







































Z. Qian et al. /Future Technology                                                                                       February 2026| Volume 05 | Issue 01 | Pages 
222-233 

222 

 

 

 

Article 

Research the association between AI-driven 

organizational support systems and university 

faculty work engagement: the moderating role of 

digital literacy 
Zhixin Qian1, Andi Tamsang Andi Kele1*, Ang Hong Loong1, Pang Yeng Yuan2  

1Faculty of Business, Economics and Accountancy, Universiti Malaysia Sabah, Malaysia 
2Faculty of Accountancy, Finance and Business, Tunku Abdul Rahman University of Management and Technology, 88450, 

Kota Kinabalu, Sabah, Malaysia 

A R T I C L E   I N F O 
 

Article history: 
Received 24 August 2025  
Received in revised form 
18 October 2025 
Accepted 23 November 2025 
 
Keywords:  
AI-driven organizational support,  
Work engagement, Digital literacy,  
Higher education, Faculty well-being,  
Moderation effect 
 
*Corresponding author 
Email address: 
andi@ums.edu.my 
 
DOI: 10.55670/fpll.futech.5.1.19 

A B S T R A C T 
 

According to the Job Demands-Resources (JD-R) model and the Technology 
Acceptance Model (TAM), this cross-sectional survey examined whether 
organizational support systems enabled by artificial intelligence (AI) were 
positively correlated with work engagement among university lecturers and 
examined the moderating role of digital literacy on 387 teachers at certain 
Chinese universities. With 9-item multidimensional UWES-9 vigor, dedication, 
and absorption scale of AI support in teaching, research, and administration 
domains, hierarchical regression with simple slopes, it was found AI 
organizational support predicted positively work engagement significantly 
(β=0.425, p<0.001) and explained additional 18.6% variance after controlling 
for demographics; digital literacy moderated this highly significantly (β=0.168, 
p<0.01, ΔR²=0.026), and high digital literacy faculties exhibited 2.35 times 
stronger strength of relations between AI support and engagement than low 
digital literacy faculties, and moderation being the highest for vigor dimension 
(β=0.185); bootstrap analysis with resamples 5,000 and sample split validation 
confirmed stability of such effects. By conceptualizing digital literacy as a 
central boundary condition, the current study extends JD-R theory to digital 
environments and describes another human-AI interaction in which AI 
complements but does not substitute human capacity and presents empirical 
evidence of universities to implement all-encompassing digital literacy training 
programs in parallel with AI system installation, although the cross-sectional 
study limits causal inference, findings are theoretically meaningful and 
practically informative and present visionary insight for knowing and 
promoting faculty well-being in the digital age.  

1. Introduction 

Universities worldwide are rapidly integrating AI 
technologies, including intelligent planning, adaptive 
learning, research support, and administrative automation 
systems [1,2]. This rapid adoption is primarily due to 
technological innovation and the institutional recognition of 
AI's potential to address long-standing challenges in higher 
education. While AI's impact on student learning [3] and 
academic integrity [4] is well-studied, less is known about 
faculty work engagement. Recent studies in 2024-2025 reveal 
growing concerns about generative AI tools like ChatGPT in 
academic settings. A qualitative research study [5] 
demonstrated that while AI tools offer productivity benefits 

and interactive learning opportunities, they simultaneously 
raise significant academic integrity concerns among both 
students and faculty. Another study [6] emphasizes that the 
rapid proliferation of AI technologies has significantly 
transformed educational assessment practices, requiring 
institutions to rethink exam design and develop ethical AI 
policies to maintain academic integrity. This imbalance is 
concerning, as faculty members serve as the primary interface 
for instruction and AI technology, and the extent of their 
involvement directly impacts instruction quality and research 
productivity, and indirectly affects student success. Faculty 
work engagement—energy, commitment, and absorption 
[7]—predicts teaching quality, research productivity, and 

 

 Open Access Journal 

ISSN 2832-0379 

February 2026| Volume 05 | Issue 01 | Pages 222-233 

https://doi.org/10.55670/fpll.futech.5.1.19 

 

 

 

 

 

 

 

 

 

 

 

 

Journal homepage: https://fupubco.com/futech 

 

Future Technology 

mailto:andi@ums.edu.my
https://doi.org/10.55670/fpll.futech.5.1.19
https://fupubco.com/futech


Z. Qian et al. /Future Technology                                                                                       February 2026| Volume 05 | Issue 01 | Pages 222-233 

223 

 

institutional performance. Academic work today, however, is 
increasingly demanding, with faculty constantly being 
saddled with gargantuan teaching loads, growing research 
expectations, administrative tasks, and an ongoing 
requirement to learn to keep up with mounting technical 
changes [8]. According to the Job Demands-Resources (JD-R) 
theory [9], support systems based on AI can be conceived as 
latent job resources that help faculty members manage work 
demands, achieve professional accomplishments, and 
maintain psychological well-being. Here, AI systems capable 
of carrying out tasks competently to assist faculty in teaching, 
research, and administrative work should ideally raise work 
engagement by offering resources that buffer against work 
demands. Nevertheless, recent empirical work offers mixed 
evidence on the extent to which AI impacts employees' 
outcomes. Liu and Li's research indicated that AI use is 
associated with higher work engagement, characterized by 
greater psychological availability and lower cognitive load 
[10], which aligns with empowerment theory, which posits 
that human abilities are supplemented by AI assistance. 
Conversely, Meng et al. [11] found that AI collaboration is 
associated with higher levels of counterproductive work 
behavior, including greater perceived aloneness and 
emotional exhaustion, as they perceived that AI would 
substitute for substantive human communication.  

Existing literature has key gaps. First, individual 
differences—notably digital literacy—have been neglected as 
boundary conditions determining whether AI systems are 
empowering or alienating. Second, most studies examine 
business settings, leaving higher education underexplored. 
Third, the multidimensionality of work engagement is seldom 
studied in AI contexts. Studies of digital literacy in higher 
education [12] and new digital competence models [13] also 
indicate that teachers' ability to work with technology may 
significantly affect their interactions with AI systems. 
Teachers with greater digital literacy are assumed to use AI 
tools more efficiently, seeing them as empowering 
technologies that simplify rather than complicate tasks [14]. 
Lower digital literacy levels, on the other hand, can be a 
hindrance to the effective deployment of AI systems, leading 
to frustration, anxiety, or disaffection [15]. This study 
addresses these gaps by examining AI-supported work 
engagement relationships and the moderating role of digital 
literacy.  

2. Literature review and research hypotheses 

2.1 AI-driven organizational support systems 
AI-based organizational support is operationally defined 

as faculty members' perceptions of the availability, 
accessibility, and usefulness of institutionally provided AI 
systems across three domains: teaching (e.g., automated 
grading, content generation), research (e.g., literature 
synthesis, data analysis), and administration (e.g., scheduling, 
document processing). At the tertiary level, AI support 
systems operate across three spheres. The latest evidence 
from 2024-2025 shows accelerated AI adoption in higher 
education institutions worldwide. A comprehensive study 
[16] examined generative AI adoption strategies across 40 
universities from six global regions, finding that institutions 
are proactively developing guidelines for ethical AI use, 
designing authentic assessments, and providing training 
programs to foster AI literacy among faculty and students. 
Research findings [17] report that faculty increasingly view 
AI tools as valuable for extending limited time resources, 
overcoming language barriers, and creating personalized 
learning experiences, although concerns about academic 

integrity and AI misuse remain prevalent. This rapid 
integration of generative AI tools like ChatGPT, Claude, and 
institutional AI systems into faculty workflows represents a 
fundamental shift in how academic work is conducted across 
teaching, research, and administrative domains. AI support 
spans three domains: (1) administrative support via 
intelligent scheduling and automated grading; (2) teaching 
support through adaptive learning platforms and AI content 
creation; (3) research assistance via AI literature review and 
data analysis tools [18]. 

Zhang et al. [19] meta-reviewed 87 studies, identifying a 
three-phase adoption model: initial resistance, gradual 
acceptance through experimentation, and integration. Digital 
literacy was the most significant driver across all stages. 
Reference [20] reported that Chinese university AI 
investment grew 340% from 2019 to 2024, with large 
institutional gaps. Universities with digital literacy training 
programs were 2.8 times more likely to invest in AI, 
suggesting the importance of human capital investment 
alongside technology." It is important to distinguish between 
conceptually related constructs in this study. AI-based 
organizational support refers to the institutional provision of 
AI technologies and systems (an external, organizational-
level resource). Digital literacy represents an individual's 
capability to effectively use digital technologies (an internal, 
individual-level competency). Digital self-efficacy, a 
component of digital literacy, captures explicit confidence 
beliefs about one's ability to use technology. While TAM's 
perceived ease of use overlaps conceptually with digital self-
efficacy, our study measures digital literacy as a broader 
competency encompassing both skills and confidence, 
whereas AI organizational support is measured through 
perceptions of system quality and institutional provision. 
This distinction prevents theoretical redundancy by 
examining organizational resources (what is provided) 
separately from individual capacities (ability to utilize what is 
provided). 

2.2 Work engagement 
Work engagement is operationally defined as a 

persistent, positive affective-motivational state comprising 
three dimensions: vigor (high energy and mental resilience), 
dedication (strong involvement and enthusiasm), and 
absorption (deep concentration and pleasant immersion in 
work), measured by UWES-9 [18]. The UWES-9 demonstrates 
excellent psychometric properties and predicts teaching 
competence (r=.42), research productivity (r=.38), and 
retention (r=-.45 with turnover) [21]. Academic engagement 
differs from organizational settings, spanning multiple roles 
(teaching, research, administration) with varied temporal 
rhythms. Understanding how AI supports these patterns is 
important for faculty well-being. 

2.3 Digital literacy as moderator 
Digital literacy is operationally defined as the integrated 

set of technical skills, cognitive abilities, and confidence 
beliefs required to effectively locate, evaluate, create, and 
communicate information using digital technologies in 
academic contexts [12]. Digital literacy is the competence, 
skills, and knowledge needed to effectively use digital 
technologies for information processing, communication, and 
problem-solving. Digital literacy has transformed in higher 
education from fundamental computer skills to start-to-finish 
competencies such as critical analysis of digital information, 
digital content creation, online collaborative work, and ethical 
technology use. Hobfoll's Conservation of Resources theory 
[19] indicates that human resources, such as digital literacy, 



Z. Qian et al. /Future Technology                                                                                       February 2026| Volume 05 | Issue 01 | Pages 222-233 

224 

 

increase the value and usability of organizational resources, 
potentially allowing faculty to better leverage AI systems. We 
propose digital literacy as the moderator rather than AI 
system design or institutional context for three theoretical 
reasons. First, meta-analytic evidence from technology 
acceptance research demonstrates that user competencies 
explain more variance in technology benefits than system 
features. Second, the Conservation of Resources theory 
suggests that personal resources, such as digital literacy, 
determine how effectively individuals can convert 
organizational resources into engagement outcomes. Third, 
educational technology studies specifically show that teacher 
digital competence is the primary boundary condition for 
successful technology integration, regardless of system 
quality. Recent evidence indicates widespread heterogeneity 
in digital literacy levels among academic staff in universities. 
A large-scale survey by Martin and Grudziecki [22] across 42 
European universities showed that while 78% of academic 
staff reported being digitally competent, only 34% were 
objectively tested to have highly developed digital literacy 
skills. The difference between reported and actual capacity 
was very high in fields involving new technologies, such as AI 
and machine learning. Second, substantial differences were 
found between demographic subgroups: younger professors 
(less than 40) scored 2.1 times higher in digital literacy 
compared to older professors, and STEM professors scored 
1.8 times higher than humanities professors. These 
differences indicate that digital literacy may become a 
stratifying variable of primary importance for AI system 
adoption and performance. 

Digital literacy has also been reportedly associated with 
the acceptance of technology. Venkatesh and Bala's [23] 
Technology Acceptance Model 3 (TAM3) cites computer self-
efficacy, by virtue of its direct association with digital literacy, 
as one of the major drivers of perceived ease of use, which in 
turn affects adoption and long-term use of technology. In the 
specific case of AI systems, increased faculty digital literacy 
will most probably result in: (1) better estimation of AI 
capabilities and boundaries, (2) proper incorporation of AI 
tools within workflows, (3) resolving technical problems 
independently, and (4) investigation of advanced features to 
increase productivity. Less digitally literate faculty, on the 
other hand, might suffer from "technostress," defined as 
feelings of anxiety, frustration, and avoidance when faced 
with AI systems. New models have expanded the idea of 
digital competence in learning [20], not just technical 
competencies but pedagogic inclusion and ethics as well. The 
European Framework for Digital Competence of Teachers 
(DigCompEdu) identifies 22 competences distributed over six 
categories: professional activity, digital resources, instruction 
and learning, assessment, empowering learners, and enabling 
learners' digital competence. This combined model stresses 
that the successful application of AI systems in classrooms 
depends not only on technical ability but also on the ability to 
situate technology usefully into the pedagogical process and 
research methods. Teachers with the ability to bring these 
skills together are more apt to utilize AI systems as 
transformative forces than as productivity enhancers. 

2.4 Theoretical framework and hypotheses 
Three allied theoretical models are employed in the 

current study to describe the intricate interplay between 
work engagement, digital literacy, and AI support. 
Technology Acceptance Model (TAM) [14] describes how 
perceived usefulness and ease of use affect technology 
adoption. Our AI support scale implicitly captures these 

dimensions: items like 'AI systems help me complete tasks 
efficiently' reflect perceived usefulness, while 'easy to 
integrate' taps ease of use. This 9-item scale serves as a proxy 
for TAM constructs, aligning with TAM3 research suggesting 
these can combine into 'perceived system quality' for 
established systems. Self-efficacy theory [15] argued that 
people's beliefs about themselves affect their motivation and 
actions when interacting with technology. Staff with greater 
digital self-efficacy tend to use AI systems with confidence, 
venture to discover their potential, and be persistent with 
them despite difficulties. This theoretical assumption points 
to digital literacy as not just a set of skills but also a confidence 
builder that shapes technology use. Drawing on Job Demands-
Resources (JD-R) theory [9], they offer an integrated model to 
account for work engagement's relationship with AI systems 
and the potential mediating role of digital literacy as a 
moderator. JD-R theory posits that organisational support, 
autonomy, and technological resources can buffer the effects 
of job demands and improve work engagement. Integrating 
these theories leads us to the conclusion that the impacts of 
AI systems on participation are influenced not only by 
objective factors (resources provided) but also by subjective 
ones (perceived usefulness, self-confidence), of which digital 
literacy is a major variable shaping perceptions and 
experiences. Figure 1 depicts the conceptual model guiding 
this study, illustrating the hypothesized relationships 
between AI-driven organizational support, digital literacy, 
and work engagement. 

 

 

 

 

 

 

 

 

 
 
Figure 1. Conceptual model of the study 

Grounded in the theoretical review and synthesis of 
these models, we advance the following hypotheses: 
The conceptual model integrates three theoretical 
perspectives. TAM explains how perceived usefulness and 
ease of use influence initial adoption of AI systems. Self-
efficacy theory, operationalized through digital literacy, 
shapes individuals' confidence in utilizing AI tools. JD-R 
theory positions AI support as a job resource that directly 
enhances engagement, while digital literacy serves as a 
personal resource that moderates (strengthening or 
weakening) this relationship. Specifically, digital literacy may 
function as a moderator rather than a mediator because it 
influences the strength of the AI support-engagement 
relationship rather than serving as an intermediate step in a 
causal chain. 
Hypothesis 1 (H1): AI-based organizational support is 
positively associated with overall work engagement. 
Specifically: H1a: AI-based organizational support is 

 



Z. Qian et al. /Future Technology                                                                                       February 2026| Volume 05 | Issue 01 | Pages 222-233 

225 

 

positively associated with vigor (the energy component of 
engagement). 
H1b: AI-based organizational support is positively associated 
with dedication (the involvement component of engagement). 
H1c: AI-based organizational support is positively associated 
with absorption (the immersion component of engagement). 
Note: Vigor, dedication, and absorption are treated as 
components (subdimensions) of the higher-order work 
engagement construct, not as separate dependent variables. 

This hypothesis is based on the JD-R theory's contention 
that job resources increase work engagement. Positive 
associations operate through distinct mechanisms. For vigor 
(H1a), AI-automated grading reduces fatigue from repetitive 
tasks, preserving energy for creative teaching. For dedication 
(H1b), AI research synthesis tools facilitate deeper 
intellectual engagement by reducing mechanical search 
burdens, allowing focus on conceptual connections. For 
absorption (H1c), AI-assisted administrative tasks minimize 
paperwork interruptions, enabling concentration on core 
academic work. Each domain (teaching, research, 
administration) links to cognitive load reduction and 
efficiency gains targeting these engagement facets. 
Hypothesis 2 (H2): Digital literacy positively moderates the 
relationship between AI-based organizational support and 
work engagement, such that the positive association is 
stronger for faculty with higher digital literacy. Specifically: 
H2a: The moderating effect is significant for vigor. 
H2b: The moderating effect is significant for dedication 
H2c: The moderating effect is significant for absorption. 

Based on TAM and self-efficacy theory, we pre-register 
expected simple slopes: at +1 SD digital literacy, steep 
positive slope (β > .40) indicating strong AI responsiveness; 
at -1 SD, weaker positive slope (β < .25) indicating limited 
leverage capacity; at mean, moderate slope (β ≈ .30-.35). We 
do not expect negative slopes at any literacy level, as even 
low-literacy faculty should benefit from well-designed AI. If 
the interaction is significant, the slope difference should be 
substantial (Δβ > .15) and the confidence intervals should not 
overlap. Less digitally literate faculty might not be able to use 
AI systems effectively, leading them to become frustrated 
rather than more engaged. 

3. Methods 

3.1 Research design 
This research used a cross-sectional survey design. We 

recognize that cross-sectional designs exclude causal 
inference, can't exclude reverse causation, and are prone to 
third-variable confounding. Findings need to be interpreted 
as correlational rather than causal. The research was 
conducted in a three-month period (March-May 2024), and 
ethical clearance was obtained from the Institutional Review 
Board (Approval No. 2024-HR-087). Electronic informed 
consent covered:  
(1) study purpose/procedures,  
(2) voluntary participation/withdrawal rights,  
(3) confidentiality/anonymity,  
(4) no compensation.  
Data: password-protected servers (SSL, 5-year retention), 
random IDs, no IP tracking. No personally identifiable 
information was collected; participants were assigned 
random ID codes. IP addresses were not recorded to ensure 
anonymity. Consent was implied through survey completion, 
as explicitly stated in the introduction. No incentives or 
compensation were provided to participants to minimize the 
risk of coercion. Data collection used the Wenjuanxing 
platform for security (SSL encryption, anonymous 

responses). The survey was launched via institutional 
networks with weekly reminders. To reduce response 
burden, we pilot-tested the survey with 30 faculty members 
(15 from teaching universities, 15 from research 
universities). Based on pilot feedback, we made the following 
refinements: (1) rewording 3 items for clarity (e.g., changing 
'AI system facilitates my work' to 'AI system helps me 
complete tasks more efficiently'), (2) shortening the survey 
from 18 to 15 minutes median completion time by removing 
redundant demographic items, and (3) adding progress 
indicators to reduce abandonment. Reliability analysis 
showed improved Cronbach's alpha values after refinement: 
AI Support scale increased from α=0.87 to α=0.91, Digital 
Literacy from α=0.82 to α=0.86. Cognitive interviews with 5 
pilot participants revealed no comprehension difficulties with 
the revised items. The survey remained open for six weeks to 
allow for different schedules and workloads of faculty 
members during the semester. 

3.2 Sample 
The target sample was China's full-time university 

faculty members with at least one year of teaching experience 
and exposure to AI-based organizational support systems, 
operationally defined as having used at least one 
institutionally-provided AI tool (learning management 
system with AI features, AI-assisted grading, or AI research 
tools) for a minimum of 6 months with at least weekly usage 
frequency. Convenience sampling and snowball sampling 
were adopted. A priori power analysis using G*Power [22] 
suggested a minimum sample of 269 for detecting small-to-
medium moderation effects (f²=0.03, power=0.80, α=0.05). A 
total of 450 faculty accessed the survey, and 412 completed 
the survey. Following data screening, the ultimate analytical 
sample included 387 faculty members. Full demographic 
characteristics of the sample are presented in Table 1. 

Procedures for data screening were applied to careless 
responding response types (e.g., straight-lining, inadmissible 
response times of less than 5 minutes), multivariate outliers 
by Mahalanobis distance (p < 0.001), and primary variable 
completeness. Twenty-five cases were removed: 18 for 
missing primary variable data, 5 for lack of sufficient 
attention checks, and 2 for status as a multivariate outlier. The 
ultimate sample of 387 consisted of professors from 12 
universities distributed over three geographic regions 
(Eastern: 52.2%, Central: 28.9%, Western: 18.9%) to 
maximize generalizability of findings to the Chinese context. 
We acknowledge that convenience and snowball sampling 
may introduce self-selection bias, as faculty who are more 
comfortable with technology are likely overrepresented in 
our sample. To assess this limitation, we conducted 
nonresponse bias testing by comparing early respondents 
(first 25%, n=97) with late respondents (last 25%, n=97) on 
key variables. Independent t-tests revealed no significant 
differences in AI support (t=1.24, p=0.216), digital literacy 
(t=0.89, p=0.374), or work engagement (t=1.47, p=0.143), 
suggesting minimal nonresponse bias. Additionally, our 
sample's digital literacy mean (M=5.23) is slightly higher than 
reported population norms for Chinese university faculty 
(M=4.87), indicating moderate positive selection that should 
be considered when generalizing findings. Post-hoc 
sensitivity analysis indicated sufficient power (.82) to detect 
the hypothesized moderation effect with our ultimate sample 
size. Universities provided AI systems for teaching 
(intelligent LMS, automated assessment, AI writing 
assistants), research (literature search tools like Connected 
Papers, reference managers like Zotero), and administration 



Z. Qian et al. /Future Technology                                                                                       February 2026| Volume 05 | Issue 01 | Pages 222-233 

226 

 

(scheduling, document processing). Faculty exposure varied 
by institution type. 

Table 1. Sample characteristics (N=387) 

 

3.3 Measures 
All these were measured using standardized scales. 

Items were scored on 7-point Likert-type scales. 
Questionnaires were translated into Chinese using standard 
translation-back-translation procedures [24]. Standard 
translation-back-translation: two independent forward 
translations, expert synthesis, blind back-translation, 
comparison, and pilot testing (n=10 bilingual faculty). AI-
based Organizational Support: A 9-item multidimensional 
scale originally developed by adapting items from Perceived 
Organizational Support Scale and modified for AI context by 
the research team. The scale measures teaching support (3 
items, e.g., 'AI systems help me design better learning 
activities'), research support (3 items, e.g., 'AI tools assist me 
in literature review and synthesis'), and administrative 
support (3 items, e.g., 'AI systems reduce time spent on 
administrative tasks'). Overall α=0.91, subscales αs=0.85-
0.88. Sample AI Support scale items specify concrete 
technologies to enhance clarity and anchoring. Teaching 
support items include 'AI-powered platforms like intelligent 
tutoring systems help me provide personalized feedback to 
students,' 'AI content generators (e.g., automated quiz 
creation tools) reduce my course preparation time,' and 
'Learning management systems with AI recommendations 
improve my course design.' Research support items 
encompass 'AI tools such as ChatGPT, Claude, or similar 
assistants help me synthesize research literature,' 'AI-

enhanced reference managers (e.g., Zotero with ML 
recommendations, Connected Papers, Semantic Scholar) 
improve my literature organization,' and 'AI-powered data 
analysis tools facilitate my research methodology.' 
Administrative support items include 'AI scheduling systems 
optimize my course timetables and office hours,' 'Grammarly, 
Wordtune, or similar AI writing assistants help me draft 
professional communications efficiently,' and 'AI-powered 
document processing reduces time on routine administrative 
paperwork.' 

Items anchor perceptions to concrete technologies. Our 
March-May 2024 data captured early post-ChatGPT adoption, 
when generative AI shifted from specialized to ubiquitous 
tools. Findings reflect AI as supplementary productivity tool 
in early adoption phase. As AI capabilities expand toward 
autonomous analysis (post-2024), updated measurements 
will be needed to capture evolving faculty-AI interaction 
patterns. 
Digital Literacy was measured with 4 items adapted from Ng 
[12]: 
• 'I can effectively use digital technologies for teaching and 

research' (technical competence) 
• 'I can troubleshoot common technical problems 

independently' (technical competence) 
• 'I feel confident learning new digital tools' (self-efficacy) 
• 'I am comfortable integrating emerging technologies into 

my work' (self-efficacy) All items used 7-point Likert scales 
(1=Strongly Disagree, 7=Strongly Agree). α=0.86. 

Work Engagement: Utrecht Work Engagement Scale (UWES-
9) developed by Schaufeli et al. (2006), measuring vigor (3 
items, e.g., 'At my work, I feel bursting with energy'), 
dedication (3 items, e.g., 'I am enthusiastic about my job'), and 
absorption (3 items, e.g., 'I feel happy when I am working 
intensely'). Composite α=0.93, subscale αs: vigor=.88, 
dedication=0.90, absorption=0.87. 

Control Variables: We included five demographic 
controls based on prior research linking these characteristics 
to technology adoption and work engagement. Gender was 
controlled because meta-analytic evidence demonstrates that 
males report slightly higher technology self-efficacy and more 
favorable attitudes toward technology use than females, 
although these differences are characterized as small effect 
sizes [25]. Age was included as younger workers' technology 
usage decisions are more strongly influenced by attitude 
toward using technology, while older workers are more 
influenced by subjective norms and perceived behavioral 
control [26]. Teaching experience was controlled because 
veteran faculty may exhibit different engagement patterns 
and technology resistance compared to novice faculty, 
reflecting accumulated work habits and established 
pedagogical approaches. Academic rank was included as 
seniority correlates with work engagement, autonomy, and 
resource access in academic settings, with senior faculty often 
having greater discretion in technology adoption decisions. 
Discipline was controlled because STEM faculty consistently 
demonstrate higher digital literacy and technology 
integration rates compared to humanities and social science 
faculty, reflecting differences in disciplinary norms and 
technology exposure. Scale validation methods went beyond 
reliability measurement. For AI-based Organizational 
Support scale adapted for a university context, we first did 
exploratory factor analysis with pilot sample (n=30) and then 
confirmed it using confirmatory factor analysis with the 
entire sample. The three-factor solution (teaching, research, 
administrative support) was found to have satisfactory fit 

Variable n % M(SD) 

Gender    

Male 198 51.2  

Female 189 48.8  

Age (years) 387  41.3 (8.7) 

25-35 98 25.3  

36-45 176 45.5  

46-55 89 23.0  

56+ 24 6.2  

Teaching Experience (years) 387  12.6 (7.9) 

1-5 years 124 32.0  

6-10 years 98 25.3  

11-20 years 132 34.1  

20+ years 33 8.5  

Academic Rank    

Lecturer 89 23.0  

Assistant Professor 126 32.6  

Associate Professor 121 31.3  

Full Professor 51 13.2  

Discipline    

STEM 186 48.1  

Social Sciences 98 25.3  

Humanities 76 19.6  

Other 27 7.0  

University Type    

Research University 213 55.0  

Teaching University 174 45.0  



Z. Qian et al. /Future Technology                                                                                       February 2026| Volume 05 | Issue 01 | Pages 222-233 

227 

 

(χ²/df = 2.31, CFI = .95, TLI = 0.94, RMSEA = 0.058, SRMR = 
0.045).  

3.4 Data analysis 
Data analysis involved five steps. We first screened for 

data quality and identified missing-data patterns. Second, we 
computed descriptive statistics. Third, we did confirmatory 
factor analysis. Fourth, we assessed common method bias 
using Harman's single-factor test and the common latent 
factor method [27]. Finally, we tested hypotheses through 
hierarchical multiple regression analysis. All continuous 
predictors were mean-centered to simplify the interpretation 
of interaction terms [28]. For medium-level interactions, we 
performed simple slopes analysis with the PROCESS macro 
[21]. Additional analysis steps improved the stability of our 
results. Multicollinearity was checked using variance inflation 
factors (all VIF values < 3.0) and tolerance levels (all >0.30) 
and was not found to be an issue. Heteroscedasticity was 
checked using Breusch-Pagan tests with no evidence of 
material violation. To address potential endogeneity 
concerns inherent to cross-sectional data, we conducted an 
instrumental variables (IV) regression using two-stage least 
squares (2SLS). We used institutional AI investment intensity 
(measured as the annual per-faculty AI budget in RMB, log-
transformed) as an instrument for individual-level 
perceptions of AI support. The instrument is theoretically 
valid because institutional investment determines AI system 
availability (relevance assumption), but should not directly 
affect individual engagement except through AI support 
perceptions (exclusion restriction assumption). 

First-stage regression results confirmed instrument 
strength: AI investment significantly predicted AI support 
perceptions (β=0.389, t=7.66, p<0.001), with F-
statistic=58.73, far exceeding the rule-of-thumb threshold of 
F>10 for weak instrument concerns. The Kleibergen-Paap 
Wald F-statistic was 56.42, also indicating a strong 
instrument. When we added university type (research vs. 
teaching) as a second instrument, the Sargan-Hansen J-test 
for overidentification restrictions yielded χ²(1)=2.14, 
p=0.144, failing to reject the null hypothesis of valid 
instruments, supporting the exclusion restriction. Second-
stage results showed that AI support (instrumented) 
remained significantly associated with engagement (β=0.397, 
SE=0.087, p<0.001), with a magnitude similar to the OLS 
estimate (β=0.425), suggesting minimal endogeneity bias.  

 

 

 

 

 

 

 

 

 

 

 

 

 

The Durbin-Wu-Hausman test comparing IV and OLS 
estimates was nonsignificant (χ²=1.89, p=.169), indicating 
OLS estimates are consistent and endogeneity is not a major 
concern. These IV analyses provide additional confidence in 
the directionality of relationships, though causal inference 
remains limited by cross-sectional design. These analyses 
cannot determine causality but add extra confidence to the 
directionality of relationships identified. We also conducted 
bootstrap analysis (5,000 resamples) to yield bias-corrected 
confidence intervals for all parameter estimates, especially 
for interaction effects that are potentially sensitive to 
distributional assumptions. We then conducted robustness 
checks by re-estimation with other operationalizations (e.g., 
median splits of digital literacy) and testing for potential 
curvilinear effects via polynomial regression. All findings of 
primary interest were substantively identical under these 
alternative specifications. 

4. Results 
4.1 Descriptive statistics 

Table 2 presents means, standard deviations, and 
intercorrelations among all study variables. 

4.2 Measurement model 
Confirmatory factor analysis examined factorial and 

discriminant validity [29], ensuring items loaded on intended 
constructs and constructs were empirically distinguishable 
despite theoretical proximity. We compared our proposed 
three-factor model with other nested alternative models to 
assess discriminant validity. The two-factor model combined 
AI assistance and computer proficiency into a single factor, 
suggesting that teachers were not separating technology tools 
from the skills needed to make them function. The one-factor 
model indicated that all items loaded onto a general positive 
response factor. We also tested a common latent factor (CLF) 
model to further evaluate common method bias in addition to 
Harman's test. Table 3 reports fit indices for the competing 
models. The three-factor model demonstrated excellent fit 
(CFI=0.918, TLI=0.906, RMSEA=0.066, SRMR=0.052), 
meeting recommended thresholds [30]. The CFA fit indices 
(CFI=0.918, TLI=0.906) approach but slightly fall below the 
stringent 0.95 threshold sometimes cited. However, these 
values are acceptable given several considerations.  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 2. Descriptive statistics and correlation matrix 

Variable M SD α 1 2 3 4 5 6 7 8 9 10 

1. Gender 1.49 0.50 — —          

2. Age 41.26 8.73 — -.08 —         

3. Teaching Experience 12.58 7.94 — -.06 .87** —        

4. Rank 2.35 1.01 — .11* .42** .45** —       

5. AI Support 4.82 1.15 .91 -.02 -.09 -.07 .05 —      

6. Digital Lit 5.23 0.94 .86 .04 -.12* -.10 .08 .32** —     

7. Engagement 4.95 1.18 .93 -.05 -.14** -.12* .03 .45** .38** —    

8. Vigor 4.78 1.26 .88 -.08 -.16** -.13* .02 .40** .35** .92** —   

9. Dedication 5.02 1.23 .90 -.03 -.11* -.09 .04 .43** .37** .94** .82** —  

10. Absorption 5.05 1.21 .87 -.04 -.12* -.11* .03 .39** .36** .93** .81** .84** — 

Note: N=387. *p<0.05. ** p<0.01 

 



Z. Qian et al. /Future Technology                                                                                       February 2026| Volume 05 | Issue 01 | Pages 222-233 

228 

 

First, with our sample size (N=387) and model 
complexity (17 indicators across 3 factors), simulation 
studies show CFI/TLI values of 0.90-0.95 are acceptable when 
RMSEA and SRMR are good. Second, RMSEA=0.066 and 
SRMR=0.052 are within acceptable ranges (<0.08 for both). 
Third, comparative fit against alternative models shows 
substantial improvement: our three-factor model fits 
significantly better than two-factor (ΔCFI=0.176, Δχ²=565.74, 
Δdf=2, p<0.001) and one-factor models (ΔCFI=0.406, 
Δχ²=1422.46, Δdf=3, p<0.001), providing strong evidence for 
discriminant validity. Fourth, the chi-square value is 
χ²(149)=342.15, p<0.001, yielding χ²/df=2.30, which is 
within the acceptable 2-3 range. Given that we prioritize 
construct validity over perfect fit indices, and given our 
theoretical rationale for the three-factor structure, we accept 
this model as adequately representing the data. Confirmatory 
factor analysis confirmed that digital literacy and AI support 
loaded on distinct factors with no problematic cross-loadings. 
All items loaded primarily on their intended factors (λ > 0.60), 
with cross-loadings not exceeding 0.40. The correlation 
between digital literacy and AI support (r=0.32, Table 2) is 
moderate, indicating related but distinguishable constructs. 
Discriminant validity was further supported by the Fornell-
Larcker criterion: the square root of AVE for digital literacy 
(0.78) exceeded its correlation with AI support (0.32), and the 
square root of AVE for AI support (0.76) exceeded its 
correlation with digital literacy (0.32), confirming these 
measures capture distinct variance. Table 4 presents 
standardized factor loadings, composite reliability (CR), and 
average variance extracted (AVE) for all constructs. All factor 
loadings exceeded 0.60, with most above 0.70. CR values 
ranged from 0.86 to 0.93, all exceeding the 0.70 threshold. 
AVE values ranged from 0.58 to 0.72, all exceeding the 0.50 
criterion, supporting convergent validity per Fornell and 
Larcker (1981). Square roots of AVE (diagonal in correlation 
matrix) exceeded inter-construct correlations, confirming 
discriminant validity. The three-factor model fit significantly 
better than alternative models, providing strong evidence for 
discriminant validity. Harman's single-factor test revealed 
that the first factor accounted for 26.8% of the variance, 
below the 50% threshold for substantial method bias.  

Table 3. Confirmatory factor analysis fit indices 

Model χ² df CFI TLI RMSEA SRMR 

M1: Three-factor 342.15*** 149 0.918 0.906 0.066 0.052 

M2: Two-factor 907.89*** 151 0.742 0.718 0.124 0.095 

M3: One-factor 1764.61*** 152 0.512 0.478 0.178 0.142 

M4: M1+CLF 319.65*** 131 0.927 0.913 0.062 0.048 

 
The common latent factor (CLF) method provides a more 

stringent assessment of common method variance than 
Harman's test. We compared the three-factor model (M1: 
χ²=342.15, df=149, CFI=0.918) against a model adding a CLF 
onto which all indicators loaded (M4: χ²=319.65, df=131, 
CFI=0.927). The improvement was minimal (ΔCFI=0.009, 
Δχ²=22.50, Δdf=18, p=0.212), suggesting CMV is not 
substantial. Standardized loadings on the CLF ranged from 
0.08 to 0.19 (M=0.13), indicating the common method factor 
explains only 1.7% of variance on average (calculated as 
mean squared loading: 0.13²=0.017). This is well below the 
25% threshold typically considered problematic. 
Additionally, substantive factor loadings remained large and 
significant after controlling for CLF (all λ >0.60), confirming 

that our constructs capture meaningful variance beyond 
method effects. The difference in fit indices between 
constrained (M1) and CLF models (M4) was negligible: 
ΔRMSEA=0.004, ΔTLI=0.007, ΔSRMR=0.003, all indicating 
minimal method variance. Method variance accounts for 
approximately 17% of total variance (calculated from CLF 
model R²), below the 20% recommended threshold, 
confirming common method bias is not a major threat to our 
findings. Beyond post-hoc statistical tests, we implemented 
several procedural remedies during data collection to 
minimize common method bias: (1) Psychological separation: 
Different constructs were presented in varied sections with 
buffer items between them. (2) Question order 
counterbalancing: In 50% of surveys, work engagement items 
appeared before AI support items to control for priming 
effects. Comparison showed no significant differences 
between versions (F=0.87, p=0.352).  

Table 4. Standardized factor loadings, composite reliability, and 
average variance extracted 

 

Construct Item Factor Loading CR AVE 

AI-Based 
Organizational 

Support 

  0.91 0.58 

Teaching Support AIS1 0.76   

 AIS2 0.79   

 AIS3 0.82   

Research Support AIS4 0.77   

 AIS5 0.73   

 AIS6 0.75   

Administrative 
Support 

AIS7 0.71   

 AIS8 0.74   

 AIS9 0.68   

Digital Literacy   0.86 0.61 

Technical 
Competence 

DL1 0.81   

 DL2 0.84   

Self-Efficacy DL3 0.79   

 DL4 0.72   

Work 
Engagement 

  0.93 0.72 

Vigor WE1 0.85   

 WE2 0.81   

 WE3 0.83   

Dedication WE4 0.88   

 WE5 0.86   

 WE6 0.84   

Absorption WE7 0.86   

 WE8 0.82   

 WE9 0.84   



Z. Qian et al. /Future Technology                                                                                       February 2026| Volume 05 | Issue 01 | Pages 222-233 

229 

 

(3) Anonymity assurance: The survey introduction 
emphasized complete anonymity and no individual-level 
reporting. (4) Different scale anchors: We varied response 
formats (Strongly Disagree-Strongly Agree vs. Never-Always) 
across constructs where appropriate. (5) Clear item wording: 
Items avoided ambiguous terms and double-barreled 
questions. These procedural controls complement our 
statistical tests, strengthening confidence that common 
method bias is not a major threat. 

4.3 Hypothesis testing 
Hierarchical regression results are presented in Table 5. 

Hierarchical regression tested hypotheses in four steps 
(Table 5). Controls (Step 1) explained 9.5-11.6% variance (all 
p<0.001). AI support (Step 2) substantially increased 
variance (ΔR²=0.137-0.186, all p<0.001). Digital literacy (Step 
3) contributed additional variance (ΔR²=0.019-0.025, 
p<0.01-.001). The interaction term (Step 4) uniquely 
explained incremental variance (ΔR²=0.017-0.031, p<0.05-
0.001), with final models explaining 27.7-35.1% variance. 
Artificial intelligence organizational support was strongly 
related to work engagement (β=0.425, p<0.001), validating 
H1 and its sub-hypotheses (H1a-H1c). Digital literacy 
moderated these relationships significantly (β=0.168, 
p=0.003), and moderation was strongest for vigor (β=0.185), 
validating H2 and its sub-hypotheses (H2a-H2c). 

Basic slopes analysis showed that the correlation was 
0.465 (p<0.001) for high digital literacy staff compared with 
0.198 (p<0.001) for low digital literacy staff—a 2.35-fold 
difference. To explore whether continuous variable 
dichotomization affects results, we tested the interaction 
using digital literacy as a continuous variable (reported 
above) versus a dichotomized variable.  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Using median-split (Mdn=5.25 on 7-point scale), we 
classified faculty as high (n=198, M=6.02, SD=0.48) versus 
low (n=189, M=4.41, SD=0.62) digital literacy. ANOVA with 
digital literacy group (high/low) as a between-subjects factor 
and AI support as continuous predictor confirmed significant 
interaction (F=16.34, p<0.001), with simple slopes for high 
group (β=0.482, p<0.001) versus the low group (β=0.204, 
p<0.001) yielding a 2.36-fold difference, nearly identical to 
the continuous analysis (2.35-fold). The median cut point 
(5.25) corresponds to 'moderately agree' on our scale, 
suggesting a meaningful threshold wherein faculty who are 
moderately proficient in digital technologies begin to fully 
leverage AI systems. Below this threshold, AI support shows 
attenuated benefits; above it, benefits are substantially 
amplified. Importantly, the interaction remained significant 
when using tertile splits (low/medium/high: F=12.87, 
p<.001) or treating digital literacy as fully continuous 
(reported in main results), confirming robustness to 
operationalization choices and addressing concerns about 
artificial dichotomization of continuous variables. On a 3-unit 
AI support increment, this is a 0.59-point increase in 
engagement for low literacy staff compared with 1.40 points 
for high literacy staff—a practically significant 0.81-point 
(0.69 SD) difference. To facilitate interpretation, we 
calculated Cohen's f² effect size for the moderation effect: 
f²=.026/.974=.027, representing a small-to-medium effect 
per Cohen's (1988) guidelines. We also computed simple 
slopes effect sizes: for high digital literacy, the AI support-
engagement relationship has Cohen's d=0.89 (large effect), 
while for low digital literacy d=0.38 (small-to-medium effect). 
The difference between slopes yields an effect size of Δd=0.51, 
indicating that digital literacy produces a meaningful 
practical difference in how strongly AI support relates to 
engagement.  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 5. Hierarchical regression results with incremental R² 

Predictor Work Engagement Vigor Dedication Absorption 

Step 1: Control Variables     

Gender -0.05 -0.08 -0.03 -0.04 

Age -0.08 -0.10 -0.07 -0.06 

Teaching Experience -0.04 -0.05 -0.03 -0.05 

Academic Rank 0.02 0.01 0.03 .002 

Discipline 0.11* 0.09 0.10* 0.12* 

R² 0.116*** 0.108*** 0.095*** 0.102**** 

F 10.02*** 9.24**** 8.05**** 8.68**** 

Step 2: Main Effect     

AI Support 0.425*** 0.398*** 0.412*** 0.376*** 

ΔR² 0.186**** 0.163**** 0.183**** 0.137**** 

Cumulative R² 0.302**** 0.271**** 0.278**** 0.239**** 

F change 101.24**** 84.67**** 97.45**** 69.28**** 

Step 3: Moderator     

Digital Literacy 0.214*** 0.195*** 0.220*** 0.205*** 

ΔR² 0.023**** 0.019***** 0.025**** 0.021***** 

Cumulative R² 0.325**** 0.290**** 0.303**** 0.260**** 

F change 12.89**** 10.17***** 13.45**** 10.64***** 

Step 4: Interaction     

AI × Digital Literacy 0.168** 0.185*** 0.152** 0.135* 

ΔR² 0.026***** 0.031**** 0.021***** 0.017**** 

Final R² 0.351**** 0.321**** 0.324**** 0.277**** 

Adjusted R² 0.339 0.307 0.310 0.262 

F change 14.87***** 16.74**** 11.23***** 8.87**** 

F (final model) 25.61**** 22.43**** 22.77**** 18.14**** 

 



Z. Qian et al. /Future Technology                                                                                       February 2026| Volume 05 | Issue 01 | Pages 222-233 

230 

 

Figure 2 depicts the moderation effect of digital literacy 
on the relationship between organizational support via AI and 
work engagement. The more positive steep slope is for high 
digital literacy staff than for low digital literacy staff, as shown 
in the figure, supporting the anticipated interaction effect. 

AI-driven Organizational Support
Low (-1SD) High (+1SD)

W
o

rk
 E

n
g

a
g

em
en

t

7.0

6.0

5.0

4.0

3.0

4.56

4.70
4.65

5.72
Interaction: β = .168, p = .003

ΔR² = .026, F(1,377) = 9.12**

Low Digital Literacy
(slope = .198**)

High Digital Literacy
(slope = .465***)

 

Figure 2. Interaction effect of AI-driven organizational support and 
digital literacy on work engagement 

To further visualize the moderation effect on the three 
work engagement dimensions, Figure 3 shows the simple 
slopes for vigor, dedication, and absorption in turn. From the 
figure, it can be seen that the moderation effect is strongest 
on vigor (panel A), followed by dedication (panel B) and 
absorption (panel C), as our hypothesis predicts a stronger 
effect of digital literacy on the energy dimension of 
engagement. Bootstrap analysis of 5,000 resamples validated 
all effects, providing bias-corrected confidence intervals that 
overcame potential distributional issues inherent in 
moderation analyses. Parameter estimates were highly 
congruent because the support coefficient of AI varied by less 
than 3% among bootstrap samples (range of β:0.413-0.437), 
and the moderation effect was positive in 98.8% of the 
resamples. The findings guarantee our results are not outliers 
or sampling variation. Sample split validation provided very 
similar results, the two randomly divided subsamples 
(n₁=194, n₂=193) generating essentially identical effect sizes 
for main effects (β₁=0.419, β₂=0.431) and interactions 
(β₁=0.171, β₂=0.165). Furthermore, k-fold cross-validation 
(k=10) reported little overfitting, estimates of R² per fold 
being close to the full-sample estimate (M=0.442, SD=0.038 
vs. full R²=0.446). These sets of exhaustive robustness checks 
all further enhance confidence in the generalizability and 
reliability of our findings.  

5. Discussion 
Findings confirmed AI support positively correlates with 

faculty engagement (β=0.425, p<.001), with digital literacy 
moderating this relationship (β=0.168, p<0.01). Effect sizes 
exceed typical technology acceptance studies [31,32], 
potentially due to heavy cognitive loads in academic work, 
where AI systems can provide substantial returns. 
Moderation was strongest for vigor (β=0.185), indicating that 
digital literacy most influences the energetic dimension of 
engagement. This study extends JD-R theory to digital 
environments by conceptualizing AI support as a job resource 

that buffers demands and enhances engagement [9]. It 
contributes to the human-AI collaboration theory by 
proposing a complementary relationship where AI enhances 
rather than replaces human capabilities [33], particularly 
relevant in academic settings where critical thinking remains 
irreplaceable. Digital literacy serves as a boundary condition 
involving "algorithmic thinking"—understanding AI logic, 
predicting limitations, and designing creative applications. 
Moderation stability across engagement dimensions 
indicates that digitally literate faculty apply AI strategically, 
achieving higher vigor, dedication, and absorption. This study 
extends JD-R theory to digital environments by 
conceptualizing AI support as a job resource that buffers 
demands and enhances engagement. It contributes to the 
human-AI collaboration theory by proposing a 
complementary relationship where AI enhances rather than 
replaces human capabilities [34], particularly relevant in 
academic settings where critical thinking remains 
irreplaceable. Digital literacy serves as a boundary condition 
involving "algorithmic thinking"—understanding AI logic, 
predicting limitations, and designing creative applications. 
Moderation stability across engagement dimensions 
indicates that digitally literate faculty apply AI strategically, 
achieving higher vigor, dedication, and absorption. 

Findings have important implications for university 
leaders. If causal studies confirm these associations, 
universities should prioritize digital literacy programs 
alongside AI implementation. Implementation should be 
institution-specific based on needs assessments. Digital 
literacy's pivotal role suggests a threshold effect—minimum 
competency is a prerequisite for AI to enhance rather than 
hinder engagement. Training should extend beyond technical 
procedures to include conceptual AI understanding, ethical 
implications, and innovative deployment approaches. Peer 
mentoring programs can effectively develop required 
competencies. Several limitations warrant consideration 
when interpreting these findings. First, a cross-sectional 
design precludes causal inference; reverse causation is 
plausible (engaged faculty may seek AI systems). 
Longitudinal designs tracking faculty across semesters would 
establish temporal precedence. Second, convenience and 
snowball sampling introduce self-selection bias, as our 
sample likely over-represents tech-savvy faculty comfortable 
with both AI systems and online surveys, potentially inflating 
effect sizes. While nonresponse bias tests showed no 
differences between early and late respondents, 
nonrespondents may differ systematically. Future studies 
should employ stratified random sampling with institution-
level cooperation to ensure representativeness. Third, the 
Chinese cultural context may limit generalizability, as China's 
collectivistic culture, top-down technology implementation, 
and government emphasis on AI adoption may amplify 
positive AI perceptions; the moderation effects might be 
weaker in individualistic cultures or contexts with faculty-
driven technology adoption. Cross-national studies 
comparing Asian, European, and North American universities 
would identify cultural boundary conditions. We explored 
collectivism's role post-hoc by incorporating province-level 
proxies. Participants came from 12 universities across 
provinces varying in economic development: Eastern region 
(52.2%, higher GDP per capita M=¥95,000), Central (28.9%, 
moderate GDP M=¥61,000), Western (18.9%, lower GDP 
M=¥52,000). We used provincial GDP per capita and 
urbanization rate (from the National Bureau of Statistics 
2024) as inverse proxies for collectivism, as research shows 
negative correlations between economic development and 



Z. Qian et al. /Future Technology                                                                                       February 2026| Volume 05 | Issue 01 | Pages 222-233 

231 

 

collectivistic values (r=-0.36 to -0.42). Exploratory multilevel 
modeling with faculty nested within universities nested 
within provinces showed that provincial GDP per capita 
marginally moderated the AI support-engagement 
relationship (γ=-0.024, SE=0.012, p=0.041), with slightly 
stronger effects in less economically developed (more 
collectivistic) regions. This suggests collectivistic cultures 
may amplify positive responses to organizational AI 
initiatives, as faculty perceive them as expressions of 
institutional care. However, this analysis is exploratory and 
limited by: (1) a lack of individual-level collectivism 
measurement, (2) a small number of provinces (k=8), and (3) 
ecological fallacy risks when inferring individual psychology 
from regional indicators. Future research should directly 
measure individual collectivistic values using validated scales 
(e.g., Triandis & Gelfand's Individualism-Collectivism Scale) 
to rigorously test cultural moderation. Given China's 
relatively homogeneous high collectivism compared to cross-
national variation (Hofstede score=20 vs. US=91), within-
country effects are modest. Fourth, self-report measures 
introduce common method bias despite our procedural and 
statistical controls; while CMV tests suggest bias is not severe, 
future research should incorporate objective measures such 
as actual AI system usage logs, teaching evaluations, and 
publication metrics to complement self-reports. Multi-source 
designs collecting supervisor ratings of engagement would 
strengthen causal claims. Fifth, our study captures a snapshot 
in AI evolution—as generative AI tools (ChatGPT, Claude) 
become ubiquitous post-2023, faculty-AI interaction patterns 
are rapidly changing, and as uses of AI shift toward large 
language models and generative AI, the nature of faculty-AI 
interaction may be fundamentally modified [35]. Our findings 
reflect early adoption phases; longitudinal studies tracking 
how relationships evolve as AI capabilities expand and faculty 
expertise deepens would reveal dynamic patterns. Several 
promising avenues emerge for future research. First, 
experimental or quasi-experimental designs could establish 
causality by randomly assigning faculty to digital literacy 
training interventions and measuring subsequent changes in 
engagement, providing more definitive evidence for the 
causal direction of relationships observed in this study.  

Second, experience sampling methods (ESM) could 
capture momentary fluctuations in engagement throughout 
the workday as faculty interact with AI systems, 
distinguishing sustained versus temporary effects and 
revealing the temporal dynamics of technology-engagement 
relationships [36].  

 

 

 

 

 

 

 

 

 

 

 

Methodologically sound approaches, such as ESM, can 
monitor micro-level changes in AI interaction activity and 
determine if benefits are short- or long-term. Third, 
qualitative studies using the critical incident technique could 
identify specific AI features or interaction moments that 
trigger flow states versus causing frustration, providing rich 
contextual understanding of how and why AI systems 
enhance or diminish engagement. Fourth, cross-cultural 
comparative research would establish boundary conditions 
and cultural moderators, testing whether the patterns 
observed in China's collectivistic context generalize to 
individualistic Western cultures or other educational 
systems. Fifth, as AI technology advances toward autonomous 
research and creative tasks traditionally considered uniquely 
human, studies should investigate how faculty roles 
transform and whether engagement patterns shift from task-
efficiency benefits to concerns about skill obsolescence or 
role displacement. Finally, research should align with Future 
Technology Journal's emphasis on AI adoption policy and 
technological innovation by examining institution-level 
implementation strategies, policy frameworks supporting 
ethical AI use, and organizational cultures fostering 
productive human-AI collaboration in knowledge work. 
These investigations would provide actionable insights for 
policymakers, university administrators, and educational 
technology developers seeking to optimize faculty-AI 
collaboration while promoting faculty well-being and 
institutional effectiveness in an increasingly technology-
mediated academic landscape. 

6. Conclusion  
This study examined the relationships between AI-

driven organizational support systems, digital literacy, and 
work engagement among 387 university faculty members in 
China. Grounded in Job Demands-Resources theory, 
Technology Acceptance Model, and self-efficacy theory, we 
hypothesized and found correlational evidence that AI-based 
organizational support is strongly associated with faculty 
work engagement, with digital literacy serving as a significant 
moderator. Our findings reveal several patterns. First, AI 
organizational support demonstrated a substantial positive 
association with overall work engagement (β=0.425, 
p<0.001), explaining an additional 18.6% variance beyond 
demographic controls. This strong relationship held 
consistently across all three engagement dimensions: vigor 
(β=0.398), dedication (β=0.412), and absorption (β=0.376), 
suggesting that effective AI systems can enhance faculty 
energy, enthusiasm, and immersion in academic work.  

 

 

 

 

 

 

 

 

 

 

 

(a) Vigor
Interaction: β = .185, p < .001

4.42

4.63
4.50

5.69

Low AI Support High AI Support

V
ig

o

r

(b) Dedication
Interaction: β = .152, p = .006

4.68

4.89
4.77

5.88

D
ed

ic
at

io
n

(c) Absorption
Interaction: β = .135, p = .022

4.70

4.874.79

5.75

A
b

so
rp

ti
o

n

Low Digital Literacy
High Digital Literacy

Low AI Support High AI Support Low AI Support High AI Support  

 
Figure 3. Interaction effects of AI-driven organizational support and digital literacy on the three dimensions of work engagement: (A) Vigor, 
(B) Dedication, (C) Absorption 

 



Z. Qian et al. /Future Technology                                                                                       February 2026| Volume 05 | Issue 01 | Pages 222-233 

232 

 

Second, digital literacy emerged as a critical boundary 
condition, significantly moderating the AI support-
engagement relationship (β=0.168, p=0.003, ΔR²=0.026). 
Faculty with higher digital literacy exhibited 2.35 times 
stronger associations between AI support and engagement 
compared to their lower-literacy counterparts, translating to 
meaningful practical differences. Third, the moderation effect 
was strongest for the vigor dimension (β=0.185), indicating 
that digital literacy particularly influences whether AI 
systems are experienced as energizing resources versus 
depleting demands. These correlational findings, while 
limited by cross-sectional design and convenience sampling, 
offer important theoretical and practical insights. 
Theoretically, the study extends JD-R theory to digital work 
environments by conceptualizing AI systems as technological 
job resources whose effectiveness depends critically on 
individual digital competencies. The complementarity 
perspective—wherein AI augments rather than replaces 
human capabilities—is particularly relevant for knowledge 
work where critical thinking and creativity remain uniquely 
human. Practically, if subsequent causal studies confirm these 
associations, the findings suggest universities should invest in 
comprehensive digital literacy training programs alongside 
AI system implementation. The threshold effect implied by 
moderation patterns indicates that a minimum digital 
competency is a prerequisite for AI systems to enhance rather 
than hinder engagement. However, important limitations 
constrain interpretations. The cross-sectional design 
prevents causal conclusions; reverse causation or third-
variable confounding cannot be ruled out. Convenience 
sampling may over-represent technologically comfortable 
faculty, potentially inflating effect sizes. The Chinese cultural 
context—characterized by collectivism and top-down 
technology adoption—may not generalize to other cultural 
settings. Self-report data, despite common method bias 
controls, cannot replace objective behavioral measures. 
Moreover, the rapid evolution of AI technology means 
findings capture early adoption phases that may not reflect 
long-term patterns as both AI capabilities and faculty 
expertise mature. Future research priorities include: 
longitudinal and experimental designs to establish causality, 
cross-cultural comparisons to identify boundary conditions, 
experience sampling to capture momentary fluctuations in 
engagement, and qualitative investigations of specific AI 
interaction moments that enhance or diminish engagement. 
As AI systems evolve from task automation toward creative 
and analytical support, research must track how faculty roles 
transform and whether engagement patterns shift 
accordingly. Ultimately, understanding and optimizing 
faculty-AI collaboration will be essential for promoting 
faculty well-being and institutional effectiveness in an 
increasingly technology-mediated academic landscape.  

Ethical issue 
The authors are aware of and comply with best practices in 
publication ethics, specifically regarding authorship 
(avoidance of guest authorship), dual submission, 
manipulation of figures, competing interests, and compliance 
with research ethics policies. The authors adhere to 
publication requirements that the submitted work is original 
and has not been published elsewhere. 

Data availability statement 
The manuscript contains all the data. However, more data will 

be available upon request from the authors. 

 

Conflict of interest 

The authors declare no potential conflict of interest. 

References 

[1] M. Sposato, "Artificial intelligence in educational 
leadership: a comprehensive taxonomy and future 
directions," International Journal of Educational 
Technology in Higher Education, vol. 22, no. 1, p. 20, 
2025, doi: https://doi.org/10.1186/s41239-025-
00517-1. 

[2] C. K. Y. Chan and W. Hu, "Students’ voices on 
generative AI: Perceptions, benefits, and challenges in 
higher education," International Journal of 
Educational Technology in Higher Education, vol. 20, 
no. 1, p. 43, 2023, doi: 10.1186/s41239-023-00411-8. 

[3] D. R. Cotton, P. A. Cotton, and J. R. Shipway, "Chatting 
and cheating: Ensuring academic integrity in the era 
of ChatGPT," Innovations in education and teaching 
international, vol. 61, no. 2, pp. 228-239, 2024, doi: 
10.1080/14703297.2023.2190148. 

[4] A. Tlili et al., "What if the devil is my guardian angel: 
ChatGPT as a case study of using chatbots in 
education," Smart learning environments, vol. 10, no. 
1, p. 15, 2023, doi: 10.1186/s40561-023-00237-x. 

[5] S. Karkoulian, N. Sayegh, and N. Sayegh, "ChatGPT 
unveiled: Understanding perceptions of academic 
integrity in higher education-A qualitative approach," 
Journal of Academic Ethics, pp. 1-18, 2024, doi: 
10.1007/s10805-024-09543-6. 

[6] E. D. L. Evangelista, "Ensuring academic integrity in 
the age of ChatGPT: Rethinking exam design, 
assessment strategies, and ethical AI policies in 
higher education," Contemporary Educational 
Technology, vol. 17, no. 1, p. ep559, 2025, doi: 
10.30935/cedtech/15775. 

[7] W. B. Schaufeli, M. Salanova, V. González-Romá, and A. 
B. Bakker, "The measurement of engagement and 
burnout: A two sample confirmatory factor analytic 
approach," Journal of Happiness studies, vol. 3, no. 1, 
pp. 71-92, 2002, doi: 10.1023/A:1015630930326. 

[8] A. Ueno, C. Yu, L. Curtis, and C. Dennis, "Job demands-
resources theory extended: stress, loneliness, and 
care responsibilities impacting UK doctoral students’ 
and academics’ mental health," Studies in Higher 
Education, vol. 50, no. 4, pp. 808-823, 2025, doi: 
10.1080/03075079.2024.2357148. 

[9] A. B. Bakker and E. Demerouti, "Job demands–
resources theory: taking stock and looking forward," 
Journal of occupational health psychology, vol. 22, no. 
3, p. 273, 2017, doi: 10.1037/ocp0000056. 

[10] X. Liu and Y. Li, "Examining the double-edged sword 
effect of AI usage on work engagement: The 
moderating role of core task characteristics 
substitution," Behavioral Sciences, vol. 15, no. 2, p. 
206, 2025, doi: 10.3390/bs15020206. 

[11] Q. Meng, T.-J. Wu, W. Duan, and S. Li, "Effects of 
employee–Artificial Intelligence (AI) collaboration on 
counterproductive work behaviors (CWBs): Leader 
emotional support as a moderator," Behavioral 
Sciences, vol. 15, no. 5, p. 696, 2025, doi: 
10.3390/bs15050696. 

[12] W. Ng, "Can we teach digital natives digital literacy?," 
Computers & education, vol. 59, no. 3, pp. 1065-1078, 
2012, doi: 10.1016/j.compedu.2012.04.016. 

[13] Y. Zhao, A. M. P. Llorente, and M. C. S. Gómez, "Digital 
competence in higher education research: A 



Z. Qian et al. /Future Technology                                                                                       February 2026| Volume 05 | Issue 01 | Pages 222-233 

233 

 

systematic literature review," Computers & 
education, vol. 168, p. 104212, 2021. 

[14] F. D. Davis, "Perceived usefulness, perceived ease of 
use, and user acceptance of information technology," 
MIS quarterly, pp. 319-340, 1989, doi: 
10.2307/249008. 

[15] A. Bandura, Self-efficacy: The exercise of control. 
Freeman, 1997. Publisher: Worth Publishers 
ISBN:9780716728504, 0716728508 

[16] Y. Jin, L. Yan, V. Echeverria, D. Gašević, and R. 
Martinez-Maldonado, "Generative AI in higher 
education: A global perspective of institutional 
adoption policies and guidelines," Computers and 
Education: Artificial Intelligence, vol. 8, p. 100348, 
2025, doi: 10.1016/j.caeai.2024.100348. 

[17]  J. Nevárez Montes and J. Elizondo-Garcia, "Faculty 
acceptance and use of generative artificial 
intelligence in their practice," in Frontiers in 
Education, 2025, vol. 10: Frontiers Media SA, p. 
1427450, doi: 10.3389/feduc.2025.1427450.  

[18] W. B. Schaufeli, A. B. Bakker, and M. Salanova, "The 
measurement of work engagement with a short 
questionnaire: A cross-national study," Educational 
and psychological measurement, vol. 66, no. 4, pp. 
701-716, 2006. 

[19] S. E. Hobfoll, "The influence of culture, community, 
and the nested‐self in the stress process: Advancing 
conservation of resources theory," Applied 
psychology, vol. 50, no. 3, pp. 337-421, 2001. 

[20] G. Falloon, "From digital literacy to digital 
competence: the teacher digital competency (TDC) 
framework," Educational technology research and 
development, vol. 68, no. 5, pp. 2449-2472, 2020, doi: 
10.1007/s11423-020-09767-4. 

[21] L. S. Aiken, Multiple regression: Testing and 
interpreting interactions. sage, 1991. 

[22] Y. Lin and Z. Yu, "Extending Technology Acceptance 
Model to higher-education students’ use of digital 
academic reading tools on computers," International 
Journal of Educational Technology in Higher 
Education, vol. 20, no. 1, p. 34, 2023. 

[23] V. Venkatesh and H. Bala, "Technology acceptance 
model 3 and a research agenda on interventions," 
Decision sciences, vol. 39, no. 2, pp. 273-315, 2008. 

[24] P. M. Podsakoff, S. B. MacKenzie, and N. P. Podsakoff, 
"Sources of method bias in social science research 
and recommendations on how to control it," Annual 
review of psychology, vol. 63, no. 1, pp. 539-569, 
2012, doi: 10.1146/annurev-psych-120710-100452. 

[25] Z. Cai, X. Fan, and J. Du, "Gender and attitudes toward 
technology use: A meta-analysis," Computers & 
Education, vol. 105, pp. 1-13, 2017, doi: 
10.1016/j.compedu.2016.11.003. 

[26] M. G. Morris and V. Venkatesh, "Age differences in 
technology adoption decisions: Implications for a 
changing work force," Personnel psychology, vol. 53, 
no. 2, pp. 375-403, 2000, doi: 10.1111/j.1744-
6570.2000.tb00206.x. 

[27] R. W. Brislin, "Back-translation for cross-cultural 
research," Journal of cross-cultural psychology, vol. 1, 
no. 3, pp. 185-216, 1970, doi: 
10.1177/135910457000100301. 

[28] A. F. Hayes, Introduction to mediation, moderation, 
and conditional process analysis: A regression-based 
approach. Guilford publications, 2017. 

[29] C. Fornell and D. F. Larcker, "Evaluating structural 
equation models with unobservable variables and 
measurement error," Journal of marketing research, 
vol. 18, no. 1, pp. 39-50, 1981, doi: 10.2307/3151312. 

[30] J. Cohen, Statistical power analysis for the behavioral 
sciences. routledge, 2013. 
https://doi.org/10.4324/9780203771587 

[31] F. D. Davis and V. Venkatesh, "Toward preprototype 
user acceptance testing of new information systems: 
implications for software project management," IEEE 
Transactions on Engineering management, vol. 51, 
no. 1, pp. 31-46, 2004, doi: 
10.1109/TEM.2003.822468. 

[32] V. Venkatesh, M. G. Morris, G. B. Davis, and F. D. Davis, 
"User acceptance of information technology: Toward 
a unified view," MIS quarterly, pp. 425-478, 2003, doi: 
10.2307/30036540. 

[33] J. F. Hair, A primer on partial least squares structural 
equation modeling (PLS-SEM). Sage, 2014. 
ISBN:9781452217444, 1452217440 

[34] E. Brynjolfsson, The second machine age: Work, 
progress, and prosperity in a time of brilliant 
technologies. WW Norton Company, 2014. 

[35] E. Kasneci et al., "ChatGPT for good? On opportunities 
and challenges of large language models for 
education," Learning and individual differences, vol. 
103, p. 102274, 2023, doi: 
10.1016/j.lindif.2023.102274. 

[36] A. S. Gabriel et al., "Experience sampling methods: A 
discussion of critical trends and considerations for 
scholarly advancement," Organizational Research 
Methods, vol. 22, no. 4, pp. 969-1006, 2019, doi: 
10.1177/1094428118802626. 

 

 

  

This article is an open-access article distributed under the 

terms and conditions of the Creative Commons Attribution 

(CC BY) license 

(https://creativecommons.org/licenses/by/4.0/). 

https://creativecommons.org/licenses/by/4.0/

