AI-Assistive Technology Adoption and Mental Health Disorders in Visually Impaired University Students

Abstract

The rapid integration of Artificial Intelligence Assistive Technology (AIAT) into higher education has generated new avenues for visually impaired university students, primarily in enhancing accessibility, self-autonomy, and academic performance. This study examined associations between AIAT-related perceptions and mental-health indicators (depression, anxiety, and stress) among visually impaired higher education students in the Kingdom of Saudi Arabia (KSA). A quantitative research approach was employed, using a self-administrated questionnaire targeting 390 visually impaired students in KSA universities. Partial least squares structural equation modelling (PLS-SEM) was employed as the main data analysis technique. The findings emphasised two important issues. First, performance expectancy (PE) of AIAT adoption, Effort expectancy (EE), and social influence (SI) are forceful psychological facilitators that can buffer against the feeling of depression and anxiety in visually impaired university students. Second, minimising the feeling of stress requires more than the existence of good infrastructure or social support; it necessitates systemic and ongoing interventions, comprising proactive university support, an accessible learning context, and personalised training programmes. These insights highlight the need for implementing inclusive support systems that combine technological, psychological, and university dimensions to promote the advantages of AIAT adoption for visually impaired students.

1. Introduction

Assistive technology (AT) refers to health, educational, and service-related technologies designed to enhance the functional capabilities of students with special needs, especially those with visual impairments [1]. Assistive technology tools improve academic performance, facilitate classroom integration, and promote participation in the labour market and social life, while reducing the need for educational interventions [2]. The WHO (2022) [3] reports that over one billion people currently require one or more assistive products, a figure projected to exceed two billion by 2050. For individuals with disabilities, AT is often indispensable, mitigating the challenges of temporary or permanent impairments [4]. Globally, over 21,000 assistive technology (AT) devices are available, with visual impairment solutions representing the most prevalent category, significantly outnumbering those for other disabilities [5]. These solutions are specifically engineered to enhance functional capabilities and promote independence among individuals with visual impairments [6].

In higher education, students with visual impairments frequently encounter mental health challenges during their academic journey, including heightened levels of depression, anxiety, and stress [7]. These psychological difficulties often emerge from the compounded pressures of navigating educational systems while managing disabilities, significantly hindering academic achievement and overall educational attainment [8,9]. Hence, this intersection of visual impairment and mental health concerns creates substantial barriers to learning, underscoring the critical need for targeted psychological interventions and accessible support systems within educational settings [10].

Despite diversity, recent studies address visually impaired students in higher education and mental health concerns, which create substantial barriers to learning, especially in Saudi Arabia. However, limited attention has examined how AI assistive technology (AIAT) usage correlates with mental health disorders in this population. Notably, there remains a critical gap in understanding how Unified Theory of Acceptance and Use of Technology (UTAUT) constructs, performance expectancy (PE), effort expectancy (EE), social influence (SI), and facilitating conditions (FCs), interact with mental health outcomes (depression, anxiety, and stress) among university students with visual impairments using AIAT. Thus, the study aims to answer the following question: RQ1: How does Performance Expectancy (PE) influence levels of depression, stress, and anxiety among visually impaired students who use AI Assistive Technology (AIAT)? RQ2: To what extent does Effort Expectancy (EE) impact depression, stress and anxiety in this population? RQ3: What is the impact of Social Influence (SI) on psychological well-being (depression, anxiety, stress) among AIAT users? and RQ4: How do Facilitating Conditions (FCs) affect depression, anxiety and stress among visually impaired students utilising AIAT? The structure of the paper will be as follows. Section 2 reviews the literature, the research methodology is presented in Section 3, followed by the study findings in Section 4. The discussion of the findings is in Section 5, and Section 6 concludes the study. Finally, Section 7 describes the future research directions.

2. Literature Review

2.1. Visual Impairments and AI-Powered Assistive Technologies

Artificial intelligence (AI) has emerged as a transformative force across all educational settings, with particularly profound implications for special needs education [11]. By revolutionising digital learning, AI-powered assistive technologies (AIATs) deliver adaptive and interactive experiences tailored to students with special needs, especially those with visual impairments [12]. Moreover, these AI-driven advancements are not only forging new pathways toward inclusive education but also actively fostering greater participation in learning processes for students with disabilities [13]. Additionally, Setiawan’s study (2024) [14] demonstrates that AI-powered assistive technology (AIAT) comprises a comprehensive range of solutions specifically designed to help individuals with disabilities, including those with spatial visual impairments, overcome daily challenges. Supporting this finding Takshara and Bhuvaneswari (2025) [15] study revealed that these tools significantly enhance autonomy in individuals with visual impairment, thereby improving access to educational resources and promoting full societal inclusion. Furthermore Vistorte, et al. (2024) [11] highlighted how the integration of AI, through advanced algorithms and big data analytics, has substantially increased the efficiency and adaptability of these technologies, making them more responsive than previous iterations. Furthermore, Thoits [16] noted that the practical benefits of the AIATs are evident in advanced voice recognition systems, which enable visually impaired students to operate devices entirely through voice commands, from basic text dictation to sophisticated application navigation. Yap et al. (2025) [17] highlighted how AI-driven writing, specifically AIATs, improves accessibility for visually impaired students with dyslexia or visual impairments by offering real-time word prediction and suggestions, streamlining the writing process, and reducing cognitive load. Moreover Rana et al. (2024) [12] note that AIATs not only enhance functional capabilities but also promote greater independence in daily activities.

Additionally, the AIATs, such as Siri, Pocket Vision, and Seeing AI, offer comprehensive features that extend beyond basic functionalities [18]. These include image-to-speech conversion, image enhancement for visually impaired students with low vision, facial and currency recognition, among others. Furthermore, by leveraging machine learning, these applications continually adapt to address the unique and evolving needs of visually impaired students [19].

Furthermore, the AIATs show especially promise for transforming higher education by empowering visually impaired students to pursue fully independent learning experiences [20,21]. Additionally, these solutions not only provide seamless access to information and essential academic services but also promote dignity and social inclusion within learning communities [19]. As well, by systematically addressing both technological and social barriers, they create equitable conditions for meaningful participation in higher education [22]. Contextually, existing research on AI-powered assistive technologies (AIATs) has primarily focused on classifying different tool types rather than critically analysing their impact on educational outcomes across diverse learning contexts. AI applications such as Siri, Pocket Vision, and Seeing AI are frequently praised for facilitating independent learning among visually impaired students; however, few studies have systematically examined their pedagogical integration, differential effectiveness across various disabilities, or measurable contributions to academic achievement. In the Saudi Arabian context, recent studies indicate growing interest in developing inclusive AI infrastructure within higher education. Nevertheless, research specifically addressing AI adoption for visually impaired students remains limited, particularly in terms of comprehensive theoretical frameworks and rigorous empirical methodologies to evaluate its impact on academic success in university settings. Given the cross-sectional design of this study, the identified relationships are interpreted as associations rather than causal effects. Accordingly, technology perceptions are conceptualised as correlations that may influence outcomes through support- and control-related mechanisms.

2.2. Mental Health in Visually Impaired Students and Research Hypotheses

Self-reported experiences of visual impairment are frequently associated with mental health challenges, particularly depression, anxiety, and stress [23]. Likewise, several pieces of evidence similarly indicated that learners with visual impairments often develop negative self-perceptions, potentially exacerbating these psychological difficulties [24]. As well, while visual impairment and mental health disorders are closely interrelated, they are mediated by distinct risk factors [25]. Hence, distress resulting from vision loss may increase susceptibility to mental health conditions among those with visual impairment [23]. Blindness remains one of the most feared health conditions globally [26]. Recently, studies have consistently highlighted the psychological impact of visual impairment among learners [27,28,29]. Elshaer et al. [5] found that a significant proportion of students express profound concerns about potential vision loss. Supporting this, Yilma et al. [30] empirically demonstrated that individuals with visual impairments face elevated risks of mental health challenges, particularly depression, anxiety, and stress. As well, the severity of this issue is further underscored by Demmin and Silverstein [31], who reported that approximately 33% of visually impaired learners experience at least one of these conditions. Additionally, disparity is particularly evident in higher education, where Wu (2025) [32] found that 10.7% of students with visual impairments reported significant mental health problems compared to 6.8% of their sighted peers. Similarly, De Schipper et al. (2017) [32] found that 35% of their visually impaired study participants exhibited significant symptoms of depression, anxiety, or stress.

Additionally, the integration of artificial intelligence (AI) into mental healthcare represents a paradigm shift in service delivery [33]. Likewise, leveraging machine learning algorithms to analyse large-scale datasets and detecting subtle psychopathological patterns, AI is transforming both diagnostic processes and therapeutic interventions [34,35]. Numerous articles have examined various models about technology acceptance theories, the “Theory of Planned Behaviour (TPB) [36], the “Value-Based Adoption Model” (VAM) [37], the “Technology Acceptance Model” (TAM) [38], and the “Unified Theory of Acceptance and Use of Technology” (UTAUT) [39]. However, UTAUT has A widespread acknowledgement for its efficiency in analysing people actions toward emerging technologies through different domains [39,40]. In an educational environment, the UTAUT framework, either in its original form or with theoretical expansions, has been implemented in several studies to understand how people interact with AI-driven technology [41,42].

Nevertheless, limited research has explored how UTAUT constructs may relate to psychological well-being, especially among visually impaired students using AI-powered technologies. As a result, this study examines psychological outcomes associated with AIAT usage, specifically depression, anxiety, and stress, through the lens of UTAUT [43], focusing on four constructs: performance expectancy (PE), effort expectancy (EE), social influence (SI), and facilitating conditions (FCs).

Additionally, UTAUT was prioritised over TAM because TAM’s two-factor structure (perceived usefulness; perceived ease of use) omits social influence and facilitating conditions—determinants central to assistive technology in university contexts [44]; and we did not adopt UTAUT2 because its added consumer-focused determinants (hedonic motivation, price value, habit) are less applicable to institution-provisioned, medically/educationally necessary AIAT in our cross-sectional design [44].

Performance Expectancy (PE) is a key dimension of the UTAUT, assessing people’s belief that using specific technology will lead to significant improvements in their daily performance and outcomes. In the context of AIAT adoption among visually impaired students, PE explicitly related to the level to which students perceive that using AI tools such as chatbots, or AI-driven self-help programmes, can efficiently decrease their mental health, disorder symptoms (i.e., depression, stress, and anxiety) [43]. When students believe that an AI mental health-related tool can make a significant difference in their overall well-being, their incentive to adopt and continue with its usage grows. As evidence, Salem et al. (2011) [44] confirmed that perceived efficiency of AIAT adoption is a strong predictor of faithfulness to mental health interventions. Similarly, AI-based interventions highlight that people’s perceptions of the potential effectiveness of these AI tools can influence their engagement levels [45] and decrease the symptoms of mental health disorders. Consequently, the following hypotheses are proposed:

H1. 

Performance Expectancy (PE) significantly correlates with depression (as a mental health dimension).

H2. 

Performance Expectancy (PE) significantly correlates with anxiety (as a mental health dimension).

H3. 

Performance Expectancy (PE) significantly correlates with stress.

Effort Expectancy (EE) is another main factor of the UTAUT framework. It describes the level to which people perceive that utilising specific technology will be easy to use [43]. In the context of AIAT adoption among visually impaired students, EE is precisely related to students’ perceptions of how easy it is to involve these AI tools, such as ChatGPT v.4.1, mobile apps, aimed at enhancing overall well-being and mitigating the symptoms of mental health disorders (i.e., depression, anxiety, and stress). When students perceive AIAT as user-friendly, they are more likely to incorporate these AI tools into their daily routines. The ease of use of AIAT can significantly influence their readiness to try AI technology, both in a general sense and to continue using it over a longer time [46]. Al-Saedi et al. [47] confirmed that when mental health AIATs are recognised as simple and easy to use, users report a higher level of satisfaction and are more likely to continue using these AI tools, which assist in managing mental health symptoms such as depression, stress, and anxiety. Therefore, the following hypotheses are suggested:

H4. 

Effort Expectancy (EE) significantly correlates with depression.

H5. 

Expectancy (EE) significantly correlates with anxiety.

H6. 

Effort Expectancy (EE) significantly correlates with stress.

Social Influence (SI) is the third dimension of UTAUT theory. It describes the level to which people feel that peers, family members, or social networks trust them to utilise specific AI technology [43]. In the context of AIAT aimed at mitigating mental health symptoms, Sl reflected the social pressures from those significant parts that can shape and reshape students’ decision to participate in utilising such interventions.

Social influence may significantly influence students’ adoption of practice. The peers’ positive feelings or endorsements of AIAT adoption can improve a student’s incentive to try and continue utilising the AI technology, particularly when mental health stigma or disorders are present [48]. Furthermore, support from peers or mental health professionals who suggest AIAT as part of a successful academic environment can foster students’ trust in these AI tools and promote greater participation [49]. Venkatesh et al. [50] found that peer impacts and social norms can play a significant role in students’ readiness to use digital AIAT. In sum, SI can serve as a social cue that may normalise and validate the use of AIAT, making disabled students more open to its adoption. This is mainly relevant where social support and validation can minimise feelings of stigma and decrease mental health disorder symptoms. Therefore, the following hypotheses are proposed:

H7. 

Social Influence (SI) significantly correlates with depression.

H8. 

Social Influence (SI) significantly correlates with anxiety.

H9. 

Social Influence (SI) significantly correlates with stress.

Facilitating Conditions (FCs) are the fourth component of the UTAUT theory. FCs describe the level to which people recognise that the technical infrastructure, and resources, are present to facilitate the actual use of AI technology [51]. FCs include elements such as a stable internet connection, user-friendly applications, accessible technical support, and training that can help students use AIAT effectively.

In the university environment, FCs could comprise access to AI devices (such as smartphones and PCs), a consistent internet connection, technical support, and informational backing on the efficient use of AIAT. The existence of adequate FCs minimises the barriers of AI adoption and continued usage, thus maximising the likelihood that students will employ these AI tools to progress their overall well-being [52]. Baumel et al. [53] argue that infrastructure and institutional support are significant in safeguarding the successful usage and continued utilisation of mental health AIAT in an educational context, particularly when targeting stress, anxiety, and anxiety symptom reduction. Additionally, FCs are particularly dominant for university students who may have some deficiencies in digital literacy regarding AI-based tools, as these elements may impact their trust and ease of use [43,54]. Ensuring acceptable training and technical aid might alleviate the recognised barriers and promote more actual usage of AIAT, thus improving its potential to mitigate symptoms of mental health disorders. Hence, the following hypotheses are posited:

H10. 

Facilitating Conditions (FCs) significantly correlate with depression.

H11. 

Facilitating Conditions (FCs) significantly correlate with anxiety.

H12. 

Facilitating Conditions (FCs) significantly correlate with stress.

Although the UTAUT framework was initially developed to explain technology adoption behaviour [1], its constructs also encompass cognitive--behavioural mechanisms that can influence mental health outcomes [2]. To strengthen this perspective, the present study integrates the Stress-Buffering Model [3] and Self-Determination Theory [4], which together suggest that technology-mediated experiences can foster psychological resilience.

Additionally, the current study hypothesises that the use of AIATs enhances perceived autonomy, competence, and social support [5,6]—key protective factors against depression and anxiety [7,8]. Moreover, the Technostress and Cognitive Load frameworks [9,10] indicate that intuitive, user-friendly technology systems, particularly AI systems, can alleviate cognitive strain and emotional stress, especially in educational settings [11].

Consequently, this study conceptualises technology attitudes as predictors of psychological well-being, while acknowledging associative—rather than causal—relationships between AIAT perceptions and mental health indicators among visually impaired students [12]. Furthermore, we adopted UTAUT as the organising framework for technology perceptions because it synthesises eight prominent adoption theories and shows comparatively higher explanatory power for behavioural intention and use in organisational settings [13].

Crucially for AIATs in higher education, UTAUT explicitly incorporates social influence (normative pressure from peers and instructors) and facilitating conditions (infrastructure, training, and accessibility services)—determinants central to students with visual impairments but absent from TAM’s two-factor structure (perceived usefulness and perceived ease of use) [14]. By contrast, TAM is a parsimonious model that omits these social and infrastructural enablers, making it less suitable for capturing university-level accessibility supports, assistive-technology training, and policy environments that shape adoption and continued use among disabled learners [15].

From a psychological standpoint, Social Support Theory posits that both the structure (availability) and the function (emotional, informational, instrumental) of support relate to better mental health via direct (main-effect) and buffering pathways [16,17]. In their seminal review, Cohen and Wills (1985) [3] showed that social support attenuates the adverse effects of stress—especially when support is responsive to the demands of the stressor (the “buffering hypothesis”). This lens aligns naturally with our UTAUT variables [16].

Specifically, SI reflects perceived interpersonal support (norms, encouragement), FCs map onto instrumental/tangible support (infrastructure, training, accessibility services), and PE and EE capture control-relevant appraisals (perceived effectiveness and ease) that bolster perceived competence and reduce cognitive-affective load. Accordingly, we expect these support-like determinants (SI, FCs) and control-relevant appraisals (PE, EE) to show direct associations with lower depression and anxiety, and to buffer the association between academic/technological demands and stress (i.e., stronger protective associations when demands are high).

3. The Research Methods

3.1. The Study Scale

All developed scale measures were derived from prior validated measures, safeguarding the accuracy and reliability of the study scale. The designed questionnaire is structured in four main parts. The first part explains the study’s aim to participants and provides them with a consent form to ensure their voluntary participation and approval of the ethical criteria. The second section was developed to obtain the necessary demographic information (i.e., student disability type, student age level, enrolled academic year, and gender). The third section was designed to measure the study’s independent latent variables (AIAT) usage. UTAUT framework, first coined by [43], was employed to operationalize AIAT usage This measure contains four main latent multidimensional factors: EE (measured by four items, sample item “Interacting with AIAT is clear and straightforward”); SI (measured by four questions, sample question “People whose opinions I value encourage me to use AIAT”); PE (measured by four variables, sample variable “ Using AIAT increases my chances of achieving academic goals”); FCs (operationalized by four items, sample variable “AIAT is compatible with the technologies I use”); and BI (operationalized by 3-item scale as reported by Ajzen and Fishbein’s [5], sample items, “I intend to keep using AIAT in the future.” All participants were directed to rate the agreement level on a five-point scale (1 = “Strongly disagree”, 5 = “Strongly agree”). The fourth section was developed to operationalise the study’s dependent latent construct (mental health disorder). To operationalise mental health disorders of disabled students, we adopted the Depression (Dpshn), Anxiety (Anzdy), and Stress (Strs) 21-Item Scale (named DASS-21), which evaluates the mental health disorder signs with 21 variables. The DASS-21 is a short version of the long DASS measure, first developed by Lovibond and Lovibond [32]. This shortened measure is currently validated by several scholars in different context (i.e., [6,33,34,35,36]). DASS-21 was designed to assess people’s negative feelings over the previous 7 days. The measure has three main dimensions, each of which has seven items. Participants were asked to assess their level of agreement with each question on a four-level Likert scale (0 “meaning no agreement” and 3 “signaling a higher level of agreement”) [32]. All scale items can be seen in Appendix A.

To ensure that the employed scale has face validity, the developed questionnaire was subjected to ten professors to emphasise the clarity, appropriateness and relevance of the questionnaire items. Additionally, a pilot test was implemented in ten higher education disabled students (with visual impairments) enrolled in King Faisal University (KFU). The outcomes from these two phases displayed that the scale items were clearly understood, and only very minimal language revisions were needed. These two steps ensured that the employed scale has shown a good face and content validity.

3.2. Study Population and Adequacy of Sample Size

Several types of disabilities are common among Saudi Arabian (SA) higher education students (i.e., hearing loss, mobility impairments, cognitive impairment, visual impairment, self-care limitations and communication challenges) (KSA census, 2022). The latest census of the SA population showed that around 1.8%, roughly 65,000 out of 36 million local residents, are suffering from a type of disability [5]. Moreover, the report showed that university students in higher education have a significant proportion (around 58%). The majority are full-time students in five governmental universities: King Abdulaziz University (KAU: 1569), King Saud University (KSU: 663), Taibah University (TU: 523), Umm Al-Qura University (UQU: 381), and King Faisal University (KFU: 330).

For the purpose of our paper, we targeted only visually impaired university students, excluding individuals with other types of disabilities. We collected the data through a convenience sampling method. To assess the adequacy of the sample size, a power analysis was conducted using the G*Power programme (version 3.1). We selected the F-test in the test family, “Linear multiple regression: Fixed model, R² deviation from zero” in the statistical test option. In the input parameter option, we selected an effect size of 0.15 (f² = 0.15), with four predictors, and a 0.95 statistical power, with a 0.05 significance level (α), the results suggested a total sample size of at least 124.

We recruited and trained 40 enumerators to help with the data collection process. The enumerators were trained on ethical procedures for collecting data from the targeted sample. These ethical procedures focus on how to obtain informed consent, ensure participants’ confidentiality, and address sensitivity concerns related to dealing with visually impaired university students. Several orientation sessions were designed to train the recruited enumerators, explaining the study’s objectives and addressing any concerns that visually impaired students might raise accurately. Out of the 950 questionnaires distributed, 390 were returned, resulting in a response rate of 41%.

As seen in Figure 1, the demographic statistics of the obtained sample (n = 390) showed a comparatively balanced representation among gender and universities, providing a solid foundation for interpreting the study results. Overall, females represented 55% (n = 214) of the respondents, while males accounted for 45% (n = 176), signifying a slight overrepresentation of female university students. This gender dispersal may mirror a wider enrolment pattern in Saudi Arabia higher education, where female involvement has been growing over the recent years. When inspected by the university, KFU donated the main proportion of respondents (26%), with females (18%) outnumbering males (9%). KSU followed with 22%, also showing a higher female participation rate. In contrast, TU had a superior proportion of male respondents (12%) compared to females (6%), which highlights some variability in gender composition across universities. Other universities, including KAU and UQU, donated similarly to the sample size (16–18%), with a somewhat higher proportion of male university students at KAU and an equally balanced distribution at UQU. Future research papers may investigate whether these demographic statistics can moderate the tested relationships, as cultural and institutional variations could impact on the experiences and results of students with disabilities. To contextualise the study results, respondents were inquired about their previous experience and adoption level of AIAT. The replies demonstrated that all respondents had some experience with AIAT applications, although the level of adoption varied. About 38% (n = 148) stated occasional use (e.g., limited to some specific academic duties such as translation or reading support), 42% (n = 164) revealed moderate usage (e.g., using AIAT two to four times per week for academic or personal support), whereas 20% (n = 78) stated an integrated and frequent usage (e.g., daily usage of AIAT for academic purposes, communication, and information access). These results demonstrated that while previous experience with the adoption of AIAT was common, the levels of usage and adoption were varied, offering a balanced representation of the study sample at diverse stages of technology usage.

3.3. Testing Common Method Variance (CMV)

The common method variance (CMV) issue can arise in social studies, as both the dependent and independent variables are typically completed by the same person [55], which can compromise the model’s validity and explanatory power [5]. Following Reio’s (2010) [55] suggestions, we implemented some precautious methods and statistical analysis to mitigate the influence of this issue (CMV). First, the questionnaire was designed to ensure balance between parts and to start with dependent variables [56], in order to minimise instructing influences, evade obvious patterns in answering the questions, and retain an adequate questionnaire length. To statistically test CMV, we conducted Harman’s single-factor test. The outcomes of this test indicated that a single combined factor can explain only 43% of the variance, suggesting that no single factor can explain the majority of the variance (>50%), and providing a strong indication that the CMV issue did not significantly influence the study results.

3.4. Ethical Approval

Given the delicate nature of our study, which targets university students with visual impairments in higher education, we formulated strict adherence to ethical concerns. Prior to the data collection process, we submitted a formal application to obtain approval from the Institutional Review Board (IRB) at Kin Faisal University (“Ethics Reference: KFU-REC-2025-APR-ETHICS3201,” approved on 6 September 2024). The approval safeguard ensures that our study aligns with our institutional standards and fulfils the ethical principles outlined in the Declaration of Helsinki.

To defend the rights of all contributing students, we implemented several safeguard procedures. The contribution was completely voluntary, with no pressure involved. Informed consent was obtained in writing from each contributor after providing detailed information about the study objectives. We also informed contributors of their right to withdraw from the study at any point in time without providing any specific reasons. Additionally, all collected data was anonymised to protect participants’ confidentiality and identities.

4. Data Analysis and Study Findings

Partial Least Squares Structural Equation Modelling (PLS-SEM) was employed as the primary analytical method for this study. PLS-SEM is a variance-based method that is mostly well-suited for predictive and exploratory research, presenting flexibility in analysis [57]. Unlike the other covariance-based SEM (CB-SEM) methods, which frequently require large sample sizes and fulfil the multivariate normality assumption, PLS-SEM has numerous advantages: it can manage smaller sample sizes effectively and does not assume data normality [58].

The analysis was run employing SmartPLS version 4 [59]. We conducted a bootstrapping process with 5000 resamples in the reflective mode to evaluate its stability and the significance of the estimation [60]. Following the procedures suggested by [59], the analysis was conducted in two consecutive phases. The first phase involved assessing the measurement properties of the outer model, including reliability and validity (see

Table 1. Fornell and Larcker metrics.

	1	2	3	4	5	6	7
1—Anxiety	0.868
2—Depression	0.851	0.828
3—EE	−0.415	−0.469	0.915
4—FCs	−0.315	−0.316	0.301	0.851
5—PE	−0.424	−0.466	0.798	0.326	0.858
6—SI	−0.323	−0.360	0.302	0.833	0.361	0.879
7—Stress	0.550	0.432	−0.289	−0.046	−0.352	−0.076	0.841

). The second phase mainly focused on testing the structural inner model to confirm or reject the proposed study hypotheses.

4.1. Measurement Inner Model Assessment

We carefully assessed the measurement outer model by reviewing four key psychometric properties: first, all variables displayed high standardised factor loading (SFL) values (i.e., >0.7); second, and third, the Cronbach’s alpha (α) values and composite reliability (C.R.) scores exceeded the value of 0.7; fourth, “Average variance extracted” (AVE) values are above the benchmark score of 0.5 [61]. These findings, when reviewed together, confirmed that the study measures have good internal consistency, adequate scale reliability, and high convergent validity. To assess the scale’s discriminant validity, we implemented three corresponding analyses. First, as recommended by Hair et al. [62], we compared the square root of each factor’s AVE (Presented in Figure 2) with all correlations between those factors; the square root of AVE should be higher (marked in bold in Table 1). This initial assessment showed adequate discriminant validity. Moreover, we also calculated the heterotrait–monotrait (HTMT) ratio [58] (

Table 2. “Heterotrait–monotrait” (HTMT) ratio—Matrix.

	1	2	3	4	5	6	7
1—Anxiety
2—Depression	0.704
3—EE	0.441	0.500
4—FCs	0.345	0.347	0.331
5—PE	0.459	0.503	0.784	0.374
6—SI	0.348	0.388	0.342	0.757	0.421
7—Stress	0.587	0.464	0.307	0.068	0.381	0.088

). This metric is considered a more advanced method than the previous Fornell–Larcker criterion; however, the HTMT metric has potential validity concerns when values exceed 0.9. As depicted in Figure 2, all HTMT values are below this cut-off point (0.9). Third, the cross-loading scores shown in Figure 3 revealed that all items are highly correlated and loaded to their programmed dimension with no cross-loading recognised. These constant results from the previous three methods presented strong evidence for adequate discriminant validity.

4.2. Structural Inner Model Findings

Before presenting and evaluating the study hypotheses, certain goodness-of-fit (GoF) criteria were reviewed, i.e., SRMR, “coefficient of determination” (R²), and “predictive relevance” (Q²). As Fornell and Larcker suggestions [57], the model’s GoF criteria should include an SRMR value lower than 0.08, an R² score of at least 0.10, and a Q² value higher than zero. Our model GoF met these conditions, demonstrating a good predictive and explanatory power. Specifically, the SRMR score is below the threshold value (SRMR = 0.071). Additionally, all endogenous latent variables have good predictive power: depression (R² = 0.284, Q² = 0.268); anxiety (R² = 0.231, Q² = 0.209); and stress (R² = 0.129, Q² = 0.106). Finally, the Variance Inflation Factor (VIF) scores were reviewed, as they should be below the value of 5, as suggested by [58]. As shown in Figure 2, all VIF scores are below 5, indicating that no multicollinearity concerns exist in our study model.

After ensuring the reliability of the outer model and the adequacy of inner model GoF, the hypothesis evaluation stage can be proceed, as depicted in

Table 3. Loadings and Factor Cross Loadings.

	Anxiety	Depression	EE	FCs	PE	SI	Stress
Anzdy_1	0.850	0.762	−0.395	−0.240	−0.384	−0.221	0.467
Anzdy_2	0.828	0.744	−0.378	−0.189	−0.360	−0.177	0.469
Anzdy_3	0.915	0.743	−0.370	−0.259	−0.395	−0.272	0.526
Anzdy_4	0.865	0.703	−0.372	−0.272	−0.366	−0.288	0.522
Anzdy_5	0.900	0.706	−0.360	−0.288	−0.380	−0.296	0.522
Anzdy_6	0.859	0.746	−0.314	−0.299	−0.333	−0.316	0.428
Anzdy_7	0.855	0.765	−0.332	−0.358	−0.357	−0.379	0.405
Dpshn_1	0.547	0.771	−0.304	−0.079	−0.325	−0.190	0.310
Dpshn_2	0.629	0.849	−0.427	−0.196	−0.426	−0.282	0.311
Dpshn_3	0.698	0.842	−0.402	−0.200	−0.426	−0.289	0.329
Dpshn_4	0.810	0.820	−0.394	−0.371	−0.387	−0.367	0.400
Dpshn_5	0.770	0.827	−0.381	−0.423	−0.392	−0.407	0.364
Dpshn_6	0.693	0.835	−0.345	−0.220	−0.315	−0.249	0.362
Dpshn_7	0.742	0.847	−0.442	−0.275	−0.402	−0.258	0.421
EE_1	−0.382	−0.440	0.924	0.259	0.703	0.254	−0.269
EE_2	−0.395	−0.439	0.916	0.282	0.724	0.266	−0.289
EE_3	−0.363	−0.433	0.918	0.295	0.754	0.306	−0.235
EE_4	−0.377	−0.405	0.903	0.266	0.742	0.282	−0.262
FC_1	−0.242	−0.330	0.307	0.893	0.327	0.776	0.009
FC_2	−0.269	−0.323	0.289	0.902	0.342	0.796	−0.045
FC_3	−0.317	−0.235	0.199	0.824	0.192	0.638	−0.052
FC_4	−0.245	−0.158	0.224	0.780	0.242	0.607	−0.081
PE_1	−0.366	−0.399	0.619	0.334	0.871	0.366	−0.297
PE_2	−0.426	−0.478	0.672	0.219	0.874	0.265	−0.340
PE_3	−0.319	−0.339	0.712	0.307	0.861	0.297	−0.293
PE_4	−0.328	−0.361	0.754	0.277	0.826	0.320	−0.268
SI_1	−0.210	−0.250	0.294	0.732	0.333	0.865	−0.074
SI_2	−0.290	−0.333	0.260	0.712	0.320	0.889	−0.074
SI_3	−0.331	−0.346	0.253	0.753	0.304	0.881	−0.056
strs_1	0.540	0.449	−0.279	−0.042	−0.344	−0.082	0.826
strs_2	0.548	0.429	−0.260	−0.061	−0.288	−0.079	0.826
strs_3	0.539	0.413	−0.234	−0.052	−0.264	−0.078	0.876
strs_4	0.545	0.433	−0.233	−0.054	−0.272	−0.085	0.871
strs_5	0.361	0.293	−0.226	−0.010	−0.310	−0.037	0.854
strs_6	0.362	0.277	−0.234	−0.008	−0.307	−0.025	0.863
strs_7	0.346	0.247	−0.226	−0.055	−0.265	−0.067	0.767

the bootstrapped results of the evaluated model exposed that performance expectancy of AIAT adoption can significant decrease the symptoms of depression (β = −0.179, t = 2.598, p < 0.01), anxiety (β = −0.197, t = 2.571, p < 0.05), and stress (β = −0.350, t = 4.939, p < 0.01) among visually impaired university students in SA, hence supporting H1, H2, and H3. Similarly, effort expectancy of AIAT adoption can significantly decrease the symptoms of depression (β = −0.261, t = 3.834, p < 0.001), and anxiety (β = −0.194, t = 2.578, p < 0.01), confirming H4 and H5. However, interestingly, effort expectancy of AIAT adoption failed to significantly decrease the symptoms of stress among visually impaired university students, rejecting H6.

However, interestingly, the PLS-SEM report also showed that social influence of AIAT adoption successfully decreases the symptoms of depression (β = −0.220, t = 3.498, p < 0.001), supporting H7, but failed to decrease the symptoms of anxiety (β = −0.107, t = 1.303, p = 0.193) and stress (β = −0.016, t = 0.193, p = 0.847), among visually impaired university students, rejecting H8, and H9. Similarly, facilitating conditions of AIAT adoption failed to decrease the symptoms of depression (β = −0.005, t = 0.073, p =0.942), anxiety (β = −0.103, t = 1.218, p = 0.223) and stress (β = −0.091, t = 1.067, p = 0.286) among visually impaired university students, rejecting H10, H11 and H12 (as seen in

Table 4. Hypothesis testing outcomes.

Tested Relationships	β	T statistics	p	Outcomes
(H1): Performance Expectancy -> Depression	−0.179	2.598	0.009	✓
(H2): Performance Expectancy -> Anxiety	−0.197	2.571	0.010	✓
(H3): Performance Expectancy -> Stress	−0.350	4.939	0.000	✓
(H4): Effort Expectancy -> Depression	−0.261	3.834	0.000	✓
(H5): Effort Expectancy -> Anxiety	−0.194	2.578	0.010	✓
(H6): Effort Expectancy -> Stress	−0.032	0.421	0.674	×
(H7): Social Influence -> Depression	−0.220	3.498	0.000	✓
(H8): Social Influence -> Anxiety	−0.107	1.303	0.193	×
(H9): Social Influence -> Stress	−0.016	0.193	0.847	×
(H10): Facilitating Conditions -> Depression	0.005	0.073	0.942	×
(H11): Facilitating Conditions -> Anxiety	−0.103	1.218	0.223	×
(H12): Facilitating Conditions -> Stress	0.091	1.067	0.286	×

Note: ✓: approved; ×: rejected.

).

5. Discussion

The findings of the tested model showed that the PE of AIAT significantly decreases the symptoms of mental health disorders (depression, anxiety, and stress) among visually impaired higher education university students in SA. These results emphasised the psychological advantages of AIAT adoption, chiefly when people perceive that such AI tools are helpful in improving academic performance, user-friendliness, and self-autonomy. These AIAT tools (i.e., screen readers, audio assistants, and adaptive education platforms) can decrease academic barriers, improve self-efficacy, and minimise psychological distress among visually impaired students.

The significant negative correlation between PE and depression is consistent with prior research demonstrating that AIAT can minimise the feelings of isolation. Empirical studies have argued that when people with disabilities perceive the sense of empowerment through AI technology, their feeling of control and self-worth are maximised, which in turn minimised the symptoms of depression [63]. Additionally, the reduction in anxiety feeling might be attributed to the reliability and predictability that AIAT offers in the academic environment. Previous empirical evidence has determined that when AIAT tools operate as expected and provide continuous support, people feel less uncertainty, which is a common indicator of anxiety [64]. Furthermore, for visually impaired university students, the expected fear of learning failure is a well-known anxiety trigger. AIAT that enhances the accessibility to academic resources may mitigate this fear. Interestingly, the strongest correlation was observed between PE and stress reduction (β = −0.350). This result may be explained by the impacts of AI devices, which minimise the mental effort required to accomplish academic tasks. According to the literature, minimising superfluous cognitive burden can decrease stress feelings [65,66]. For university students with visual impairments, the capability to accomplish academic duties more effectively via AI minimises time pressure and academic overload, which are two major indicators of stress in the learning environment.

The PLS-SEM results also indicated that EE of AIAT adoption can significantly reduce depression and anxiety symptoms among visually impaired higher education students. This aligns with the UTAUT theory, which argues that EE is a primary antecedent of technology usage [67]. When people feel that AIAT is easy to acquire and learn, they encounter fewer emotional and cognitive restrictions, which in turn can improve self-efficacy and self-autonomy—protective factors against anxiety and depression [10,67]. Previous evidence has confirmed that user-friendly AIAT can mitigate negative emotions by minimising AI-related frustration and operational anxiety [68,69]. PLS-SEM report also declared that SI can significantly decrease the feeling of depression. This result can be explained by understanding the conceptualisation of stress as operationalised in the DASS-21 scale, which describes continuous tension and irritability rather than contextual anxiety or a bad mood [70]. While AIAT ease of use might minimise the feeling of depression and anxiety tied among visually impaired university students in SA, stress may be more forcefully driven by other systemic elements such as study load, accessibility of university infrastructure, or insufficient university support. Previous evidence suggested that stress reduction feeling need not only valid systems but also university support, such as continuous training, technical help, and inclusive practice [71].

The PLS-SEM report also declared that SI can significantly decrease the feeling of depression. For visually impaired university students, support from colleagues, close family, or lecturers to use AIAT may promote a feeling of attachment to university, thus minimising the sense of isolation that is a common indicator of depression [63]. Social support from peers and family has long been established as a protective element against symptoms of depression among students with disabilities [72]. In contrast, SI cannot significantly decrease the feeling of anxiety and stress. This divergence might be due to the variance mechanisms in the SI process. While social support can alleviate the feeling of depression, which is commonly linked to loneliness, it may not be able to mitigate the feeling of anxiety and stress, which are frequently tied to learning pressures, study overload, and other barriers in the university setting [70]. Moreover, prior evidence argues that SI can maximise anxiety, as expectations from colleagues or family may generate pressure to perform well [43,51]. Additionally, the study findings revealed that FCs of AIAT adoption failed to significantly minimise the symptoms of depression, anxiety, and stress among visually impaired students in SA. While FCs can naturally predict actual adoption behaviour, their impact on overall psychological well-being seems less clear. The current findings suggest that the mere accessibility of university resources and support does not inevitably translate into minimised emotional distress. These results are consistent with prior evidence that FCs tend to have low effects compared to factors such as PE or EE, particularly when examining individual-level psychological consequences [51,54]. A reasonable explanation is that mental health consequences can be shaped by perceived control and serviceability rather than by institutional availability of support. Prior evidence has shown that accessibility infrastructure in university contexts is frequently under-utilised due to the lack of strong awareness, self-stigma, or inadequate training [9,73,74,75]. Therefore, even when FCs exist, visually impaired university students may still feel depression, stress and anxiety, if disability infrastructure is fragmented, inadequately applied, or fails to fulfil the person’s needs.

The outcomes of this research paper have numerous practical implications for university policymakers, administrators, and AI developers in Saudi Arabia (SA). First, university managers should prioritise improving PE and EE by capitalising on AIAT devices that are accessible, reliable, and tailored to fulfil the specific requirements of visually impaired university students. Second, SI appeared as a main element in minimising the feeling of depression, signifying the importance of generating inclusive peer and family networks that support and encourage the usage of AIAT. Finally, the non-significant impact of FCs on mental health disorders indicated that infrastructure alone is not enough. University managers must move beyond providing university resources and focus on cohesive support ecosystems that emphasise user-friendliness, study load management programmes, and psychosocial welfare.

Based on our results, enhancing AIAT functionality elevates performance expectancy (PE), reduces cognitive load, and strengthens effort expectancy (EE). Likewise, key design practices include streamlining workflows with consistent navigation and minimal steps, using progressive disclosure for complex tasks, and providing forgiving error recovery. Together, these measures lower extraneous cognitive load and stress, thereby improving perceived ease of use and reducing frustration.

Beyond usability, stress should also be addressed through demand-side measures implementing Universal Design for Learning (UDL), ensuring accessible LMS pages, providing lecture capture with transcripts/captions, and allowing flexible assessment timing—recognising that stress often reflects chronic academic demands that usability alone cannot offset.

Additionally, because facilitating conditions (FCs) showed no direct mental-health effect, prioritise integrated, low-friction mental-health pathways—opt-in referrals and periodic check-ins—within AIAT onboarding, coordinated with counselling and disability services to counter the under-utilisation of care among people with visual impairment.

Furthermore, to enhance the effectiveness of AIAT interventions on the mental health of visually impaired students, future system development can draw on feature-learning approaches that progress from simple to complex scenes, as demonstrated in From Simple to Complex Scenes: Learning Robust Feature Representations for Accurate Human Parsing [18]. This approach emphasises hierarchical representation learning, in which models first capture low-level perceptual details and then integrate higher-level contextual cues.

Additionally, translating this into AIAT design suggests building adaptive, multi-layered interaction modules that evolve with users’ proficiency and emotional state, enabling more resilient, context-aware, and personalised support systems.

Hence, integrating such progressive feature-learning into AIATs may therefore improve both engagement and psychological well-being by aligning cognitive challenge with emotional readiness and learning progression.

6. Conclusions

This paper tested the relationships between AIAT adoption and mental health outcomes (depression, anxiety, and stress) among visually impaired university students in SA. Employing the UTAUT framework and PLS-SEM data analysis technique, the results confirmed that PE, EE, and SI can play a significant and main role in minimising specific psychological symptoms, mainly anxiety and depression. However, the findings also demonstrated that some factors (PE and SI) were less efficient in mitigating stressful feelings, while FCs failed completely to impact any of the mental health disorders. These outcomes emphasised two important issues. First, perceptions of usefulness (PE), ease of use (EE), and social influence (SI) are forceful psychological facilitators that can buffer against the feeling of depression and anxiety in visually impaired university students. Second, minimising the feeling of stress requires more than the existence of good infrastructure or social support; it necessitates systemic and ongoing interventions, comprising proactive university support, an accessible learning environment, and personalised training programmes.

7. Limitations and Future Research Opportunities

Like other social science studies, the present paper has several limitations that should be acknowledged. First, because all measures were collected at a single time point, it was not possible to determine the directionality between technology perceptions and mental health indicators. Both causal directions remain theoretically plausible; disentangling these relationships would require longitudinal, cross-lagged, or experimental research designs.

Second, this study employed a cross-sectional design, which limits the ability to establish causal inferences between AIAT adoption and mental health disorders. Future research, using longitudinal or experimental approaches, would provide stronger evidence regarding the temporal and causal dynamics.

Third, the data were collected through self-reported measures, which may be influenced by social desirability bias or respondent subjectivity. The use of more objective or mixed-method assessments (e.g., behavioural tracking, physiological measures, or triangulation with instructor reports) could enhance the validity of future findings. Moreover, the study’s theoretical framework, centred on UTAUT perceptions, did not include explicit demand-related constructs central to stress mechanisms (e.g., workload, deadlines). Future models should incorporate such demand-side variables and explore mediated or moderate pathways (e.g., Effort Expectancy × demands). Additionally, future research papers could integrate advanced robustness mechanisms, such as multiple reversible robust watermarking, to improve AIAT system stability and ensure accurate emotion monitoring and mental health support for visually impaired students across diverse usage environments.

Fourth, the sample was drawn from five major Saudi universities, which may constrain the generalizability of the findings to other institutional or cultural contexts. Future studies could include a more diverse and representative population across different regions or educational systems to enhance external validity. Fifth, another limitation of the current study is that university students who had completely given up employing AI-assistive technologies were excluded from the sample. While this can safeguard that all participants had sufficient knowledge with technology to offer meaningful replies, it may have introduced bias in sample selection. Future research papers should broaden the scope of the sample to include students with discontinued technology usage, as their perspectives would offer a more comprehensive understanding of the complex relationship between AIAT adoption and mental health disorders. Furthermore, future research papers could explore some moderating or mediating impacts, such as gender or academic year, to explore how these elements can influence the relationships between UAAUT factors and mental health disorders. A further limitation is that contextual factors such as workload or academic pressures were not measured, and future research should integrate these variables to strengthen the explanatory power of the model.

Finally, adopting qualitative or mixed-method approaches—such as in-depth interviews or focus groups—could provide richer and more nuanced insights into the lived experiences of university students using AIATs. These methods may uncover contextual and emotional dimensions that quantitative approaches alone cannot fully capture.