 2025; Wang and Zhou, 2025). Conversely, ambient documentation and AI-tailored interventions have generally been associated with lower burnout or exhaustion when they reduce documentation and workload demands (Afshar et al., 2025; Baek and Cha, 2025; Chowdhury et al., 2026; Olson et al., 2025; You et al., 2025). Related quality-improvement evidence has also examined clinicians’ perceptions of work burden, burnout, and job satisfaction following ambient AI documentation (Albrecht et al., 2025). This mixed evidence suggests that AI does not have a uniform effect on employee well-being, but instead produces context-dependent outcomes. Review and conceptual sources are used in this article to establish the broader theoretical context but are not counted among the 43 primary empirical studies included in the systematic-review synthesis.

This divergence gives rise to a central paradox. AI can simultaneously reduce workload by automating routine tasks while increasing psychological strain through new demands such as continuous monitoring, uncertainty, and accelerated work pace. Empirical findings illustrate this contradiction. Workers exposed to automation risk may experience lower immediate stress yet report poorer health and reduced job satisfaction (Nazareno and Schiff, 2021), while AI adoption increases burnout indirectly through job stress mechanisms (Kim and Lee, 2024). Broader evidence has similarly linked automation probability with psychosocial working conditions and worker health (Cheng et al., 2021), while employee perceptions of smart technologies, AI, robotics, and algorithms have been examined as an important aspect of anticipated workplace change (Brougham and Haar, 2018). AI-related identity threat, job-automation awareness, and perceived replacement risk may also increase cognitive job insecurity and undermine employee well-being (Gull et al., 2023; Chung et al., 2025; Lingmont and Alexiou, 2020). AI adoption has also been examined in relation to psychological contracts, job engagement, and employee trust (Braganza et al., 2021). These findings raise a critical question: under what conditions does AI reduce burnout, and when does it amplify it?

Although prior research on technology and work has established links between digitalization and burnout, primarily through technostress and job demand frameworks, AI represents a qualitatively different class of technology. Unlike traditional systems, AI is adaptive, opaque, and capable of allocating tasks and monitoring performance in real time (Zerilli et al., 2022; Zheng et al., 2025). These characteristics introduce novel stressors, including algorithmic opacity, techno-unpredictability, and perceived loss of control (Issa et al., 2024; Kumar et al., 2024). Importantly, while existing reviews have examined AI and burnout in specific sectors such as healthcare (Dave et al., 2026) and have provided multilevel perspectives on AI in organizations (Bankins et al., 2024), no synthesis has integrated AI-specific mechanisms across occupations within a dynamic, multi-pathway theoretical framework. Consequently, the literature does not yet provide an integrated account of the conditions under which AI exposure is associated with burnout-related outcomes across occupational contexts.

This gap has significant theoretical and practical implications. Without a clear understanding of the mechanisms, moderators, and boundary conditions linking AI to burnout, organizations risk implementing AI systems in ways that undermine employee well-being and long-term productivity. Burnout is associated with substantial organizational costs, including turnover, reduced performance, and increased healthcare burden (Shanafelt et al., 2017; Martinez et al., 2025). Evidence suggests that poorly implemented AI systems, characterized by excessive monitoring, inadequate training, or unrealistic expectations, may exacerbate stress and disengagement, whereas human-centered implementations may mitigate these risks (Meduri et al., 2024). Thus, understanding how AI shapes burnout is both a theoretical imperative and a managerial necessity.

To address these limitations, we conceptualize AI not solely as a job demand or a job resource but as a work system that may simultaneously alter demands, resources, and control. The Job Demands–Resources framework provides a useful theoretical foundation for examining these domains (Bakker et al., 2023), while the proposed AIMS framework makes their concurrent and interactive operation explicit in AI-enabled work. The AIMS framework is therefore presented as a theory-building extension of existing models rather than as evidence that established occupational-health frameworks are inadequate.

Building on this premise, this study develops a dynamic, multi-pathway, contingency-based framework to explain how, when, and why AI-driven work systems influence employee burnout. It synthesizes interdisciplinary evidence to identify key mechanisms, including technostress, job insecurity, emotional strain, dependency, skill degradation, AI-related fear of missing out, work–family interference, and organizational-support pathways (Högemann et al., 2025; Méndez-Suárez et al., 2026; Zheng and Zhang, 2025; Wang and Zhou, 2025). It further examines moderators and boundary conditions, including organizational support, self-efficacy, and system design characteristics, while identifying critical gaps such as the lack of longitudinal designs and AI-specific measurement.

The primary conceptual contribution of this paper is the AI as Meta-System (AIMS) framework. The framework treats AI as potentially operating as a demand amplifier, a resource provider, and a control system and organizes proposed pathways involving workload, identity-related concerns, and temporal adaptation. It also specifies candidate individual and organizational moderators and proposes a dynamic adaptation trajectory requiring direct longitudinal testing. Rather than establishing causal mechanisms, the AIMS framework integrates the observed associations and generates research propositions for future experimental, longitudinal, and multi-wave studies.

2 Methods

2.1 Protocol and reporting standards

This systematic review was conducted in accordance with the PRISMA 2020 guidelines (Page et al., 2021) to ensure transparency, rigor, and reproducibility. A review protocol was developed a priori to define the research objectives, eligibility criteria, and analytical procedures. Due to the interdisciplinary nature of the study, spanning organizational behavior, information systems, and occupational health, the protocol was not registered in PROSPERO; however, all procedures were predefined to minimize selection and reporting bias.

2.2 Search strategy

A comprehensive literature search was conducted across four bibliographic databases, Web of Science, Scopus, PubMed/MEDLINE, and APA PsycINFO, supplemented by Google Scholar and backward and forward citation searching. The search strategy combined keywords related to artificial intelligence, workplace contexts, and burnout outcomes using Boolean operators. The core search string included terms such as “artificial intelligence,” “machine learning,” “algorithmic management,” and “automation,” combined with “burnout,” “emotional exhaustion,” “stress,” and “well-being,” as well as “employee,” “worker,” and “professional.”

Backward and forward citation tracking was also conducted to ensure comprehensive coverage, and database-specific adaptations of the search syntax were applied. Complete database-specific search strategies, limits, and search fields are provided in Supplementary Table S1.

2.3 Eligibility criteria

Studies were included if they met the following criteria: (1) an empirical quantitative, qualitative, mixed-methods, randomized, or non-randomized design; (2) publication as a peer-reviewed journal article; (3) examination of an identifiable AI-driven, algorithmic, robotic, automation, or machine-learning system in a workplace context; (4) inclusion of employees, workers, professionals, or occupational groups; and (5) reporting either (a) burnout or an established burnout dimension or (b) an occupational psychological outcome used specifically to examine a mechanism, moderator, boundary condition, or implementation factor theoretically relevant to burnout.

Eligible outcomes were classified a priori into two groups. Direct or core-dimension outcomes included burnout, emotional exhaustion, work exhaustion, interpersonal disengagement, depersonalization, and professional efficacy. Supporting outcomes included technostress, job stress, job insecurity, work–family interference, occupational well-being, career sustainability, psychological health, and related mechanisms or boundary conditions. Supporting outcomes were synthesized separately and were not interpreted as equivalent to a validated burnout endpoint.

Studies were excluded if they were reviews, conceptual articles, editorials, protocols, commercial reports, conference proceedings, non-peer-reviewed sources, studies of non-working populations, studies without an identifiable workplace AI or automation exposure, or studies without an eligible occupational psychological outcome. Only English-language reports were considered.

2.4 Study selection

All identified records were imported into EndNote (X9) for reference management. Duplicate records were removed automatically and verified manually. Titles and abstracts were screened independently by two reviewers, and disagreements were resolved through discussion or consultation with a third reviewer.

The search initially yielded 1,248 records. After removal of 312 duplicate records, 936 records underwent title and abstract screening, of which 742 were excluded. The full texts of 194 reports were assessed for eligibility, and 151 reports were excluded because they did not meet the empirical-design, population, exposure, outcome, publication-status, or reporting criteria. A total of 43 peer-reviewed primary empirical studies were included in the final synthesis, comprising 20 studies measuring burnout or an established burnout dimension and 23 supporting empirical studies addressing mechanisms, moderators, proxy outcomes, measurement, or implementation conditions. The complete study-selection process is presented in Figure 1, while the operational exclusion criteria and potentially eligible reports excluded after full-text assessment are provided in Supplementary Tables S2A,B, respectively.

Figure 1

2.5 Data extraction

A standardized data extraction protocol was applied to all included studies. Extracted information included study characteristics (author, year, country, and industry), sample size and population, study design, type and functional role of the AI system, outcome or construct measured, measurement instrument, and reported relationships between AI exposure and burnout, established burnout dimensions, or supporting occupational psychological outcomes. Where available, effect sizes, including standardized coefficients, odds ratios, correlations, confidence intervals, and statistical significance, were recorded.

In addition, studies were systematically coded to identify underlying mechanisms (e.g., technostress, job insecurity, identity threat), moderating variables (e.g., organizational support, self-efficacy), and contextual boundary conditions. The complete study-level extraction matrix is provided in Supplementary Tables S3A,B. Studies measuring burnout or an established burnout dimension are reported in Supplementary Table S3A, whereas studies addressing mechanisms, moderators, proxy outcomes, measurement, or implementation conditions are reported separately in Supplementary Table S3B.

2.6 Quality assessment

Methodological quality was assessed using the Mixed Methods Appraisal Tool (MMAT 2018; Hong et al., 2018), which permits design-specific appraisal of qualitative, randomized, non-randomized, descriptive, and mixed-methods studies. The two MMAT screening questions were applied to all studies, followed by the five criteria corresponding to the relevant design category. Two reviewers completed the appraisal independently, and disagreements were resolved through discussion. Criterion-level judgments of yes, no, or cannot tell are reported in Supplementary Table S4. Consistent with MMAT guidance, no aggregate numerical quality score was calculated. Recurrent limitations involved sample representativeness, nonresponse reporting, attrition, self-reported outcomes, and control of confounding, whereas randomized, longitudinal, and repeated-measures studies generally provided stronger evidence.

2.7 Data synthesis

Due to substantial heterogeneity in study designs, AI operationalizations, and burnout measures, a quantitative meta-analysis was not feasible. Instead, a theory-driven narrative synthesis was conducted.

The synthesis proceeded in three stages. First, empirical findings were organized according to the core dimensions of the proposed AIMS framework: demands, resources, and control. Second, a thematic coding approach was used to identify recurring mechanisms and moderators across studies. Third, cross-study comparisons were conducted to identify consistent patterns, contradictions, and boundary conditions, which informed the development of a dynamic, multi-pathway theoretical framework. Direct or core-dimension burnout evidence and supporting evidence were synthesized separately to prevent proxy psychological outcomes from being interpreted as equivalent to burnout.

3 Results

3.1 Characteristics of included studies

A total of 43 peer-reviewed primary empirical studies met the inclusion criteria. Twenty studies measured burnout or an established burnout dimension, whereas 23 examined mechanisms, moderators, proxy outcomes, measurement, or implementation conditions relevant to AI-related employee well-being. The evidence base comprised three randomized controlled studies, 19 quantitative non-randomized studies, 16 quantitative descriptive studies, three qualitative studies, and two mixed-methods studies. Sample sizes ranged from 15 participants in qualitative interviews to 27,252 workers in a large European survey.

AI exposure was operationalized in three broad functional roles: assistive systems, including ambient documentation and AI-tailored interventions; monitoring or control systems, including algorithmic management and automated task allocation; and perceptual exposures, including AI awareness, replacement concerns, trust, or perceived threat. Table 1 summarizes all 20 studies measuring burnout or an established burnout dimension. Supplementary Table S3A provides the corresponding study-level extraction information, while Supplementary Table S3B reports the 23 supporting studies addressing mechanisms, proxy outcomes, measurement, or implementation conditions.

Table 1

Characteristics of studies measuring burnout or an established burnout dimension.

Table presents the 20 included studies that measured burnout or an established burnout dimension. The complete evidence inventory for all 43 included primary empirical studies is provided in Supplementary Tables S3A,B. Supporting outcomes such as technostress, job insecurity, job satisfaction, career sustainability, and occupational well-being are reported separately and are not interpreted as equivalent to a validated burnout endpoint. AI, artificial intelligence; CBI, Copenhagen Burnout Inventory; CI, confidence interval; HRM, human resource management; MBI, Maslach Burnout Inventory; OR, odds ratio; RCT, randomized controlled trial; SEM, structural equation modeling.

3.2 Synthesis of quantitative findings

The quantitative findings, summarized in Table 2, indicated a role-dependent rather than uniform pattern. Studies measuring burnout or an established burnout dimension were analyzed separately from studies reporting supporting occupational psychological outcomes, including technostress, occupational stress, job insecurity, job satisfaction, and general well-being.

Table 2

Selected quantitative findings for direct burnout, core burnout dimensions, and supporting proxy outcomes.

Direct outcomes comprise burnout or an established burnout dimension, including emotional exhaustion, work exhaustion, and interpersonal disengagement. Supporting proxy outcomes are included to explain mechanisms, boundary conditions, or adjacent psychological consequences and are not interpreted as equivalent to burnout. CI, confidence interval; HRM, human resource management; NR, not reported; NS, not statistically significant; OR, odds ratio; RCT, randomized controlled trial; SEM, structural equation modeling.

Assistive AI systems were generally associated with lower burnout or exhaustion when they reduced documentation, workload, or cognitive demands. Randomized and pre–post healthcare studies reported reductions in burnout or established burnout dimensions following the use of ambient AI scribes or AI-tailored interventions (Afshar et al., 2025; Baek and Cha, 2025; Chowdhury et al., 2026; Cho et al., 2024; Misurac et al., 2025; Olson et al., 2025; You et al., 2025). However, these findings were not uniform. Stults et al. (2025), for example, reported a reduction in burnout from 42.1 to 35.1%, but the difference was not statistically significant. Moreover, several implementation studies lacked concurrent control groups, and some had low follow-up response or unpaired pre- and post-implementation samples.

Monitoring and algorithmic-control exposures were generally associated with greater burnout or related psychosocial strain. Algorithmic management was positively associated with burnout among delivery riders (Dong et al., 2025), and burnout mediated the adverse association between perceived algorithmic control and service performance among gig workers (Gao et al., 2025). Large-scale European evidence further linked AI-enabled monitoring and task allocation with time pressure, reduced autonomy, and stress-related outcomes (Payá Castiblanque and Pizzi, 2024). Perceptual AI exposure produced more heterogeneous results. AI awareness was positively associated with burnout or emotional exhaustion in Kong et al. (2021), Zheng and Zhang (2025), and Wang and Zhou (2025); Kim and Lee (2024) identified an indirect pathway through job stress rather than a significant direct association; and Thorpe et al. (2026) found no statistically significant direct relationship between generative-AI use and burnout. Three randomized studies provided the strongest evidence concerning assistive interventions, but all were conducted in healthcare contexts. Evidence for nonlinear effects was limited to supporting proxy outcomes, including the inverted U-shaped association between AI exposure and job satisfaction reported by Wu et al. (2026), and did not directly establish a nonlinear AI–burnout relationship.

3.3 Mechanisms linking AI to burnout

The mechanisms and mediating pathways identified across the included studies are summarized in Table 3. The evidence indicated that AI exposure was seldom associated with burnout through a single pathway. Instead, the included studies reported associations involving AI-related demands, resources, control conditions, and psychological or organizational mechanisms. In Chuang et al. (2025), AI technostress was positively associated with exhaustion, whereas AI efficacy was negatively associated with exhaustion. Kim and Lee (2024) reported a significant indirect association between AI adoption and burnout through job stress. Zheng and Zhang (2025) reported a serial indirect association through job insecurity and work interference with family, while Wang and Zhou (2025) identified indirect associations involving perceived organizational support and organizational commitment. Work–life balance was examined as a potential resource pathway in AlOqaily et al. (2025), and Gao et al. (2025) reported an indirect association between algorithmic control and service performance through burnout. These mediation findings should be interpreted in light of the observational designs used in most of the relevant studies.

Table 3

Mechanisms and mediating pathways relevant to AI-related burnout and occupational strain.

Mechanisms derived from studies that did not directly measure burnout are identified as supporting pathways. Their outcomes should not be interpreted as equivalent to burnout. AI, artificial intelligence; FoMO, fear of missing out; HRM, human resource management.

The supporting and qualitative studies identified additional AI-specific mechanisms, including dependency, regulatory ambiguity, skill degradation, autonomy loss, surveillance precarity, and AI-related fear of missing out (Högemann et al., 2025; Méndez-Suárez et al., 2026; Zheng et al., 2025). These constructs help explain how AI may generate occupational strain, but they were not interpreted as direct burnout outcomes unless burnout or an established burnout dimension was explicitly measured.

3.4 Moderators and boundary conditions

The individual, organizational, and system-level moderators and boundary conditions identified across the included studies are summarized in Table 4. These factors help explain why similar forms of AI exposure may be associated with different psychological outcomes across employees and organizational settings.

Table 4

Moderators and boundary conditions relevant to AI-related burnout and occupational well-being.

Several factors in table were examined in relation to supporting outcomes such as job insecurity, career sustainability, or workplace well-being rather than burnout itself. They are therefore interpreted as boundary conditions relevant to the AIMS framework, not as confirmed moderators of every AI–burnout relationship. AI, artificial intelligence.

At the individual level, AI self-efficacy weakened the adverse indirect pathway from AI adoption to burnout through job stress (Kim and Lee, 2024). Resilience and career resilience buffered adverse pathways involving AI awareness, job stress, job insecurity, and employee well-being (Jin et al., 2024; Chung et al., 2025). At the organizational level, perceived organizational support and organizational commitment formed protective pathways between AI awareness and burnout (Wang and Zhou, 2025). Organizational culture also shaped employee responses: authoritarian culture strengthened the association between job-automation awareness and job insecurity (Lingmont and Alexiou, 2020). At the system and implementation levels, algorithmic transparency weakened the adverse indirect pathway associated with algorithmic control (Gao et al., 2025), while training, personalization, and implementation support were associated with more favorable employee outcomes in human-centered AI systems (Meduri et al., 2024). Appraisal and coping further determined whether AI unpredictability was experienced as distress or eustress (Issa et al., 2024).

3.5 Qualitative findings

Three qualitative or qualitative-dominant studies provided detailed evidence concerning employees’ experiences of AI-enabled work. Högemann et al. (2025) identified dependency, regulatory ambiguity, monitoring shifts, and perceived skill degradation among young professionals using generative AI. Kumar et al. (2024) described an automation–augmentation paradox in which AI and machine learning could simultaneously generate technostress and provide opportunities for socio-technical coping. Zheng et al. (2025) identified autonomy erosion, surveillance precarity, algorithmic bias, and personalized discontentment in AI-enabled human-resource systems.

These findings complement the quantitative evidence by explaining how AI-related demands, resources, and control structures are experienced by employees. However, because these studies primarily examined technostress, well-being, or broader occupational experiences, their findings should be interpreted as explanatory evidence concerning mechanisms and implementation conditions rather than as quantitative estimates of AI effects on burnout. Complete characteristics of these and the other supporting studies are reported in Supplementary Table S3B.

3.6 Heterogeneity and contradictory evidence

The apparent heterogeneity in the evidence can be partly explained by differences in AI role, outcome definition, study design, and implementation context. Assistive systems were generally associated with lower burnout or exhaustion when they reduced documentation or workload demands, but the magnitude and statistical significance of change varied across implementation studies. Monitoring and algorithmic-control exposures were associated with greater burnout or related psychosocial strain, although the relevant evidence was predominantly cross-sectional. Perceptual AI exposure produced positive, indirect, and null findings depending on appraisal, self-efficacy, organizational support, resilience, and job insecurity. Thus, the evidence supports a role-dependent pattern but not a universal beneficial or harmful effect of AI.

3.7 Risk of bias and methodological quality

The evidence base was dominated by observational designs, particularly cross-sectional surveys, which limited causal inference and increased vulnerability to common-method bias. Three randomized studies (Afshar et al., 2025; Baek and Cha, 2025; Chowdhury et al., 2026) and several longitudinal or pre–post studies provided stronger temporal evidence, although attrition, self-report measurement, low follow-up response, and the absence of concurrent controls remained important limitations in some implementation studies. Criterion-level MMAT judgments for all 43 studies are reported in Supplementary Table S4.

4 Discussion

4.1 Summary of key findings

This systematic review indicates that the association between AI-driven work systems and employee burnout depends on the functional role, implementation, and appraisal of the technology. Assistive systems were generally associated with lower burnout or exhaustion when they reduced documentation and workload demands, particularly in healthcare implementation studies. Monitoring and algorithmic-control systems were associated with greater burnout or psychosocial strain through intensified demands and reduced autonomy. Perceptual exposure, including AI awareness and replacement concerns, produced direct, indirect, and null associations through pathways involving job stress, job insecurity, work–family interference, perceived organizational support, organizational commitment, and resilience. Because most studies were observational, these patterns should be interpreted as associations rather than universal causal effects.

The review identified technostress, exhaustion, job insecurity, work–family interference, work–life balance, organizational support, and autonomy as recurrent pathways. AI-specific dependency, ambiguity, skill degradation, and FoMO emerged as additional potential stress mechanisms. Evidence for a nonlinear AI–burnout relationship remained limited. Wu et al. (2026) reported an inverted U-shaped association for job satisfaction, which was classified as a supporting proxy outcome rather than a direct burnout endpoint. The proposed shock–adjustment–mastery trajectory should therefore be interpreted as a theoretical proposition requiring direct longitudinal testing.

4.2 Theoretical contributions

4.2.1 The AIMS model: AI as a meta-system beyond JD–R

The findings support a reconceptualization of AI as a meta-system that simultaneously reshapes job demands, resources, and control, extending the Job Demands–Resources (JD–R) model. Figure 2 presents the conceptual AI as Meta-System (AIMS) framework, in which AI is theorized to reshape job demands, job resources, and control structures simultaneously. These pathways may be associated with burnout through mechanisms including technostress, exhaustion, job insecurity, and emotional labor, while individual and organizational conditions may modify their strength or direction. The pathways shown in Figure 2 represent theory-derived relationships rather than causal effects established across all included studies.

Figure 2

Empirical evidence illustrates this dual role. Empirical findings are consistent with this dual pattern. Chuang et al. (2025) reported that AI-related technostress was positively associated with emotional exhaustion (γ = 0.33), whereas AI efficacy was negatively associated with exhaustion (γ = −0.16). Similarly, Kim and Lee (2024) identified a significant indirect association between AI adoption and burnout through job stress, while the direct association was not statistically significant. These findings are consistent with multiple psychological pathways linking AI exposure to burnout-related outcomes but do not, by themselves, establish universal causal effects.

The AIMS framework extends JD–R by introducing three key elements: (1) simultaneous influence on demands, resources, and control; (2) non-linear pathways; and (3) dynamic adaptation over time. This shifts the conceptualization of work from task execution to continuous meta-cognitive regulation, where employees must interpret and respond to algorithmic systems.

4.2.2 A role-based typology: assistive, monitoring, and perceptual AI

A central contribution of this review is the introduction of a role-based typology that reconciles conflicting findings in the literature. Distinguishing between assistive, monitoring, and perceptual AI clarifies why prior studies have reported divergent effects.

Assistive AI was generally associated with lower burnout or exhaustion when it reduced documentation, workload, or cognitive demands, particularly in healthcare intervention and implementation studies (Afshar et al., 2025; Baek and Cha, 2025; Chowdhury et al., 2026; Olson et al., 2025; You et al., 2025). Monitoring and algorithmic-control exposures were associated with greater burnout or related psychosocial strain through time pressure, reduced autonomy, surveillance, and intensified control (Dong et al., 2025; Gao et al., 2025; Payá Castiblanque and Pizzi, 2024). Perceptual exposure produced more heterogeneous findings and operated through mechanisms such as job stress, job insecurity, work–family interference, organizational support, and self-efficacy (Kim and Lee, 2024; Zheng and Zhang, 2025; Wang and Zhou, 2025). This typology advances prior research by explicitly linking AI functionality to psychological outcomes, rather than treating AI as a homogeneous construct.

4.2.3 Dynamic and non-linear pathways

The included evidence does not directly establish an inverted U-shaped temporal trajectory between AI exposure and employee burnout. Wu et al. (2026) identified a nonlinear association between AI exposure and job satisfaction, but job satisfaction was treated as a supporting proxy outcome rather than a burnout measure. The proposed shock–adjustment–mastery curve is therefore presented as a conceptual model rather than an empirical estimate. It proposes that initial uncertainty and learning demands may increase strain, while subsequent adaptation, training, system familiarity, and organizational support may reduce strain over time.

AI-specific longitudinal evidence remains limited. Existing multi-wave, randomized, and pre–post studies generally cover relatively short implementation periods and do not test the complete proposed trajectory. Direct examination of within-person burnout before implementation, during adjustment, and after sustained use is therefore required. Figure 3 presents this theory-derived three-phase trajectory as a conceptual illustration requiring direct longitudinal testing.

Figure 3

4.2.4 Mechanisms and moderators

The review identified both established and emerging mechanisms relevant to AI-related burnout. Established pathways included technostress and exhaustion (Chuang et al., 2025), job stress (Kim and Lee, 2024), job insecurity and work interference with family (Zheng and Zhang, 2025), work–life balance (AlOqaily et al., 2025), and perceived organizational support and organizational commitment (Wang and Zhou, 2025). Emerging mechanisms included dependency, skill degradation, regulatory ambiguity, surveillance precarity, and AI-related FoMO (Högemann et al., 2025; Zheng et al., 2025; Méndez-Suárez et al., 2026).

Individual-level boundary conditions included AI self-efficacy, resilience, career resilience, and protean career orientation (Kim and Lee, 2024; Jin et al., 2024; Chung et al., 2025; Kong et al., 2023). Organizational and system-level conditions included perceived organizational support, organizational commitment, organizational culture, implementation quality, and algorithmic transparency (Meduri et al., 2024; Gao et al., 2025; Wang and Zhou, 2025; Lingmont and Alexiou, 2020). Complementary evidence also highlights trust, threat, and protean orientation in employee–AI collaboration and career sustainability (Duong et al., 2026), the role of generative AI in employees’ work-goal progress (Zhang et al., 2026), and the relationship between leadership AI awareness, hindrance/challenge stressors, and employee voice (Zhou and Lyu, 2025).

4.2.5 Research propositions derived from the AIMS framework

The following propositions are theory-derived deductions from the AIMS framework and were not tested in the present systematic review.

P1: AI-driven job demands, including monitoring intensity, cognitive overload, and replacement concerns, are expected to be positively associated with employee burnout through technostress and job insecurity.

P2: AI-driven job resources, including automation of routine work and decision support, are expected to be negatively associated with burnout when they reduce workload and cognitive demands.

P3: AI-driven control mechanisms, including algorithmic monitoring and automated task allocation, are expected to be positively associated with burnout through reduced perceived autonomy.

P4: Professional identity threat is expected to mediate the association between perceived AI replacement risk and emotional exhaustion.

P5: AI self-efficacy is expected to weaken adverse associations between AI-driven demands and burnout.

P6: Perceived organizational support is expected to weaken adverse associations between AI exposure and burnout.

P7: AI-related burnout may follow nonlinear adaptation trajectories over time, with their direction and shape contingent on implementation quality, learning, job redesign, and organizational support.

P8: Algorithmic transparency is expected to weaken adverse associations between AI monitoring and burnout when it is accompanied by explanation, employee voice, and meaningful recourse.

4.3 Practical implications

The findings should be interpreted as design and implementation guidance rather than evidence that one form of AI will produce the same outcome in every organization. Assistive systems may reduce burnout or exhaustion when they demonstrably reduce documentation, workload, and cognitive demands. Organizations should therefore evaluate changes in workload and burnout before and after implementation rather than assuming that automation automatically improves well-being.

Monitoring and algorithmic-control systems require particular caution because observational studies associated these exposures with greater psychosocial strain, time pressure, reduced autonomy, or burnout. Where monitoring is necessary, organizations should provide transparent decision rules, employee voice, appeal or recourse mechanisms, clear accountability, and limits on data collection. These recommendations are also consistent with broader organizational research emphasizing the implications of transparency for privacy, organizational learning, operational control, and motivation (Bernstein, 2012; Brandes and Darai, 2017).

Training, personalization, technical support, and realistic performance expectations may improve implementation outcomes. Organizations should also monitor job insecurity, work–family interference, AI self-efficacy, resilience, and organizational support because these factors may shape employee responses. Burnout imposes substantial organizational costs, but future research should directly quantify whether responsible AI implementation reduces these costs over time.

4.4 Limitations and future research

Several limitations should be considered. First, the evidence base was dominated by observational and cross-sectional studies, limiting causal inference and increasing vulnerability to common-method and self-report bias. Three randomized studies were identified, but they were all conducted in healthcare and primarily examined assistive AI systems. Their findings may not generalize to monitoring, algorithmic-control, or non-healthcare contexts.

Second, direct comparisons of AI-based monitoring and human supervision using validated burnout outcomes were not identified. Gu et al. (2025) compared algorithmic and human supervision but was excluded from the systematic-review sample because it reported performance rather than an eligible psychological outcome. Future studies should compare these supervisory arrangements using longitudinal or experimental designs and validated occupational psychological endpoints.

Third, AI-specific longitudinal evidence remains insufficient to establish a shock–adjustment–mastery trajectory. Future studies should use repeated within-person measurements before implementation, during initial adjustment, and after sustained use. Fourth, cross-national and cross-occupational comparisons remain limited despite likely differences in regulation, labor-market conditions, organizational culture, and employee voice.

Fifth, measurement remains fragmented. Cross-national measurement research further underscores the importance of measurement invariance in burnout assessment (De Beer et al., 2020). Burnout instruments were not always AI-specific, and supporting outcomes such as technostress, job insecurity, job satisfaction, and well-being were operationalized inconsistently. Contextual scale-development studies outside the included AI-specific sample indicate growing interest in technology-related burnout measurement, but no comprehensive instrument currently integrates AI-specific demands, resources, control, identity threat, and adaptation.

Finally, experimental evidence concerning transparency, recourse, organizational support, resilience, and human-versus-algorithmic supervision remains limited. Future research should test these moderators directly rather than inferring their effects from cross-sectional associations.

5 Conclusion

AI was not uniformly associated with either beneficial or adverse burnout outcomes. Across the 20 studies measuring burnout or an established burnout dimension, assistive systems, examined predominantly in healthcare settings, were generally associated with lower burnout or exhaustion, whereas monitoring and algorithmic-control exposures were generally associated with greater burnout in observational studies. Perceptual AI exposures produced positive, indirect, and null associations. The 23 supporting studies clarified potential mechanisms, moderators, and implementation conditions but were not treated as direct burnout evidence. The AIMS framework organizes these findings into theory-derived pathways involving demands, resources, and control. Longitudinal, cross-occupational, and experimental studies are required before causal or temporal conclusions can be established.

Statements

Data availability statement

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/Supplementary material.

Author contributions

ZF: Investigation, Formal analysis, Writing – review & editing, Methodology. LH: Formal analysis, Investigation, Writing – review & editing. TM: Investigation, Formal analysis, Writing – original draft.

Funding

The author(s) declared that financial support was not received for this work and/or its publication.

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpsyg.2026.1922281/full#supplementary-material

References

• 1

  AfsharM.BaumannM. R.ResnikF.HintzkeJ.Gravel SullivanA.WillsG.et al. (2025). A pragmatic randomized controlled trial of ambient artificial intelligence to improve health practitioner well-being. NEJM AI2:AIoa2500945. doi: 10.1056/AIoa2500945,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 2

  AlbrechtM.ShanksD.ShahT.HudsonT.ThompsonJ.FilardiT.et al. (2025). Enhancing clinical documentation with ambient artificial intelligence: a quality improvement survey assessing clinician perspectives on work burden, burnout, and job satisfaction. JAMIA Open8:ooaf013. doi: 10.1093/jamiaopen/ooaf013,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 3

  AliT.HussainI.HassanS.AnwerS. (2024). Examine how the rise of AI and automation affects job security, stress levels, and mental health in the workplace. Bull. Bus. Econ.13, 1180–1186. doi: 10.61506/01.00506

  • CrossRef
  • Google Scholar

• 4

  AlOqailyA. N.QawasmehE. F.TawalbehJ. (2025). The effect of implementing AI on job burnout through the mediating role of work–life balance in the context of HRM. Economics13, 465–484. doi: 10.2478/eoik-2025-0049

  • CrossRef
  • Google Scholar

• 5

  ArbohF.ZhuX.AtingabiliS.YeboahE.DrokowE. K., (2025). From fear to empowerment: the impact of employees’ AI awareness on workplace well-being—a new insight from the JD–R model. J. Health Organ. Manag.40, 1148–1172. doi:10.1108/JHOM-06-2024-0229

  • CrossRef
  • Google Scholar

• 6

  BaekG.ChaC. (2025). AI-assisted tailored intervention for nurse burnout: a three-group randomized controlled trial. Worldviews Evid.-Based Nurs.22:e70003. doi: 10.1111/wvn.70003,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 7

  BakkerA. B.DemeroutiE.Sanz-VergelA. I. (2023). Job demands–resources theory: ten years later. Annu. Rev. Organ. Psych. Organ. Behav.10, 25–53. doi: 10.1146/annurev-orgpsych-120920-053933

  • CrossRef
  • Google Scholar

• 8

  BankinsS.FormosaP.GriepY.RichardsD. (2024). A multilevel review of artificial intelligence in organizations: implications for organizational behavior research and practice. J. Organ. Behav.45, 159–182. doi: 10.1002/job.2735

  • CrossRef
  • Google Scholar

• 9

  BernsteinE. S. (2012). The transparency paradox: a role for privacy in organizational learning and operational control. Adm. Sci. Q.57, 181–216. doi: 10.1177/0001839212453028

  • CrossRef
  • Google Scholar

• 10

  BraganzaA.ChenW.CanhotoA.SapS. (2021). Productive employment and decent work: the impact of AI adoption on psychological contracts, job engagement and employee trust. J. Bus. Res.131, 485–494. doi: 10.1016/j.jbusres.2020.08.018,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 11

  BrandesL.DaraiD. (2017). The value and motivating mechanism of transparency in organizations. Eur. Econ. Rev.98, 189–198. doi: 10.1016/j.euroecorev.2017.06.014

  • CrossRef
  • Google Scholar

• 12

  BroughamD.HaarJ. (2018). Smart technology, artificial intelligence, robotics, and algorithms (STARA): employees’ perceptions of our future workplace. J. Manag. Organ.24, 239–257. doi: 10.1017/jmo.2016.55

  • CrossRef
  • Google Scholar

• 13

  ChengW. J.PienL. C.ChengY. (2021). Occupation-level automation probability is associated with psychosocial work conditions and workers’ health: a multilevel study. Am. J. Ind. Med.64, 108–117. doi: 10.1002/ajim.23210,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 14

  ChoA.ChaC.BaekG. (2024). Development of an artificial intelligence–based tailored mobile intervention for nurse burnout: single-arm trial. J. Med. Internet Res.26:e54029. doi: 10.2196/54029,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 15

  ChowdhuryA.CaseyM.WilsonJ.PollakK. I.GoldsteinB. A.BedoyaA.et al. (2026). Comparing ambient scribes: a randomized crossover clinical trial addressing ambient scribe technologies’ impact on physician burnout. J. Am. Med. Inform. Assoc.33, 990–999. doi: 10.1093/jamia/ocag018,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 16

  ChuangY. T.ChiangH. L.LinA. P. (2025). Insights from the job demands–resources model: AI’S dual impact on employees’ work and life well-being. Int. J. Inf. Manag.83:102887. doi: 10.1016/j.ijinfomgt.2025.102887

  • CrossRef
  • Google Scholar

• 17

  ChungY. W.ImS.KimJ. E.YunJ. K. (2025). Artificial intelligence awareness, career resilience, job insecurity and behavioural outcomes. Aust. J. Psychol.77:2559910. doi: 10.1080/00049530.2025.2559910,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 18

  DaveB.MartinP.DavidS. S.KumarS.ChakrabortyT. (2026). Enhancing healthcare worker mental health via artificial intelligence-driven work process improvements: a scoping review. Int. J. Med. Inform.205:106122. doi: 10.1016/j.ijmedinf.2025.106122,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 19

  De BeerL. T.SchaufeliW. B.De WitteH.HakanenJ. J.ShimazuA.GlaserJ.et al. (2020). Measurement invariance of the burnout assessment tool across seven cross-national representative samples. Int. J. Environ. Res. Public Health17:5604. doi: 10.3390/ijerph17155604

  • CrossRef
  • Google Scholar

• 20

  DongJ.ZhangG.WuL. (2025). Life against algorithmic management: burnout and its influencing factors among food-delivery riders. Front. Public Health13:1531541. doi: 10.3389/fpubh.2025.1531541

  • CrossRef
  • Google Scholar

• 21

  DuongN. H.HungH. N.AnN. H.BaoL. G.NguyenT. T.QuangD. X. (2026). Psychological drivers of employee–AI collaboration and career sustainability: the role of trust, threat, and protean orientation. Acta Psychol.262:106074. doi: 10.1016/j.actpsy.2025.106074,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 22

  GaoX.WuW.GuoY.LuL. (2025). Promotion or suppression? Effect of perceived algorithmic control on service performance of gig workers. Soc. Behav. Personal. Int. J.53, 1–13. doi: 10.2224/sbp.14812

  • CrossRef
  • Google Scholar

• 23

  GullA.AshfaqJ.AslamM. (2023). AI in the workplace: uncovering its impact on employee well-being and the role of cognitive job insecurity. Int. J. Bus. Econ. Affairs8, 79–91. doi: 10.24088/IJBEA-2023-84007

  • CrossRef
  • Google Scholar

• 24

  GuW.LiM.ZhangS. (2025). Algorithmic Supervisor and Employee Performance. Prod. Oper. Manag.34, 3101–3118. doi: 10.1177/10591478251331095

  • CrossRef
  • Google Scholar

• 25

  HögemannM.HeinL.BritscheJ. O.ThomasO. (2025). Technostress and generative AI in the workplace: a qualitative analysis of young professionals. Front. Art. Intell.8:1728881. doi: 10.3389/frai.2025.1728881,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 26

  HongQ. N.FàbreguesS.BartlettG.BoardmanF.CargoM.DagenaisP.et al. (2018). The mixed methods appraisal tool (MMAT) version 2018 for information professionals and researchers. Educ. Inf.34, 285–291. doi: 10.3233/EFI-180221

  • CrossRef
  • Google Scholar

• 27

  IssaH.JaberJ.LakkisH. (2024). Navigating AI unpredictability: exploring technostress in AI-powered healthcare systems. Technol. Forecast. Soc. Change202:123311. doi: 10.1016/j.techfore.2024.123311

  • CrossRef
  • Google Scholar

• 28

  JinG.JiangJ.LiaoH. (2024). The work affective well-being under the impact of AI. Sci. Rep.14:25483. doi: 10.1038/s41598-024-75113-w,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 29

  KimB. J.LeeJ. (2024). The mental health implications of artificial intelligence adoption: the crucial role of self-efficacy. Humanit. Soc. Sci. Commun.11:1561. doi: 10.1057/s41599-024-04018-w

  • CrossRef
  • Google Scholar

• 30

  KochM.LodefalkM. (2025). Artificial intelligence and worker stress: evidence from Germany. Digit. Soc.4:5. doi: 10.1007/s44206-025-00160-3

  • CrossRef
  • Google Scholar

• 31

  KongH.YinZ.BaruchY.YuanY. (2023). The impact of trust in AI on career sustainability: the role of employee–AI collaboration and protean career orientation. J. Vocat. Behav.146:103928. doi: 10.1016/j.jvb.2023.103928

  • CrossRef
  • Google Scholar

• 32

  KongH.YuanY.BaruchY.BuN.JiangX.WangK. (2021). Influences of artificial intelligence awareness on career competency and job burnout. Int. J. Contemp. Hosp. Manag.33, 717–734. doi: 10.1108/IJCHM-07-2020-0789

  • CrossRef
  • Google Scholar

• 33

  KumarA.KrishnamoorthyB.BhattacharyyaS. S. (2024). Machine learning and artificial intelligence-induced technostress in organizations: a study on the automation–augmentation paradox with socio-technical systems as coping mechanisms. Int. J. Organ. Anal.32, 681–701. doi: 10.1108/IJOA-01-2023-3581

  • CrossRef
  • Google Scholar

• 34

  LingmontD. N. J.AlexiouA. (2020). The contingent effect of job automating technology awareness on perceived job insecurity: exploring the moderating role of organizational culture. Technol. Forecast. Soc. Change161:120302. doi: 10.1016/j.techfore.2020.120302

  • CrossRef
  • Google Scholar

• 35

  LiuH.DingN.LiX.ChenY.SunH.HuangY.et al. (2024). Artificial intelligence and radiologist burnout. JAMA Netw. Open7:e2448714. doi: 10.1001/jamanetworkopen.2024.48714,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 36

  MartinezM. F.O’SheaK. J.KernM. C.ChinK. L.DinhJ. V.BartschS. M.et al. (2025). The health and economic burden of employee burnout to US employers. Am. J. Prev. Med.68, 645–655. doi: 10.1016/j.amepre.2025.01.011

  • CrossRef
  • Google Scholar

• 37

  MeduriK.NadellaG. S.GonayguntaH.KumarD.AddulaS. R.SatishS.et al. (2024). Human-centered AI for personalized workload management: a multimodal approach to preventing employee burnout. J. Infrastruct. Policy Dev.8:6918. doi: 10.24294/jipd.v8i9.6918

  • CrossRef
  • Google Scholar

• 38

  Méndez-SuárezM.ĆukušićM.Ninčević-PašalićI. (2026). AI FoMO (fear of missing out) in the workplace. Technol. Soc.84:103052. doi: 10.1016/j.techsoc.2025.103052

  • CrossRef
  • Google Scholar

• 39

  MisuracJ.KnakeL. A.BlumJ. M. (2025). The effect of ambient artificial intelligence notes on provider burnout. Appl. Clin. Inform.16, 252–258. doi: 10.1055/a-2461-4576,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 40

  NazarenoL.SchiffD. S. (2021). The impact of automation and artificial intelligence on worker well-being. Technol. Soc.67:101679. doi: 10.1016/j.techsoc.2021.101679

  • CrossRef
  • Google Scholar

• 41

  OlsonK. D.MeekerD.TroupM.BarkerT. D.NguyenV. H.MandersJ. B.et al. (2025). Use of ambient AI scribes to reduce administrative burden and professional burnout. JAMA Netw. Open8:e2534976. doi: 10.1001/jamanetworkopen.2025.34976,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 42

  PageM. J.McKenzieJ. E.BossuytP. M.BoutronI.HoffmannT. C.MulrowC. D.et al. (2021). The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ372:n71. doi: 10.1136/bmj.n71,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 43

  Payá CastiblanqueR.PizziA. (2024). Relationship between certain uses of artificial intelligence and psychosocial risk factors in European work environments. Archivos Prevención de Riesgos Laborales27, 233–249. doi: 10.12961/aprl.2024.27.03.02,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 44

  ShanafeltT. D.GohJ.SinskyC. (2017). The business case for investing in physician well-being. JAMA Intern. Med.177, 1826–1832. doi: 10.1001/jamainternmed.2017.4340,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 45

  StamateA. N.SauvéG.DenisP. L. (2021). The rise of the machines and how they impact workers’ psychological health: an empirical study. Hum. Behav. Emerg. Technol.3, 942–955. doi: 10.1002/hbe2.315

  • CrossRef
  • Google Scholar

• 46

  StultsC. D.DengS.MartinezM. C.WilcoxJ.SzwerinskiN.ChenK. H.et al. (2025). Evaluation of an ambient artificial intelligence documentation platform for clinicians. JAMA Netw. Open8:e258614. doi: 10.1001/jamanetworkopen.2025.8614,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 47

  ThorpeD.BeanC.KriegJ. (2026). Associations between generative artificial intelligence usage, job demands, job control and burnout. AI Soc.41, 4077–4086. doi: 10.1007/s00146-025-02764-2

  • CrossRef
  • Google Scholar

• 48

  WangD.ZhouX. (2025). The impact of AI awareness on employees’ job burnout: a chain mediation of perceived organizational support and organizational commitment. SAGE Open15:21582440251400532. doi: 10.1177/21582440251400532

  • CrossRef
  • Google Scholar

• 49

  WuW.YangY.YangL.SuL.ChenZ. (2026). The inverted U-shaped relationship between AI and job satisfaction: empirical evidence from China’s CLDS microdata. Emerg. Mark. Finance Trade. 62, 2554–2569. doi: 10.1080/1540496X.2025.2548427

  • CrossRef
  • Google Scholar

• 50

  YouJ. G.DboukR. H.LandmanA.TingD. Y.DuttaS.WangJ. C.et al. (2025). Ambient documentation technology in clinician experience of documentation burden and burnout. JAMA Netw. Open8:e2528056. doi: 10.1001/jamanetworkopen.2025.28056,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 51

  ZerilliJ.BhattU.WellerA. (2022). How transparency modulates trust in artificial intelligence. Patterns3:100455. doi: 10.1016/j.patter.2022.100455,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 52

  ZhangR.LiuY.QasimM.WangX. (2026). More than a tool: how generative AI shapes employees’ work-goal progress. Pers. Rev.55, 418–454. doi: 10.1108/PR-05-2025-0478

  • CrossRef
  • Google Scholar

• 53

  ZhengJ.ZhangT. (2025). Association between AI awareness and emotional exhaustion: the serial mediation of job insecurity and work interference with family. Behav. Sci.15:401. doi: 10.3390/bs15040401,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 54

  ZhengJ.ZhangJ. Z.KamalM. M.LiangX.AlzeibyE. A. (2025). Unpacking human–AI interaction: exploring unintended consequences on employee well-being in entrepreneurial firms through an in-depth analysis. J. Bus. Res.196:115406. doi: 10.1016/j.jbusres.2025.115406

  • CrossRef
  • Google Scholar

• 55

  ZhouY.LyuB. (2025). How does leadership AI awareness shape employee voice behavior? A study based on the framework of hindrance and challenge stressors. Work82, 289–305. doi: 10.1177/10519815251341816,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

Keywords

algorithmic management, artificial intelligence, employee burnout, job demands–resources model, systematic review, technostress

Citation

Fang Z, Han L and Ma T (2026) Artificial intelligence-driven work systems and employee burnout: a systematic review of mechanisms, moderators, and a dynamic meta-system framework. Front. Psychol. 17:1922281. doi: 10.3389/fpsyg.2026.1922281

Received

06 July 2026

Revised

31 July 2026

Accepted

06 August 2026

Published

02 October 2026

Volume

17 - 2026

Edited by

Suryo Wibowo, Krida Wacana Christian University, Indonesia

Updates

Copyright

© 2026 Fang, Han and Ma.

This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

*Correspondence: Zicheng Fang, 4032530641@smail.lnu.edu.cn; Te Ma, mate987654321@163.com

† These authors share first authorship

Disclaimer

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.