Digital Adoption of Generative AI Tools: A Multi-Theory Model Linking Cognitive Load, User Perceptions, and System Attributes

Abstract

The rapid diffusion of “(GenAI)” Generative Artificial Intelligence systems has reshaped everyday activities, yet their adoption remains uneven and cognitively demanding for many users. Existing research has largely relied on conventional technology acceptance models, providing limited insight into cognitive burden and GenAI-specific system characteristics. To address this gap, this study develops an integrated framework combining the Technology Acceptance Model, Cognitive Load Theory, and the DeLone and McLean Information Systems Success Model to explain GenAI adoption among ordinary users. Survey data from 1001 active GenAI users were analyzed using partial least squares structural equation modeling (PLS-SEM). The results indicate that all core technology acceptance relationships are statistically significant (p < 0.001), while mental load negatively affects perceived usefulness and user attitudes. Moreover, GenAI system attributes—output quality, transparency, friction reduction, and system integration—significantly moderate key adoption pathways and strengthen the translation of behavioral intention into actual use. Predictive assessment indicates that the proposed model outperforms the baseline technology acceptance model, with stronger explanatory power and superior out-of-sample predictive performance (Q²predict > 0.35). The findings offer actionable insights for designing cognitively efficient, trustworthy, and sustainable GenAI systems.

1. Introduction

The use of generative artificial intelligence (GenAI) models like ChatGPT as digital assistants has rapidly evolved from an experimental technology to a mainstream one [1,2]. These tools have become essential in the daily lives of millions of people in writing, learning, solving problems, and making decisions [3]. Current research indicates that ChatGPT and other systems are being rapidly adopted in education, health care, and business. Yet, users still have concerns about trust, the reliability of outputs, cognitive overload, and ethical consequences. As GenAI becomes part of our daily lives rather than a tool used by technical professionals, determining how average users make decisions to adopt and persist in using these tools has become a research topic of interest. Beyond technical performance, the sustainability of generative artificial intelligence (GenAI) systems increasingly depends on their capacity to support long-term, responsible, and human-centered use. Sustainable digital transformation requires technologies that minimize unnecessary cognitive burden, promote transparency, and enable efficient human–AI interaction over time. When GenAI systems impose excessive mental load, lack interpretability, or generate friction in use, they risk digital fatigue, disengagement, and inefficient use of human and organizational resources, thereby undermining sustainable adoption. In this context, sustainability is not limited to environmental outcomes but extends to the durability, inclusiveness, and effectiveness of digital systems within everyday decision-making environments. This study, therefore, conceptualizes GenAI adoption as a digital sustainability challenge, emphasizing cognitive efficiency, system quality, and user trust as essential conditions for sustained, responsible technology use.

However, most current studies on adoption continue to regard GenAI as a conventional information system, despite the emerging interest [4]. Many researchers have used the Technology Acceptance Model (TAM) [5] to investigate the adoption of ChatGPT and other AI tools, with perceived usefulness as a leading predictor of behavioural intention and actual use, alongside perceived ease of use and attitude. The model developed by [6] based on TAM and introducing mental load as an aspect, demonstrates that cognitive effort is a determinant of acceptance. Nevertheless, these models fail to describe the uniqueness of GenAI. Unlike traditional software, GenAI produces dynamic, contextual, and sometimes ambiguous responses and does not necessarily explain how it arrived at those responses. This implies that usefulness and ease of use are not the only factors in user adoption, as system-level features such as output quality, transparency, interaction friction, and the degree of integration with the user’s digital space also play a role. Recent studies highlight the significance of system quality and transparency, as well as the cognitive effort required to deploy GenAI most productively. Yet, this aspect is rarely incorporated into a single behavioural model.

This creates a research gap. Most existing research is based on the classical TAM framework or its simplistic extensions and lacks a combination of psychological perceptions, cognitive load, and system quality. Cognitive load is a significant concern in AI-assisted work, which is recognized but not explicitly modeled in the studies of GenAI adoption. Similarly, even though the DeLone and McLean [7] IS Success Model focuses on the quality of systems and information, these parameters have not been systematically examined as GenAI-specific moderators, including quality, transparency, reduced friction, and system integration. Moreover, most adoption studies aim to elucidate relationships (rather than assess the extent to which extended models can forecast the actual use of GenAI).

To fill these gaps, this paper draws on three complementary theoretical viewpoints. TAM establishes a platform on which the perceived ease of use, perceived usefulness, and attitude explain behavioural intention and actual use. Cognitive Load Theory (CLT) [8] is a theory that can be used to explain how mental strain and the complexity of the information may compromise perceptions of usefulness, as well as diminish positive attitudes—a particularly important consideration in cases where users are brought to work with long, complex, or unclear AI-generated content. The DeLone and McLean IS Success Model adds a system-quality lens, which justifies the introduction of GenAI Quality, Transparency, Reduction in Friction, and System Integration as key characteristics that could impact user reviews and outcomes. Having a combination of these three approaches provides a more comprehensive view of GenAI adoption by connecting the psychological and cognitive dimensions with the tech determinants.

The novelty of the present study is that it extends the Bashir and Zhou frameworks by four GenAI-specific system moderators that act at different phases of the adoption process. The moderators that exist are mental load to perceived usefulness, perceived usefulness to attitude, attitude to behavioural intention, and behavioural intention to actual use. Mental load is also incorporated into the study as a central construct within a single system of TAM–CLT–D&M to capture the frequently ignored cognitive load of GenAIs’ interactions. In addition, the model is tested not only with explanatory SEM but also with predictive methods, including Q2 predict and CVPAT, which allow the study to evaluate whether the extended model predicts real user behaviour better than a baseline TAM.

The research questions to be answered in the study, based on these motivations, include the following:

RQ1: What is the joint effect of perceived ease of use, perceived usefulness, attitude, and mental load on behavioural intention and actual use of GenAI tools with everyday users?

RQ2: In what way do the psychological relationships put forward by TAM mediated by the GenAI system attributes, including quality, transparency, reduction in friction, and system integration?

RQ3: Does the longer TAM-CLT-D&M model have more explanatory and predictive strength in explaining and predicting GenAI adoption compared to a baseline TAM?

This study aims to develop and test a model that is more comprehensive in capturing the process of everyday users using GenAI by integrating psychologically perceived sources, cognitive expenditure, and system-based features. The current research is necessary, as GenAI is a relatively common occurrence in everyday life. At the same time, established theories provide important insights into technology use, but none of the dominant models independently explain how individuals simultaneously balance perceived value, cognitive burden, and system-level behavior in the context of Generative Artificial Intelligence. The Technology Acceptance Model emphasizes evaluative beliefs such as perceived usefulness and ease of use, but does not account for the cognitive strain imposed by complex generative systems. Cognitive Load Theory explains how mental effort affects information processing, yet it does not model technology adoption or continued use outcomes. Similarly, the DeLone and McLean Information Systems Success Model focuses on system quality and performance but largely overlooks users’ cognitive constraints during interaction. As a result, existing models offer fragmented explanations of GenAI adoption. Addressing this gap, the present study integrates these perspectives to conceptualize GenAI adoption as a cognitively constrained and system-dependent process, thereby explaining how perceived value, cognitive burden, and system behavior jointly shape adoption outcomes. Rather than incrementally extending TAM, this study advances theory by synthesizing complementary perspectives to explain GenAI adoption under conditions of cognitive intensity and system complexity.

2. Literature Review

2.1. Mental Load

Mental load represents a multidimensional cognitive condition that shapes how users evaluate and respond to Generative Artificial Intelligence (GenAI). During interaction with GenAI systems, users are exposed to varying degrees of task complexity, information intensity, time pressure, and emotional demands, all of which influence cognitive processing and evaluative judgments. Prior studies consistently show that increased task complexity elevates cognitive load, slows information processing, and induces feelings of frustration or overwhelm, thereby reducing perceived usefulness of GenAI systems [9,10]. While these studies highlight the negative consequences of complexity, they rarely consider whether GenAI design features—such as adaptive interfaces or context-aware prompts—can mitigate cognitive burden, leaving a notable gap in existing research.

Sustained mental effort can also lead to cognitive fatigue, which reduces users’ readiness to engage with GenAI and weakens their attitudes toward its use [11]. However, much of the literature does not clearly distinguish between short-term cognitive fatigue and longer-term technology avoidance, limiting the generalizability of current findings. Cognitive burden is further intensified by information overload, which occurs when users are confronted with excessive or poorly structured information. Such overload depletes cognitive resources, increases confusion, and undermines confidence in judgment, thereby diminishing the perceived usefulness and trustworthiness of GenAI systems [12]. Importantly, many studies implicitly attribute overload to users’ cognitive limitations while overlooking system-level responsibility, introducing a theoretical bias in the interpretation of cognitive strain.

Empirical evidence further shows that difficulty in interpreting complex or ambiguous AI outputs exacerbates cognitive overload, leading to poorer task performance and more negative attitudes toward GenAI [13,14,15,16]. At the same time, tolerance for cognitive demand may vary across user groups. Individuals with higher digital literacy or expertise may be better equipped to manage elevated mental load, suggesting that cognitive strain does not uniformly affect all users. Time pressure also contributes to mental exhaustion, particularly when users perceive insufficient time to meaningfully evaluate or apply AI-generated content, further weakening perceived usefulness [11,13]. Notably, most studies examine time pressure in controlled experimental contexts, which may not adequately reflect real-world multitasking environments where GenAI is commonly used.

Beyond cognitive strain, emotional load—manifested as anxiety, frustration, or stress—emerges when users encounter unforeseen or ambiguous AI responses. These emotional reactions negatively influence perceived usefulness and overall attitudes toward GenAI [17]. Emotional responses may also vary depending on personality traits, prior experience, and digital self-efficacy, factors that remain underexplored in the literature. Ambiguity in AI-generated outputs further intensifies cognitive and emotional demands, requiring greater interpretive effort and leading users to perceive GenAI as less supportive or more difficult to understand [18,19]. Consistent with recent findings, sustained cognitive load in AI-assisted environments is associated with reduced efficiency, burnout, and negative user evaluations [20,21].

Cognitive Load Theory distinguishes between intrinsic cognitive load, which stems from task complexity, and extraneous cognitive load, which arises from system design and information presentation. In GenAI contexts, intrinsic load reflects task-related demands, whereas extraneous load is shaped by interface design, output transparency, interaction friction, and information overload. Although mental load is measured as a single construct in the present study, it primarily captures extraneous cognitive load induced by GenAI interaction rather than task-inherent complexity. This conceptualization aligns with prior technology adoption research and enables examination of how system-level features can either alleviate or intensify cognitive strain during GenAI use.

2.2. Perceived Ease of Use and Perceived Usefulness in GenAI Contexts

Perceived Ease of Use (PEOU) is a central construct in the Technology Acceptance Model (TAM) and is widely recognized as a determinant of both Perceived Usefulness (PU) and attitude toward technology use. Prior research consistently demonstrates that users who perceive GenAI as easy to use are more likely to regard it as useful, supporting a positive relationship between PEOU and PU [22,23]. However, most empirical evidence is derived from studies of conventional digital systems rather than complex, generative technologies. This raises important questions about the applicability of ease-of-use assumptions to systems characterized by probabilistic, ambiguous, and context-dependent outputs.

Beyond its direct influence, PEOU also affects PU indirectly through experiential factors such as enjoyment and trust [24,25]. While this suggests that user experience mediates evaluative judgments, many studies treat enjoyment and trust as stable attributes rather than as dynamic responses shaped by GenAI output quality or transparency. This represents a critical oversight, as generative systems are prone to errors, hallucinations, and inconsistencies that may undermine trust even when interaction is technically easy. Consequently, prior research may overestimate the indirect effects of PEOU on PU by neglecting the variability inherent in GenAI interactions.

Similarly, existing studies often position PU as the primary mediator between PEOU and attitude, assuming that ease of use leads to positive attitudes primarily through enhanced usefulness perceptions [13,26,27]. While empirically supported, this logic presumes that usefulness is the dominant lens through which users form attitudes. In GenAI contexts, however, attitudes may also be shaped by cognitive load, perceived risk, ethical concerns, and uncertainty, even when systems are easy to use. This suggests that traditional mediation structures may insufficiently capture the complexity of attitude formation in generative environments.

Some studies further indicate a direct link between PEOU and attitude, whereby intuitive systems foster positive evaluations independent of perceived usefulness [23,28,29]. Yet these findings are often based on controlled or narrowly defined use cases that lack the open-ended, conversational characteristics of GenAI. As a result, their ecological validity remains limited, particularly in real-world contexts where users face cognitive overload, ambiguous responses, and multitasking demands.

Finally, PU continues to exert a strong positive influence on attitude toward GenAI [30,31,32,33]. However, most PU measures emphasize productivity and efficiency, whereas GenAI offers broader forms of value, including creativity support, ideation, decision augmentation, and emotional interaction. Existing PU scales often fail to capture these dimensions, constraining the explanatory scope of prior findings and underscoring a conceptual gap in GenAI adoption research.

2.3. Attitude Toward Using GenAI

Attitude toward using GenAI reflects users’ overall evaluative orientation and remains a key antecedent of behavioral intention in technology adoption research. Empirical evidence consistently shows that positive attitudes toward GenAI are associated with stronger intentions to use it. This pattern is evident across structured contexts, including supply chain settings, where task–technology fit and subjective norms influence intention through attitude [34,35], and programming education, where favorable attitudes toward ChatGPT predict continued usage intentions [36]. These findings align with the core TAM assumption that attitude functions as a central motivational mechanism.

However, the presumed universality of the attitude–intention relationship warrants caution. Much of the supporting evidence originates from highly structured environments such as design practices [27,36,37] and higher education settings [38,39,40,41], where users benefit from institutional support, guidance, and clearly defined tasks. In contrast, everyday GenAI users often operate without training or technical assistance, limiting the extent to which favorable attitudes translate into intention. This contextual mismatch reveals a methodological weakness in generalizing findings from controlled settings to broader populations.

Attitude formation is further influenced by multiple antecedents, including perceived usefulness, ease of use [42,43], social influence, and perceived knowledge [44], as well as psychological traits such as resilience, innovativeness, and self-efficacy [26,45,46]. Yet prior research predominantly emphasizes positive enablers while neglecting inhibiting conditions. Cognitive strain, algorithmic opacity, trust concerns, privacy risks, ethical uncertainty, and weak system integration can all disrupt the transformation of positive attitudes into behavioral intention. These boundary conditions remain underexplored, particularly among non-expert users, highlighting the need to examine moderating mechanisms that constrain or enable attitudinal effects.

2.4. Actual Use of GenAI

Actual use of GenAI by non-routine or non-expert users is increasingly evident across education, public services, and healthcare, yet adoption remains uneven, context-dependent, and often non-habitual. Such users tend to engage with GenAI intermittently, focusing on low-risk or exploratory tasks rather than embedding the technology into structured workflows. This pattern reflects both the potential and limitations of GenAI adoption.

In educational contexts, non-traditional university students use tools such as ChatGPT 5.2 primarily for clarification, time management, or writing support, often treating GenAI as an auxiliary aid rather than a primary solution [47,48,49]. Similar patterns are observed among teachers, who frequently rely on GenAI for basic tasks such as translation or information retrieval, while more advanced applications remain limited due to variations in digital competence and attitudes [50,51]. These findings suggest that GenAI use remains relatively superficial.

Comparable challenges arise in public services and healthcare, where GenAI is expected to enhance efficiency and reduce administrative burden, yet adoption remains fragmented and weakly institutionalized due to limited policy guidance [52,53]. Vague organizational direction further discourages consistent use, challenging assumptions that awareness alone leads to frequent adoption.

Non-routine users may also disengage from GenAI due to output quality concerns, ethical risks, emotional discomfort, and perceived erosion of human interaction, all of which contribute to non-use decisions [54,55]. Transparency and explainability emerge as critical trust-related factors, as users increasingly demand systems that can justify or clarify their outputs [56,57,58]. However, ambiguous or conflicting responses from GenAI systems often undermine consistent usage.

Cognitive load further constrains actual use, particularly when users struggle to interpret unforeseen outputs or manage excessive information [59,60]. Empirical evidence shows that overwhelming or unclear AI outputs negatively affect engagement and decision quality [55,58,61]. Moreover, problematic behavioral patterns such as impatience and overreliance have been observed, with some users abandoning GenAI following imperfect responses, while others rely on it excessively without verification [60]. These dynamics highlight the instability and unpredictability of real-world GenAI usage.

Despite a growing body of research, limited attention has been paid to the conditions under which non-routine users convert awareness and positive perceptions into sustained, real-life use. Existing studies rarely integrate cognitive strain, system quality, explainability, usefulness, ease of use, and emotional response into a unified explanation of actual GenAI use, particularly outside structured institutional contexts. This gap motivates the present study’s integrative approach.

The results of previous research indicate that the average user is aware of the potential usefulness of GenAI, but its use is uneven and compromised by cognitive, emotional, and trust-related factors. This gap explains why your study is necessary, to determine the overall effect of TAM factors and GenAI-specific system attributes on actual use in general, non-routine users.

Hence, the following hypotheses are proposed:

Direct Effects (TAM & CLT)

H1: 

Perceived Ease of Use (PEOU) has a positive and significant effect on Perceived Usefulness (PU) of Generative AI tools.

H2: 

Perceived Ease of Use (PEOU) has a positive and significant effect on users’ Attitude toward using Generative AI tools.

H3: 

Perceived Usefulness (PU) has a positive and significant effect on users’ Attitude toward using Generative AI tools.

H4: 

Users’ Attitude toward using Generative AI tools has a positive and significant effect on Behavioral Intention (BI) to use them.

H5: 

Behavioral Intention (BI) has a positive and significant effect on the Actual Use (AU) of Generative AI tools.

H6: 

Mental Load (ML) has a negative and significant effect on Perceived Usefulness (PU) of Generative AI tools.

H7: 

Mental Load (ML) has a negative and significant effect on users’ Attitude toward using Generative AI tools.

Moderating Effects (GenAI System Attributes)

H8a: 

GenAI Quality positively moderates the relationship between Mental Load (ML) and Perceived Usefulness (PU), such that higher GenAI Quality weakens the negative effect of Mental Load on Perceived Usefulness.

H8b: 

GenAI Transparency positively moderates the relationship between Perceived Usefulness (PU) and Attitude, such that higher transparency strengthens the positive effect of Perceived Usefulness on Attitude.

H8c: 

GenAI Friction Reduction positively moderates the relationship between Attitude and Behavioral Intention (BI), such that reduced interaction friction strengthens the translation of positive attitudes into usage intentions.

H8d: 

GenAI System Integration positively moderates the relationship between Behavioral Intention (BI) and Actual Use (AU), such that better system integration strengthens the conversion of intention into actual usage.

Model Comparison/Predictive Hypothesis

H9: 

The extended TAM–CLT–D&M model incorporating GenAI system attributes demonstrates significantly higher explanatory and predictive power (higher R² and Q² values and lower RMSE/MAE) compared to the baseline TAM.

2.5. Theoretical Foundation of the Study

To determine how and why general, non-routine users use Generative AI (GenAI), one needs a theoretical framework that not only reflects perceptions of AI usefulness and usability but also accounts for the cognitive load required and the quality of AI outputs that influence actual system use. Technology Acceptance Model (TAM) is the best place to start since it offers a simple, yet potent description of the role of perceived ease of use (PEOU) and perceived usefulness (PU) in influencing attitudes and behavioural intention, which are two variables which have been repeatedly found to be the main prediction of technology adoption in uncertain and changing settings such as GenAI. GenAI generates variable, conversational, and even ambiguous output, unlike traditional tools with well-known functionality. This renders subjective ratings much more significant than objective system attributes, and TAM is more indicative of such perceptions than any other model. Nonetheless, GenAI adds a mental aspect, which traditional TAM fails to discuss explicitly. GenAI interactions introduce variability, ambiguity, and unpredictability for users, which creates mental load in the form of task complexity, information overload, cognitive strain, and emotional pressure. This is why applying Cognitive Load Theory (CLT) strengthens TAM, as it explains how cognitive load affects PU and attitudes. When the mental load is high, GenAI can be less useful or more difficult to interact with, despite the ease or power of the system. CLT thus serves as a psychological intermediary, explaining why the same GenAI tool can be viewed as either positive or negative by different users, depending on their cognitive resources when using it.

To further improve explanatory depth, the DeLone and McLean IS Success Model (D&M) is added to TAM and CLT, which adds system-level qualities of the system, including system quality, information quality, and service quality, which are especially pertinent to GenAI. Due to their probabilistic nature, the output of GenAI can vary widely across users and situations. D&M enables the model to include the roles of GenAI quality, transparency, reduced friction, and system integration to moderate the relationship between perceptions (PEOU, PU, attitude) and outcomes (intention, actual use). This is particularly important for non-routine, general users, where adoption is critically dependent on the system acting stably and fitting well with day-to-day chores. Combining TAM + CLT + D&M thus generates a multi-level model: TAM explains why users opt to use GenAI, CLT explains when cognitive strain undermines perceptions, and D&M explains which system attributes may reinforce or cushion the psychological mechanisms.

This integrated model has better explanatory power for general users than the other theories. UTAUT, despite its strength, gives prominence to social influence, facilitating conditions, and performance expectancy, which are better applied in organizational contexts with mandatory use or semi-compulsory use. It fails to describe cognitive strain or system-level uncertainties that GenAI is susceptible to. Task-Technology Fit (TTF) focuses on the fit between tasks and technology but fails to account for users’ perceptions and emotional barriers as primary concerns in GenAI adoption. Expectation-Confirmation Theory (ECT) applies to post-adoption satisfaction procedures. It presupposes the persistence of usage patterns, which is not the case with GenAI, as it is an exploratory, trial-and-error technology for general users. Social Cognitive Theory (SCT) focuses on self-efficacy and observational learning and does not include constructs associated with system quality or cognitive effort. The integrated TAM + CLT + D&M paradigm, in turn, is versatile, perception-based, cognitively attentive, and quality-aware, and hence, much more capable of describing the uncertain, dynamic, and psychologically challenging process of GenAI adoption by non-routine users. Figure 1 illustrates the study’s conceptual framework.

Based on the synthesis of prior literature, this study proposes an original conceptual framework that integrates Technology Acceptance Model constructs with Cognitive Load Theory and the DeLone and McLean Information Systems Success Model to explain Generative Artificial Intelligence adoption. Figure 1 presents the proposed research framework guiding the empirical analysis, illustrating the relationships among technology acceptance beliefs, mental load, GenAI system attributes, and adoption outcomes.

Despite growing interest in the adoption of Generative Artificial Intelligence (GenAI), existing models remain theoretically fragmented and share several limitations. Most prior studies rely on conventional technology acceptance frameworks that emphasize perceived usefulness and ease of use while implicitly treating GenAI as a deterministic system. As a result, they provide limited insight into the cognitive burden associated with generative, probabilistic, and often ambiguous AI outputs. Cognitive load, emotional strain, and information overload are frequently under-theorized, while GenAI-specific system attributes—such as output quality, transparency, interaction friction, and system integration—are commonly examined descriptively or in isolation rather than theorized as boundary conditions shaping adoption pathways. Moreover, prior research largely prioritizes explanatory relationships, offering limited evidence on whether extended models improve the prediction of actual use or explain persistent intention–behavior gaps. Addressing these limitations, the present study advances an integrated TAM–CLT–D&M framework that conceptualizes GenAI adoption as a cognitively constrained and system-dependent process, thereby providing a more comprehensive and theoretically grounded explanation with enhanced explanatory and predictive power.

2.6. Conceptual Framework

The current research extends the theoretical framework initially formulated by Bashir and Zhou by investigating the impact of mental load, perceived ease of use, and perceived usefulness on users’ attitudes towards ChatGPT and how these attitudes affect behavioral intention and actual use of the system. Their model could explain the psychological mechanisms of technology adoption; however, they failed to account for the system-level features of modern Generative AI tools. GenAI generates dynamic, context-dependent, and occasionally unpredictable results; in other words, users’ reactions are influenced not only by cognitive factors but also by the quality, clarity, and integration of the system itself. This shortcoming inspired the formulation of a new conceptual framework that adds these new system characteristics.

Mental Load, Perceived Ease of Use (PEOU), and Perceived Usefulness (PU) are retained in the adapted model as the main antecedents of user attitude. Mental load is the mental effort required to use AI-generated content, and its effect can undermine users’ perceptions of usefulness and decrease positive attitudes. Perceived ease of use is a metric that describes the extent to which the interaction between users and GenAI is simple and understandable, and it influences usefulness and attitude. Perceived usefulness is one of the focal predictors of attitude, in line with the Technology Acceptance Model (TAM), thereby providing continuity with the previous model and addressing the GenAI-specific difficulties.

The model also maintains the developed TAM pathway, in which Attitude influences Behavioral Intention, and Behavioral Intention, in turn, predicts Actual Use. This flow is consistent with empirical evidence indicating that positive attitudes among users positively affect continued usage and the adoption of GenAI tools into their routines. The immediate predictor of actual usage is behavioral intention, suggesting that users may translate their intentions into actual use.

The primary improvement to this framework is the addition of four GenAI-specific moderators that affect a particular aspect of the adoption process. The quality of GenAI output can moderate the relationship between mental load and perceived usefulness, as high-quality output can negate the negative effect of cognitive load. GenAI Transparency moderates the relationship between perception of usefulness and attitude, as users are more likely to form strong positive attitudes when AI responses are provided with clear explanations or justifications. GenAI Friction Reduction balances this attitude toward the behavioral intention pathway by enabling users to act on their positive attitudes through less effortful interactions. Lastly, the connection between behavioral intention and actual use is mitigated by GenAI System Integration, such that intentions to use GenAI tools are more readily converted into habitual use when they are seamlessly incorporated into users’ digital spaces.

Combining the psychological determinants of TAM, the cognitive strain factors of the Cognitive Load Theory, and the system-quality variables of the DeLone and McLean IS Success Model, this theoretical framework provides a unified explanation of GenAI adoption among regular users. It not only describes the reasons users tend to find GenAI beneficial and user-friendly, but also how system-level features affect whether these perceptions translate into real-world use. This combined strategy is indicative of how the contemporary interaction between human beings and AI has become multifaceted, and of how GenAI usage is a multilevel process influenced not only by the human mind but also by the system’s abilities.

Although the Technology Acceptance Model provides a foundational explanation of users’ evaluative beliefs, its application to Generative Artificial Intelligence remains theoretically incomplete. GenAI systems impose substantial cognitive demands and rely on complex system-level characteristics that extend beyond perceived usefulness and ease of use. This study advances adoption theory by integrating Cognitive Load Theory to conceptualize mental load as a cognitively grounded constraint that shapes users’ evaluations, and by incorporating the DeLone and McLean Information Systems Success Model to theorize GenAI system attributes as contextual boundary conditions rather than direct determinants. In doing so, the study moves beyond linear TAM extensions and offers a cognitively and systemically informed explanation of GenAI adoption that better reflects the realities of human–AI interaction.

3. Research Methodology

3.1. Research Design

The present study builds on a prior study by [6] that examined the relationships among mental load, perceived ease of use, perceived usefulness, attitude, behavioral intention, and actual system use. Their model had been used to describe the psychological aspects of adopting the technology, but never the distinctive behavior of modern Generative AI tools, which generate conversational, adaptive, and even random responses. To fill in this gap, the current research modifies the initial model. It posits GenAI-related system features, such as quality, transparency, friction reduction, and system integration, as moderators in various phases of the adoption process. It adopted a quantitative, cross-sectional design, as it is effective for capturing users’ perceptions, cognitive states, and digital behaviors at a single point in time. The method is suitable for rapidly evolving systems, such as GenAI, whose user experiences can be effectively quantified through structured surveys. The quantitative study design is also acceptable, as the study will test relationships among variables, the effects of moderators, and the calculation of pathway strength in a complex adoption model.

The modified model maintains the relationships among the variables, including the effect of perceived ease of use on perceived usefulness, the effect of attitude on intention, and the relationship between intention and actual use. The extended model assumes that four moderators exist: GenAI Quality (moderating Mental Load—perceiving Usefulness), GenAI Transparency (moderating Perceived Usefulness—attitude), GenAI Friction Reduction (moderating Attitude—behavioral intention), and GenAI System Integration (moderating behavioral intention—actual use). These features are based on real-life concerns that current AI products should be trustworthy, interpretable, painless, and easily incorporated into users’ electronic lifestyles. The analysis of this expanded model is done using Partial Least Squares Structural Equation Modeling (PLS-SEM). The approach is appropriate due to its management of multidimensional models with numerous constructs and interactions, lack of normal distribution requirements, and focus on prediction, an imperative feature in testing emerging technologies such as GenAI. It is also possible to test the model for both explanatory power and predictive accuracy using PLS-SEM, which will aid in replicating the original model and assessing the newly included moderators.

3.2. Population, Sampling, and Data Collection

The group of individuals who used GenAI tools in their daily routine, rather than for professional purposes, was the target population. This sample was selected deliberately because their actions can reflect the broader population of GenAI adopters. The information was gathered through a web-based survey shared online. Online administration is also reasonable, as GenAI users are new media natives, and the approach is natural and accessible to the population. The participants were selected through a non-probability sampling method to represent a wide range of participants. The survey questionnaires, totaling 1500, were sent to potential respondents in Saudi Arabia. This was a premeditated higher sample size than the minimum requirement, since previous methodological recommendations suggest large sample sizes to achieve a robust SEM-PLS analysis. The study instrument included 23 measurement items; a rule of thumb of at least 20 responses per item was used, indicating a minimum sample requirement of 460 responses. A larger target distribution was thus adopted to account for incomplete responses, missing data, and potential outliers. In all, 1238 responses were obtained. Following data screening, which included steps such as eliminating incomplete questionnaires and those with missing values, among others, 1001 fully cleaned and valid responses were obtained and stored for analysis. This is a large enough sample that would be well above the recommended thresholds and would offer high statistical power and reliable findings. The final sample met the minimum requirements for PLS-SEM analysis, making it statistically robust. The study employed a non-probability, online survey approach targeting ordinary GenAI users in Saudi Arabia. While this sampling strategy does not aim to produce globally representative estimates, it is appropriate for examining user-level adoption mechanisms within a rapidly digitalizing national context characterized by high technology penetration and increasing everyday exposure to artificial intelligence tools. The final sample comprised users with diverse demographic profiles.

In terms of age, the majority of respondents were in early adulthood and mid-career, reflecting the most active segment of GenAI users. Most participants had at least an undergraduate education, and respondents reported varying levels of prior GenAI experience, ranging from occasional exploratory use to regular task-oriented use. This diversity enhances the internal validity of the findings while providing a transparent basis for assessing their contextual generalizability. In terms of age, most respondents were between 21 and 30 years (34.6%) and 31–40 years (38.2%), followed by those aged 41–50 years (18.5%), while respondents below 21 and above 50 years accounted for smaller proportions of the sample. Regarding educational attainment, 68.4% of participants held at least a bachelor’s degree, 19.7% possessed postgraduate qualifications, and the remainder had completed secondary or diploma-level education. Regarding industry affiliation, respondents were drawn from multiple sectors, including services, education, information technology, healthcare, retail, and public administration, indicating broad exposure to digital tools across occupational contexts. This diversity enhances the representativeness of the sample and supports the robustness of the study’s conclusions concerning everyday GenAI adoption.

3.3. Instrument Development

The survey tool comprised pre-existing scales based on the Technology Acceptance Model (TAM), Cognitive Load Theory (CLT), and the DeLone and McLean IS Success Model (Refer to Appendix A). The use of validated scales is also justified, as it enhances measurement reliability and aligns with previous studies. The constructs were assessed using various items on a 5-point Likert scale, and this choice was made because it was easy to administer and allowed for finer details of user perceptions. The tool included perceived usefulness, perceived ease of use, attitude, behavioral intention, actual use, mental load, and GenAI system characteristics such as quality, transparency, reduced friction, and system integration. The choice of these constructs was based on the concept of capturing psychological perceptions and system-level factors that affect GenAI adoption.

3.4. Data Analysis Technique

Data analysis was carried out using Partial Least Squares Structural Equation Modeling (PLS-SEM) in SmartPLS. This approach was selected due to the fact that the research is a multi-construct, moderating-relationships model. The use of PLS-SEM is supported by the fact that it is appropriately used in predictive-oriented studies, can handle non-normal data, and can be used to estimate and test measurement and structural models simultaneously. Reliability and validity were first evaluated using Cronbach’s alpha, composite reliability, AVE, and discriminant validity tests. These measures make every construct internally consistent and empirically different. Once the measurement quality was confirmed, the structural model was tested to examine the presence of direct effects, the negative effects of mental load, and the influence of system-related moderators.

3.5. Moderation and Structural Testing

The moderation effects were incorporated to test the impact of GenAI system characteristics on the magnitude of the psychological relationships hypothesized by TAM and CLT. For example, system transparency was examined as a moderator of the perceived usefulness-attitude relationship, whereas system integration was examined as a moderator of the behavioral intention-actual use relationship. The inclusion of moderators is reasonable, since the adoption of AI nowadays does not rely solely on user perceptions but also on system functionality in terms of transparency, reliability, reduced effort, and digitality. By experimenting with these effects, it is possible to develop a more realistic and comprehensive model of how users deploy GenAI in the real world.

3.6. Predictive Assessment

To assess the model’s predictive capability, the Cross-Validated Predictive Ability Test (CVPAT) has been conducted. It compares the prediction errors of a baseline TAM with a more extended model that adds GenAI-specific moderators. The reason for using CVPAT was that current AI adoption studies focus on both prediction and explanation. The fact that the extended model demonstrates improved predictive accuracy indicates that it not only explains but also predicts user behavior more accurately. The application of CVPAT, therefore, enhances the value and significance of the research.

3.7. Common Method Bias Assessment

Given that the data were collected using a self-reported, cross-sectional survey, potential common method bias (CMB) was assessed using both procedural and statistical approaches. Procedurally, respondent anonymity was ensured, and measurement items were carefully worded to reduce evaluation apprehension and ambiguity. Statistically, Harman’s single-factor test examined whether a single factor accounted for the majority of the variance in the measurement model. The results indicated that no single factor dominated the variance, suggesting that CMB is unlikely to be a serious concern. In addition, the full collinearity variance inflation factor (VIF) approach was employed, following established PLS-SEM guidelines. All full collinearity VIF values were below the recommended threshold of 3.3, providing further evidence that common method bias does not significantly inflate the estimated path relationships. Collectively, these results indicate that the observed structural relationships are robust and not substantially affected by common method effects.

3.8. Ethical Considerations

During the research, ethical principles were observed. Participation was voluntary, and informed consent was obtained, with respondents understanding the objectives of the study, confidentiality, and the anonymity of the data collection. No personally identifiable information was gathered. Information was properly stored and used only for academic purposes. Such measures are reasonable, as participants’ rights are protected and human-centered research standards are adhered to.

4. Results

This section presents the results of the PLS-SEM analysis. First, the measurement model is assessed for reliability and validity. In the next step, model evaluation was conducted using a structural model and hypothesis testing. In the final phase, the results of predictive validity and model comparison are reported.

4.1. Measurement Model

The results, shown in Figure 2a,b, indicate that the survey measures are reliable and consistent. All Cronbach’s alphas and reliabilities are well above the minimum required, indicating that the questions for each construct fit together strongly. The AVE values are also above 0.50, indicating that each construct captures sufficient variance in its indicators. When we look at the relationships between constructs (HTMT values), they are mostly below the recommended limit of 0.85. This means the constructs are different enough from each other and do not overlap too much. As expected, some constructs that are closely related, such as Attitude, Behavioural Intention, and Perceived Usefulness, show stronger connections, which makes sense within the TAM. On the other hand, constructs like Mental Load and the GenAI moderators are more distinct, showing low correlations, which is good because it confirms they add new information. This assesses whether the models are reliable, valid, and distinct, so they are ready for structural model testing, as shown in

Table 1. Construct Reliability and Validity.

Construct	α	rho_A	rho_C	AVE	1	2	3	4	5	6	7	8	9	10	11	12	13	14
1. Actual Use of System	0.895	0.897	0.923	0.705	—	0.837	0.861	0.284	0.642	0.286	0.504	0.290	0.617	0.808	0.146	0.109	0.150	0.109
2. Attitude Towards Using GenAI	0.910	0.912	0.933	0.736	0.837	—	0.798	0.096	0.668	0.128	0.584	0.275	0.644	0.833	0.042	0.126	0.148	0.006
3. Behavioural Intention	0.891	0.892	0.920	0.696	0.861	0.798	—	0.356	0.578	0.096	0.472	0.290	0.602	0.774	0.037	0.093	0.161	0.185
4. GenAI Friction Reduction	0.881	0.884	0.913	0.678	0.284	0.096	0.356	—	0.025	0.017	0.024	0.019	0.309	0.122	0.015	0.056	0.012	0.023
5. GenAI Quality	0.869	0.870	0.905	0.656	0.642	0.668	0.578	0.025	—	0.166	0.351	0.129	0.399	0.671	0.019	0.020	0.014	0.017
6. GenAI System Integration	0.877	0.878	0.910	0.670	0.286	0.128	0.096	0.017	0.166	—	0.197	0.025	0.017	0.124	0.029	0.031	0.032	0.025
7. GenAI Transparency	0.886	0.887	0.916	0.687	0.504	0.584	0.472	0.024	0.351	0.197	—	0.028	0.251	0.477	0.022	0.019	0.038	0.027
8. Mental Load	0.889	0.890	0.919	0.693	0.290	0.275	0.290	0.019	0.129	0.025	0.028	—	0.227	0.339	0.025	0.069	0.011	0.023
9. Perceived Ease of Use	0.816	0.818	0.879	0.645	0.617	0.644	0.602	0.309	0.399	0.017	0.251	0.227	—	0.652	0.055	0.018	0.022	0.017
10. Perceived Usefulness	0.923	0.923	0.942	0.765	0.808	0.833	0.774	0.122	0.671	0.124	0.477	0.339	0.652	—	0.034	0.028	0.147	0.020
11. GenAI System Integration × Behavioural Intention	—	—	—	—	0.146	0.042	0.037	0.015	0.019	0.029	0.022	0.025	0.055	0.034	—	0.126	0.031	0.028
12. GenAI Transparency × Perceived Usefulness	—	—	—	—	0.109	0.126	0.093	0.056	0.020	0.031	0.019	0.069	0.018	0.028	0.126	—	0.093	0.054
13. GenAI Quality × Mental Load	—	—	—	—	0.150	0.148	0.161	0.012	0.014	0.032	0.038	0.011	0.022	0.147	0.031	0.093	—	0.010
14. GenAI Friction Reduction × Attitude	—	—	—	—	0.109	0.006	0.185	0.023	0.017	0.025	0.027	0.023	0.017	0.020	0.028	0.054	0.010	—

.

The VIF results indicate that all measurement items (1.61–2.93) and the interaction term (1.00) are well within the acceptable range, confirming the absence of collinearity concerns. The model fit indices also meet recommended thresholds: SRMR and NFI values indicate a good fit, while d_ULS and Chi-square are slightly higher due to sample size and model complexity, but remain acceptable in PLS-SEM. Overall, the model demonstrates sound measurement quality and satisfactory structural fit as shown in

Table 2. Collinearity, Model Fit.

Indicator/Construct	VIF Range		Remarks
Measurement Items	1.61—2.93		All within acceptable threshold (<5; [61] → No collinearity concern)
Interaction Terms	1.00		Ideal (no multicollinearity)
Model Fit Indices	Saturated Model	Estimated Model	Acceptable Thresholds	Remarks
SRMR	0.026	0.076	<0.08	Both values acceptable
d_ULS	0.845	7.281	Closer to 0 is better	The estimated model is higher, but still acceptable
d_G	0.308	0.427	Closer to 0 is better	Good fit
Chi-square	4270.83	5235.366	Lower is better	Large sample leads to significance, common in PLS-SEM
NFI	0.943	0.931	>0.90	Acceptable

.

The results clearly show that integrating GenAI features greatly strengthened the model’s predictive power. With GenAI included, all constructs—actual use, attitude, behavioral intention, and perceived usefulness—achieved higher Q²predict values (indicating stronger predictive relevance) and lower RMSE/MAE values (indicating better accuracy). The R² values also improved, indicating that GenAI explains a greater proportion of the variance in user outcomes. Likewise, the CVPAT analysis confirmed that prediction errors were consistently lower with GenAI than without it, and these improvements were statistically significant (p < 0.000). In short, adding GenAI moderators improved the extended TAM’s accuracy, reliability, and generalizability in forecasting adoption behavior, as shown in

Table 3. CVPAT Results Comparing Models With and Without GenAI Integration.

Construct	Q2predict (No GenAI)	Q2predict (With GenAI)	RMSE (No GenAI)	RMSE (With GenAI)	MAE (No GenAI)	MAE (With GenAI)	R2 (No GenAI)	R2 (With GenAI)	Adj. R2 (No GenAI)	Adj. R2 (With GenAI)	Remarks
Actual Use of System	0.249	0.506	0.867	0.703	0.721	0.580	0.593	0.642	0.592	0.641	Significant improvement with GenAI
Attitude Towards Using GenAI	0.329	0.616	0.820	0.620	0.668	0.503	0.607	0.677	0.607	0.676	Stronger predictive power with GenAI
Behavioural Intention	0.273	0.524	0.853	0.691	0.710	0.567	0.518	0.615	0.517	0.615	Better accuracy and explained variance
Perceived Usefulness	0.361	0.560	0.800	0.664	0.656	0.536	0.362	0.562	0.362	0.561	Improved prediction with GenAI

.

4.2. Structural Model

The results in

Table 4. Hypothesis Testing.

Hypothesis	Path	β (O)	t-Value	p-Value	f2	Effect Size	Result
H1	PEOU → PU	0.379	26.012	0.000	0.283	Moderate	Supported
H2	PEOU → Attitude	0.389	30.833	0.000	0.075	Small	Supported
H3	PU → Attitude	0.530	35.251	0.000	0.464	Large	Supported
H4	Attitude → BI	0.697	66.317	0.000	1.253	Very Large	Supported
H5	BI → AU	0.749	83.093	0.000	1.553	Very Large	Supported
H6	ML → PU	−0.180	12.805	0.000	0.071	Small (Negative)	Supported
H7	ML → Attitude	−0.159	12.176	0.000	0.011	Negligible (Negative)	Supported
H8a	GenAI Quality × ML → PU	0.133	10.710	0.000	0.041	Small	Supported
H8b	GenAI Transparency × PU → Attitude	0.144	13.555	0.000	0.064	Small	Supported
H8c	GenAI Friction Reduction × Attitude → BI	0.179	14.821	0.000	0.084	Small	Supported
H8d	GenAI System Integration × BI → AU	0.115	11.027	0.000	0.038	Small	Supported
H9	Extended Model vs. Baseline (↑ R2 = 0.677, ↑ Q2 = 0.614, ↓ RMSE/MAE)	–	–	–	–	Improved Predictive Power	Supported

provide strong support for the extended TAM framework with GenAI moderators. Starting with the core TAM relationships, H1 (PEOU → PU) was supported (β = 0.379, t = 26.012, f² = 0.283, moderate), showing that when users found GenAI easy to use (simple, clear, and quick to learn), they also perceived it as useful in improving their everyday efficiency. Similarly, H2 (PEOU → Attitude) was significant (β = 0.389, t = 30.833, f² = 0.075, small), indicating that ease of interaction was associated with more favorable attitudes toward GenAI, though its effect was weaker than that of usefulness. The strongest driver of user attitudes was H3 (PU → Attitude) (β = 0.530, t = 35.251, f² = 0.464, large). This indicates that users’ positive attitudes toward GenAI were primarily shaped by whether they saw it as helpful in completing daily tasks, making decisions, and saving time. These attitudes had powerful downstream effects. H4 (Attitude → BI) and H5 (BI → AU) were both very strong, with path coefficients β = 0.697 and β = 0.749 and very large effect sizes (f² = 1.253 and 1.553, respectively). In practical terms, this means that when users felt good about GenAI, they intended to use it more often, and these intentions almost automatically translated into actual use—captured in behaviors such as regularly relying on GenAI, using it for different purposes, and integrating it into their daily routines. Mental load added an interesting contrast. H6 (ML → PU) was negative but significant (β = −0.180, t = 12.805, f² = 0.071, small), and H7 (ML → Attitude) was also negative (β = −0.159, t = 12.176, f² = 0.011, negligible). This suggests that when users experienced cognitive strain—such as stress, fatigue, or information overload—they were less likely to see GenAI as useful or to feel positive about using it. Unlike some prior studies where stress pushed people to seek digital help, here, mental strain acted as a barrier to adoption. The moderating effects of GenAI system attributes (H8a–H8d) provided additional insights. H8a (GenAI Quality × ML → PU) (β = 0.133, t = 10.710, f² = 0.041) showed that reliable and accurate GenAI responses helped reduce the negative influence of mental load on usefulness. H8b (GenAI Transparency × PU → Attitude) (β = 0.144, t = 13.555, f² = 0.064) indicated that when GenAI explained its logic clearly, the connection between usefulness and positive attitudes became stronger. H8c (GenAI Friction Reduction × Attitude → BI) (β = 0.179, t = 14.821, f² = 0.084) demonstrated that when GenAI simplified everyday tasks and reduced effort, favorable attitudes more easily converted into usage intentions. Finally, H8d (GenAI System Integration × BI → AU) (β = 0.115, t = 11.027, f² = 0.038) showed that when GenAI tools fit smoothly into users’ existing apps and digital environments, intentions were more likely to become actual behavior. At the overall model level, H9 (Extended vs. Baseline TAM) was clearly supported, with higher explanatory power (R² for Attitude = 0.677; Actual Use = 0.642) and stronger predictive power (Q² = 0.614) compared to the baseline. This demonstrates that including GenAI attributes as moderators not only explains more variance in adoption but also improves predictive accuracy, making the extended TAM framework more robust for general everyday users.

From a theoretical perspective, the relatively small yet statistically significant effect of mental load on user attitude can be explained through Cognitive Load Theory. While cognitive strain negatively influences users’ evaluative responses, GenAI users may partially tolerate increased mental effort due to the perceived instrumental value of generative systems. This suggests that cognitive burden does not fully deter favorable attitudes when users anticipate substantial performance or productivity gains. Accordingly, mental load operates as a constraining rather than a dominant force in shaping attitudes, highlighting the nuanced role of cognitive strain in technologically advanced, high-utility systems such as GenAI.

4.3. Importance—Performance Map Analysis

The findings show that behavioral intention (β = 0.749, f² = 1.553, performance = 49.289) and attitude toward GenAI (β = 0.522, f² = 1.253, performance = 49.347) are the strongest drivers of actual use but perform slightly below average, so they should be priority areas for improvement. Perceived usefulness (β = 0.277, f² = 0.464, performance = 50.464) and perceived ease of use (β = 0.203, f² = 0.283, performance = 50.107) already perform above average and should be maintained. System features such as friction reduction (β = 0.195, f² = 0.084, performance = 50.006) and system integration (β = 0.192, f² = 0.038, performance = 49.879) can be optimized to smooth adoption. Transparency (β = 0.136, f² = 0.060, performance = 49.874) and quality (β = 0.125, f² = 0.041, performance = 49.921) are stable but supportive trust factors. Finally, mental load (β = −0.083, f² = 0.011, performance = 49.926) is a small barrier, so reducing stress and complexity will further encourage adoption, as shown in Figure 3 and illustrated in

Table 5. Construct Importance, Performance, Effect Sizes, and Practical Implications for GenAI Adoption.

Rank	Construct	Importance (β)	Performance	f2 Effect Size	Category	Performance Gap	Implication
1	Behavioral Intention	0.749	49.289	1.553	High	−0.6	Key driver, focus on sustaining intention → actual use
2	Attitude Towards Using GenAI	0.522	49.347	1.253	High	−0.5	Strengthen attitudes through positive user experiences
3	Perceived Usefulness	0.277	50.464	0.464	Medium	+0.6	Already strong, continue highlighting usefulness
4	Perceived Ease of Use	0.203	50.107	0.283	Medium	+0.3	Ease is adequate, less urgent for improvement
5	GenAI Friction Reduction	0.195	50.006	0.084	Medium	+0.2	Supports adoption, can be optimized
6	GenAI System Integration	0.192	49.879	0.038	Medium	−0.1	Needs improvement to ensure seamless adoption
7	GenAI Transparency	0.136	49.874	0.060	Low	−0.1	Transparency builds trust, invest moderately
8	GenAI Quality	0.125	49.921	0.041	Low	0.0	Stable, less immediate priority
9	Mental Load	−0.083	49.926	0.011	Negative	0.0	Barrier—should be minimized with supportive features

.

5. Discussion

5.1. Integration of TAM with GenAI System Characteristics

This study builds on prior research on ChatGPT and GenAI adoption, including the work of Bashir and Zhou, by showing that the core psychological mechanisms proposed by the Technology Acceptance Model (TAM) remain relevant in the context of Generative AI [6,57,58]. However, the findings clearly demonstrate that these psychological drivers do not operate in isolation. Instead, their effects are strongly shaped by GenAI-specific system characteristics such as quality, transparency, friction reduction, and system integration. In other words, while users’ perceptions and attitudes matter, the way GenAI systems behave in practice can either reinforce or weaken these psychological influences. By linking classical technology acceptance research with emerging work on AI quality and explainability, this study offers a unified and more context-sensitive explanation of GenAI adoption [7,61].

5.2. Validation and Extension of TAM Relationships

Consistent with decades of TAM research, perceived ease of use significantly influences perceived usefulness and users’ attitudes toward GenAI [57,58]. Systems that are easier to interact with are perceived as more useful and evaluated more positively, confirming findings from both the original TAM literature and subsequent extensions. Perceived usefulness emerges as the strongest predictor of attitude, reinforcing the notion that users form favorable evaluations when they recognize clear performance benefits [59]. Furthermore, attitude strongly predicts behavioral intention, which in turn predicts actual system use. These results align with prior information systems research, including studies on AI adoption, that identify behavioral intention as the most proximal determinant of actual usage in voluntary contexts [59,60]. Overall, these findings confirm that TAM continues to provide a robust foundation for explaining GenAI adoption, while also highlighting the need for contextual enrichment.

5.3. Role of Mental Load and Cognitive Load Theory

The results also support Cognitive Load Theory as a meaningful lens for understanding GenAI adoption. Mental load negatively affects both perceived usefulness and attitude, indicating that cognitively demanding interactions can undermine users’ evaluations of GenAI systems. This finding is consistent with prior research on digital learning environments and complex technologies, which shows that excessive cognitive effort reduces perceived value and satisfaction [8,10]. In the GenAI context, this insight is particularly important because responses can be lengthy, abstract, or difficult to interpret, potentially overwhelming users rather than empowering them. At the same time, the effect of mental load on attitude is relatively small, though statistically significant. This suggests that while cognitive strain constrains users’ evaluations, it does not dominate them. One plausible explanation is that users may tolerate higher mental effort when GenAI systems deliver strong instrumental benefits, such as effective problem-solving or decision support. In such cases, perceived usefulness compensates for cognitive burden, allowing users to maintain generally positive attitudes despite interaction challenges [57,59]. Thus, mental load functions as a limiting condition rather than a decisive deterrent in GenAI adoption.

5.4. Moderating Role of GenAI System Attributes

The strongest theoretical contribution of this study lies in demonstrating how GenAI system characteristics condition the classic TAM pathways. GenAI quality mitigates the negative effect of mental load on perceived usefulness, indicating that accurate, reliable outputs help users maintain a sense of value even when interactions are cognitively demanding. This finding aligns with the DeLone and McLean IS Success Model, which emphasizes the central role of system and information quality in shaping user outcomes [7].

GenAI transparency strengthens the relationship between perceived usefulness and attitude, supporting prior research on explainable AI and trust in information systems [61]. When users understand how and why GenAI produces certain outputs, perceived usefulness is more likely to translate into positive evaluations. Similarly, GenAI friction reduction enhances the conversion of attitudes into behavioral intentions, suggesting that smooth, low-effort interactions help users act on their favorable evaluations. This finding is consistent with prior research on user experience and technology continuance, which shows that reduced interaction barriers facilitate sustained usage [61,62,63]. GenAI system integration further strengthens the link between behavioral intention and actual use. Systems that fit well within existing workflows and digital ecosystems are more likely to be used regularly, reflecting principles of task–technology fit and system integration [14,62]. Collectively, these findings show that DeLone and McLean’s [63] quality dimensions do not merely supplement TAM; instead, they meaningfully shape the strength and effectiveness of traditional acceptance pathways in GenAI environments.

5.5. Interpretation of IPMA and Predictive Findings

The Importance–Performance Map Analysis reveals that attitude and behavioral intention exhibit relatively lower performance despite their high importance. This pattern suggests that although users value GenAI and intend to use it, their evaluative and intentional states may not be fully optimized due to interaction friction, limited transparency, or contextual constraints. In everyday settings, GenAI usage may increasingly be driven by task necessity or habitual reliance rather than deliberate attitudinal reflection, which can suppress the performance of attitudinal constructs even when their theoretical importance remains high [64,65]. Improving system transparency, reducing friction, and enhancing integration can therefore help strengthen the alignment between users’ positive perceptions and their actual usage behavior. From a predictive perspective, the extended model demonstrates clear advantages over the baseline TAM. Higher R² values, stronger Q²predict results, and lower RMSE and MAE indicate that incorporating GenAI system attributes improves both explanatory and out-of-sample predictive power. The CVPAT results further confirm that the extended model significantly outperforms the traditional TAM-based model. These findings respond to recent calls in the PLS-SEM and information systems literature to move beyond explanatory adequacy and evaluate models based on predictive performance [63,64].

5.6. Theoretical, Practical, and Sustainability Implications

Overall, this study supports the core insights of the TAM tradition while demonstrating that psychological perceptions alone are insufficient to explain GenAI adoption. Cognitive constraints highlighted by Cognitive Load Theory and system quality factors emphasized by the DeLone and McLean IS Success Model interact with TAM pathways to produce a more nuanced, multi-layered explanation of user behavior [7,8,10]. This integrated perspective advances current debates in AI and information systems research by showing that acceptance of generative technologies depends not only on beliefs and attitudes, but also on quality, explainability, effort, and system integration within real digital ecosystems [6,61]. The findings also contribute to discussions on digital sustainability. Sustainable GenAI adoption depends on cognitively efficient, transparent, and well-integrated systems that reduce digital fatigue and conserve users’ cognitive resources. High mental load and frictional interactions threaten long-term viability, whereas thoughtful system design promotes sustained engagement and better decision quality. By identifying the conditions under which GenAI use becomes both effective and sustainable, this study enhances understanding of how digital technologies can generate enduring organizational and societal value.

Table 6. Logical Implications, Limitations, and Future Research Directions.

Aspect	Insights from the Present Study	Limitation Addressed	Logical Future Research Direction
Practical Implications	The study demonstrates that GenAI output quality, transparency, friction reduction, and system integration significantly strengthen key Technology Acceptance Model (TAM) relationships. Users are more likely to adopt GenAI when outputs are reliable, explanations are clear, effort is minimized, and tools integrate smoothly into daily routines.	Moderating effects were examined at a single point in time using self-reported perceptions.	Future studies may validate these effects using longitudinal usage data, real-time interaction logs, or experimental manipulations of transparency, quality, and interface design.
Theoretical Implications	By integrating TAM, Cognitive Load Theory, and the DeLone and McLean IS Success Model, the study shows that psychological drivers (perceived usefulness, perceived ease of use, attitude) interact with cognitive strain and system-level attributes to explain GenAI adoption. The integrated model improves both explanatory and predictive power.	The framework includes a limited set of GenAI-specific moderators. Other AI-related constructs were not examined.	Future research may extend the framework by incorporating additional GenAI-specific constructs, such as AI safety, personalization, explainability depth, or hallucination control, and testing their mediating or moderating roles.
Social Implications	Reducing cognitive load and enhancing transparency improves accessibility for non-expert users, older individuals, and users with lower digital literacy, supporting more inclusive adoption of GenAI technologies.	Demographic heterogeneity was not the primary analytical focus of the study.	Future research should examine how age, education, digital literacy, and socio-cultural expectations shape cognitive load, transparency needs, and adoption behavior.
Managerial Implications	Organizations implementing GenAI should prioritize system quality and friction reduction, as these factors help translate favorable attitudes into behavioral intention and actual usage.	Organizational and workplace-specific contexts were not explicitly examined.	Future studies may apply the proposed model in professional settings (e.g., healthcare, education, corporate environments) to identify industry-specific adoption barriers and facilitators.
Technological/System Design Implications	The findings highlight that system design choices—such as high-quality outputs, clear explanations, and seamless integration—directly enhance user trust, satisfaction, and continued use of GenAI systems.	General-purpose GenAI systems were examined without experimental control over design features.	Controlled experiments comparing interface designs, transparency levels, and prompt structures could quantify their effects on mental load, usefulness, and trust.
Predictive and Analytical Implications	The extended model significantly outperforms the baseline TAM in predictive accuracy, demonstrating that system attributes are essential for forecasting real GenAI usage behavior.	Cross-sectional data limit the ability to capture behavioral change over time.	Longitudinal predictive modeling or machine-learning approaches may be used to track evolving patterns of reliance, trust, and repeated use as users gain experience.
General Limitations	The study is based on cross-sectional, self-reported data from a single national context with a limited set of moderators.	—	Each limitation opens avenues for future research, including longitudinal designs, multi-source data, cross-cultural comparisons, richer GenAI constructs, and multi-sector analysis.

illustrates the implication limitations, and future research direction of the research.

6. Conclusions

This research aimed to examine the adoption of Generative AI tools by everyday users, building on the familiar model created by Bashir and Zhou. However, whereas the initial model emphasized the roles of mental load, perceived ease of use, perceived usefulness, and attitude, the current study indicates that GenAI adoption relies solely on psychological perceptions. Since GenAI systems act dynamically and unpredictably, the properties of the system, including the quality of the output, transparency, reduced friction, and integration, are critical to defining the way users assess and apply them.

The results affirm the notion that the conventional TAM relationships hold in the GenAI scenario: perceived ease of use increases perceived usefulness and attitude; perceived usefulness is the strongest predictor of attitude; and positive attitudes are strong predictors of behavioural intention, which eventually leads to actual use. The mental load remains a significant obstacle, with diminished usefulness and a negative attitude. The findings further support the need to create GenAI systems that are user-friendly, cognitively simple, and usable for non-expert users.

One of the greatest contributions of this research is demonstrating how the attributes of a GenAI system reinforce major psychological pathways. The quality of the outputs keeps users in a positive frame of mind even when the tasks seem complicated. Transparency strengthens the relationship between usefulness and attitude by fostering trust. Friction reduction increases the likelihood that users will act on their positive attitudes, and high levels of system integration enhance the likelihood that intentions will be converted into actual use. Combined, these findings indicate that psychological and system-level aspects collaborate to define contemporary AI adoption.

The extended model was also much more predictive than the baseline TAM, with higher R² and Q² values and lower prediction errors. This shows that system-quality aspects are not merely auxiliary but must be considered when accurately predicting user interactions with advanced AI systems.

The research, in general, offers a more holistic and realistic view of the adoption of GenAI by integrating psychological drivers, cognitive strain, and system quality into a single framework. It emphasizes that future research on AI adoption, as well as the design of AI systems, must consider not only how people think about AI but also how they feel about it and how the system will act in actual circumstances. This model can provide a solid basis for enhancing the user experience, informing system design, and encouraging responsible, meaningful adoption of Generative AI in daily life by connecting these dimensions.