Study on the influencing factors of generative AI tool usage effect among college students
Yihan Zhang 1,*
- School of Management, Xi’an Polytechnic University, Shaanxi 710600, China
- Correspondence: 13633813289@ 163.com
Abstract: Breakthroughs in generative artificial intelligence technology are reshaping the higher education ecosystem. This study, based on the theoretical framework of technol- ogy acceptance model and integrated technology acceptance and use model, constructs a multi-level influence mechanism model by combining the technical characteristics of generative AI. Through structural equation modeling analysis of questionnaire survey data from college students, it reveals the formation path and group-specific features of the effects of using generative AI tools. The study finds that usage purpose and trust mechanisms can significantly enhance perceived usefulness and perceived ease of use, which in turn positively influence usage attitudes, behavioral intentions, and actual usage behaviors, significantly impacting user outcomes. The research provides a theoretical basis for universities to build a generative AI educational ecosystem and proposes systematic optimization strategies from functional adaptation to ethical regulation
Keywords: generative artificial intelligence; technology acceptance theory; structural equation model; influencing factors; use effect
- Introduction
With the rapid development of generative artificial intelligence (Generative AI) tech- nology, AI tools represented by ChatGPT have demonstrated tremendous potential in education, research, business, and other fields. Generative AI not only generates high- quality text, images, and code but also interacts with users through natural language processing techniques, significantly enhancing the efficiency of information acquisition and processing [1]. In recent years, the application of generative AI tools among college students has become increasingly widespread, with students using these tools to assist in learning,complete assignments, and conduct scientific explorations. The penetration rate of generative AI technology among faculty and students in domestic universities has reached 99% [2], but there are significant differences in student usage effects [3]. Most exist- ing research focuses on technical characteristics while neglecting the interaction between subjects and environmental factors, leading to theoretical explanatory gaps. However, despite the many conveniences brought by the use of generative AI tools, their actual effects and impacts on students’ learning behaviors, academic performance, and innovative capabilities remain highly controversial.
In the current educational environment, college students, as the primary user group of generative AI tools, are influenced by various factors in their usage effectiveness. To delve into how these factors impact students’ use of generative AI tools, this study will analyze the actual effects and influencing factors from multiple dimensions. Through this research, we aim to provide targeted recommendations for educators in higher education
- Use the effect theory model to construct
2.1. Structural equation model
Structural Equation Modeling (SEM) is a statistical method based on covariance struc-ture analysis. It constructs a multivariate modeling framework by integrating measurement models (the mapping relationship between observed variables and latent variables) withstructural models (causal paths between latent variables). Structural Equation Modeling features the inclusion of measurement errors in independent variables, support for simul-taneous modeling of multiple dependent variables, and the ability to handle both factormeasurements and inter-factor structures within a single model[4]. In a study on user intention to use generative AI tools in higher education, the authors integrated Technology Acceptance Model planned behavior theory to construct a structural equation model. Theyanalyzed the formation mechanisms of generative AI usage intentions from three dimen-sions: user attitudes, perceived behavioral control, and subjective norms, and examined the differential impacts of internal cognition and external environment on usage intentions[5]. Measurement model refers to the relationship between indicators and latent variables.Structural modelrefers to the relationships among latent variables. Measurement model:The measurement model aims to measure latent variables, using manifest variables toreflect these latent variables. Specifically { , the measurement equation can be expressed as: observed variable = factor loading * latent variable + measurement error.It can also be represented by the formula:, where X,Y represent observed variables; Λx, Λy represents the strength of the relationship between observed variables and latent variables,reflecting the extent to which latent variables explain observed variables; ξ, η represents the value of latent variables; δ, ε represents the measurement error of observed variables, reflecting the part that cannot be fully explained by latent variables. Structural Equation: The structural model reveals the causal relationships and inter-actions between variables. Specifically η = Λη + Γξ + ζ, it can be expressed as follows: Λ represents the value of a latent variable; η represents the degree to which one laten tvariable influences other latent variables; Γ represents the direct impact of an external variable on a latent variable; ξ represents the value of the external variable; ζ represents the unobservable random error term [6] that cannot be explained by other variables amonglatent variables.
2.2. Variable determination
Dr. Davis proposed the Technology Acceptance Model (TAM) in 1989, which estab-lished four core variables: perceived usefulness, perceived ease of use, attitude toward use, and behavioral intention [7], laying the theoretical foundation for subsequent research.In 2003, Venkateshand colleagues further introduced the Unified Technology Acceptance Theory (UTAUT), innovatively incorporating “use purpose” as a key moderating variable for behavioral intention [8]. Notably, a survey study conducted by Zhejiang University among college students validated the moderating effect of “use purpose” in this model.The data showed that generative AI was used most frequently in research activities, with significant differences in use purposes among different genders, majors, and grade levels[9]. Of particular concern is that perceived risk, as a core element of trust assessment systems, significantly influences user decision-making mechanisms [10]. Specifically, in educational settings, students ’trust in AI tools and their perception of academic integrity risks have become important preconditions for shaping attitudes toward use and actual HYPERLINK \l “bookmark20” behavior. Additionally, empirical studies show that the differentiated usage patterns of generative AI among student groups heterogeneously affect their critical thinking develop- ment and autonomous learning capabilities [11]. According to Venkateshetal.’ s [8] model,the perceived usefulness after technology adoption isoperationalized through three core observed dimensions: the extent to which task execution efficiency improves, the degree to which set goals are achieved, and the level of workflow optimization.Based on the comprehensive analysis of the theoretical evolution and empirical re-search results, this study identifies potential variables affecting the effectiveness of gen-erative AI tool usage among college students, including: purpose of use, trust and risk, perceived usefulness, perceived ease of use, attitudes toward use, behavioral intentions, and actual usage behavior. These elements collectively form the theoretical framework of the research model through complex mechanisms. The specific meanings of the eight potential variables are as follows:
|
2.3. Research hypothesis
Through a systematic literature review, it was found that in the current educational environment, college students, as the primary user group of generative AI tools, are influenced by various factors in their usage effectiveness. These factors mainly include subject characteristics (such as gender, age, education level, major category, school type, etc.), usage purpose, external variables such as trust and risk, as well as internal variables like perceived usefulness, perceived ease of use, usage attitude, behavioral intention, and actual usage behavior. Based on this, this study identifies the factors affecting thesupply-demand matching level of health care services for the elderly, proposes 18 basic hypotheses,and constructs a theoretical structural model of the factors influencing the usage effectiveness of generative AI tools among college students (see Figure 1). According to the meaning and interrelationships of the model variables, the following hypotheses areproposed:1) Intention to Use: This is to reflect the goal-oriented use of generative AI tools by college students (such as for work, study, etc.). Li Yongqi et al. [12] argue that the purpose of use by live e-commerce platform users largely determines their extraction of platform functions and content, as well as their perception of the usefulness of these functions. Based on task technology fit theory, the perceived usefulness of generative AI tools by college students depends on the degree of alignment between their use purposes and tool functionalities. Different application scenarios (such as learning assistance, content creation, efficiency improvement) directly influence value judgments by meeting specific task requirements, which are also moderated by individual differences such as profes-sionalbackground and technical proficiency. Based on the above analysis, the following hypothesis is proposed.
Figure 1. Theoretical structural model of factors influencing the use effect of generative AI tools for college students.
H1: The purpose of use positively influences perceived usefulness 2) Gender: In the analysis of the impact of learner characteristics on the acceptance of online learning, Jiang Xue et al. [13] proposed that gender shows significant differences in perceived ease of use and perceived usefulness. Male students have higher perceived usefulness and ease of use compared to female students, indicating that male students find online learning easier to operate than female students. Male students typically engage with technical tools (such as programming and gaming) earlier and adapt to new technologies more quickly, feeling more confident in their ability to handle AI tools, thus they are more likely to recognize their usefulness. Female students, on the other hand, may have less exposure to technical fields, leading to insufficient understanding of the functions and potential of AI tools, which in turn may result in an underestimation of their value. Based on the above analysis, the following hypothesis is proposed. H2: Gender positively influences perceived usefulness;
H3: Gender positively influences perceived usability 3) Age: The “2024 Report on the Attitudes of Adolescents and Young Adults Toward Generative AI” [14] points out that teenagers under 20 are in a period of rapid techno- logical growth, making them more likely to engage with generative AI tools early on. They explore these tools more actively and can quickly identify their value in assisting learning and creation, thereby enhancing perceived usefulness. In contrast, students over 21 face pressures from graduation or employment, focusing more on the practical utility of generative AI tools in academic research and resume optimization. Research by Jiang Xue et al. [13] indicates that older students have a significantly higher perception of ease of use in online learning compared to younger students. Based on the above analysis, the following hypothesis is proposed.
H4: Age positively affects perceived usefulness;H5: Age positively affects perceived usability;4) Educational Background: Students at different educational levels have varyinglevels of perception regarding the usefulness and ease of use of information technology[15]. Whether users are willing to use a particular technology is primarily determined by their perceived usefulness and ease of use, which can be influenced by their ability to adapt to technology, leading to different evaluations of tool practicality and ease of operation [16]. Based on the above analysis, the following hypothesis is proposed.
H6: Educational background positively affects perceived usefulness;
H7: Educational background positively affects perceived usability;5) Types of School: Different types of school significantly influence the perceived
usefulness and perceived ease of use of generative AI tools for college students, primarilyin terms of resource allocation, professional orientation, and technical support. There are 176 significant differences in perceived ease of use and perceived usefulness among studentsfrom different school types[15]. Based on the above analysis, the following hypothesis is proposed.
H8: School category positively influences perceived usefulness;
H9: School category positively influences perceived usability;
6)Types of Major: Research by Li Yan et al. [9] indicates that there are significantdifferences in the familiarity with generative AI among students from different majorcategories, and these differences also manifest in usage behavior and scenarios. Studentsfrom science and engineering backgrounds, who have long been exposed to programmingand data analysis tools, are more familiar with the interaction modes of generative AI toolsand rate them as more user-friendly. In contrast, humanities students, due to less technical operation experience, may find the tools “difficult to use” because of complex interfaces or difficulties in optimizing prompts. The impact of different professional categories on theperceived usefulness and usability of generative AI tools among college students mainlystems from disciplinary attributes, differences in technical requirements, and learning goal orientation. Based on the above analysis, the following hypothesis is proposed.
H10: Positive influence of professional category on perceived usefulness;
|
H11: Professional categories positively influence perceived usability;
- Trust and Risk: Luhmann posits that trust is a mechanism of social operation, often
|
- implying the ability to recognize and accept risks brought about by uncertainty, thereby
|
- reducing societal complexity[17]. Therefore, trust in generativeAIis a prerequisite for
|
- human-machine collaboration in knowledge production.Human trust in generative AI
|
- directly influences perceived usefulness andeaseof use, which in turn determines the
|
- adoption of AI tools[18]. The degree of students ’trust in AI tools and their perception of academic integrity risks are key variables affecting perceived usefulness and ease of
|
- Trust enhances users’ perceived usefulness and ease of use, leading them to accept AI forknowledge creation[19]. Based on the above analysis, the following hypothesis is proposed.
- H12: Trust and risk positively influence perceived usefulness;
H13: Trust and risk positively influence perceived usability;
8) Perceived usefulness and perceived ease of use: Perceived usefulness refers toan individual’s trust in the extent to which using a specific system will improve their performance or achievement [20]. Simply put, it is whether people believe that using a certain technology can enhance their work efficiency or quality of life. Perceived ease of use 209 refers to an individual’s perception of how easy or difficult it is to use a specific information
system [21]. Perceived usefulness and perceived ease of use are two core concepts in the technology acceptance model, both of which are crucial for influencing people’s adoption and use of technology. The premise of technology acceptance lies in perceived ease of use, as users tend to embrace technologies that require minimal effort to use. College students,when using generative AI tools, expect to master basic platform functions in as short a5 time as possible in a simple and quick manner, and to easily understand the knowledgeand information generated during interactions with the tool. Existing research has shown that perceived ease of use is a significant factor in the acceptance and adoption of new technologies in higher education students [22]. LuoFeiet al. [23] pointed out that perceived usefulness and perceived ease of use jointly influence behavioral attitudes when exploring how DeepSeek’s technical characteristics can empower educational digital transformation. ChatGPT and Gemini et al.noted that large language models have clear limitations in handling cross-linguistic tasks, which are particularly prominent in different language environments, directly affecting users’ perceptions of usefulness and ease of use, thereby influencing their behavioral attitudes [23]. Based on the above analysis, the following hypothesis is proposed:H14: Perceived usefulness has a positive effect on use attitude; H15: Perceived usability has a positive effect on use attitude9) Attitude towards Using: As a core variable in TAM, usage attitude significantly influences college students ’behavioral intentions toward generative AI tools [24]. Specif- ically, positive attitudes toward generative AI tools directly enhance their willingness to use them, manifesting as proactive exploration of tool functions, continuous use, or recommending others to use them; conversely, negative attitudes may suppress behavioral intentions, leading to avoidance or restriction of use. For example, when students believe that generative AI tools can efficiently assist academic research and are easy to operate, they tend to form a positive attitude and are more willing to integrate these tools into their daily learning processes; however, if students are concerned about the risk of academic misconduct associated with the content generated by the tool, ethical anxiety may reduce usage frequency. Research by Liu Qipinget al. [5] also shows that college students’ atti-tudes have a positive and significant impact on their intention to use generative AI. Based on the above analysis, the following hypothesis is proposed: H16: Use attitude positively influences behavioral intention;10) Behavioral Intention: The core assumption of TAM is that actual usage behavior(such as frequency of use, depth of function exploration) is primarily driven directly by individual behavioral intention [24]. Behavioral intention is the direct antecedent variable of actual behavior, reflecting users’ acceptance of technology and their motivation to use it proactively. Based on the above analysis, the following hypothesis is proposed: H17: Behavioral intention positively affects actual use behavior; 11) Actual System Use: The actual usage behavior directly impacts the effectiveness of university students ’use of generative AI tools through frequency and depth of functionexploration. Specifically, high-frequency and deep usage can enhance the tool’s alignment with tasks, thereby significantly boosting academic efficiency or innovative outcomes;hallow usage, however, may result in underutilization of the tool’s potential, leading to limited effects. Students who frequently use generative AI tools show a significant improvement in academic task efficiency, with both the quantity and quality of their generated content surpassing those of low-frequency users. A survey indicates that students who consistently use generative AI tools perform 25%[11] better in critical thinking andself-directed learning compared to non-consistent users. Based on the above analysis, thefollowing hypothesis is proposed: H18: Actual use behavior has a positive impact on the effect of use.
Table 1. Index system.
| latent variable | Observe variables | Questionnaire items | reference
documentation |
|
Perceive usefulness |
PU1 | Generative AI tools help me to complete learning tasks more efficiently |
Davis, F. D. (1989).[21] |
| PU2 | Generative AI tools have significantly improved the quality of my assignments or projects | ||
| PU3 | Generative AI tools provide me with new ideas to solve complex problems | ||
| PU4 | Generative AI tools make me more competitive in exams or research | ||
|
Perceive ease of use |
PEOU1 | The interface design of generative AI tools is intu- itive and easy to operate |
Davis, F. D. (1989).[21] |
| PEOU2 | Using generative AI tools doesn’t take much effort for me | ||
| PEOU3 | I was able to quickly learn how to use generative AI tools | ||
| PEOU4 | I can operate the various functions of generative AI tools with ease | ||
| Attitude towards using | ATT1 | Overall, I’m positive about using generative AI tools | Fishbein, M.,
& Ajzen,I. (1975).[25] |
| ATT2 | Ithink it’s a pleasure to use this generative AI tool | ||
| behavioral intention | BII | I intend to continue using generative AI tools in the future | Davis, F. D. (1989).[21] |
| BI2 | I plan to increase the frequency of use of generative AI tools | ||
| Actual
system use |
USE1 | Over the past month, I’ve used this technology a lot | Venkatesh, V.,
& Davis, F. D. (2000).[20] |
| USE2 | The time it takes me to complete a task using this technique is usually | ||
|
Trust and risk |
TR1 | I believe generative AI tools can reliably perform their functions |
Mayer, R. C., Davis,J. H., & Schoorman, F. D. (1995).[26] |
| TR2 | The services and features of generative AI tools are trustworthy | ||
| TR3 | When using generative AI tools, I worry that my personal information maybe compromised | ||
| TR4 | Using generative AI tools makes me feel insecure about my privacy | ||
|
Intention to use |
IU1 | Using generative AI tools is critical to my ability to get information |
Katz, E., et al. (1973).[27] Venkatesh, V., et al. (2003).[8] UTAUT Deci, E. L., & Ryan, R. M. (1985).[28] |
| IU2 | Improving efficiency is my core goal when using generative AI tools | ||
| IU3 | Generative AI tools are essential to my creative work | ||
| IU4 | I rely on generative AI tools to learn new skills or knowledge | ||
| IU5 | Entertainment is one of the main reasons I use gener- ative AI tools | ||
|
Effect of use |
XG1 | Using generative AI tools has improved my effi- ciencyin completing tasks |
Venkatesh, V., Morris, M. G., Davis, G. B., & Davis, F. D. (2003).[8] |
| XG2 | Using generative AI tools has helped me better achieve my goals | ||
| XG3 | Generative AI tools have helped me optimize my work / study process |
In Amos, a generative AI tool SEM model is constructed (see Figure 2). e1-e26 arerandom error terms.
Figure 2. SEM path of generative AI tool.
- Data collection and analysis
3.1. Data collection
According to the index system design see Table 1, the questionnaire of this study mainly includes the personal information of college students, the influencing factors includ- ing 26 variables and the indicators related to the use effect, and the question measurement adopts the Likert 5-point scale.The data were obtained through questionnaires, which were conducted among college students on campus and participated in by scanning QR codes online and offline. A total of 213 valid questionnaires were received see Table 2.
Table 2. Descriptive statistical analysis.
| project | option | frequency | percentage |
| Gender | man | 104 | 48.8 |
| woman | 109 | 51.2 | |
| Age | 18-25 | 128 | 60.1 |
| 26-30 | 85 | 39.9 | |
| Educational background | junior college education | 74 | 34.7 |
| undergraduate course | 73 | 34.3 | |
|
Types of school |
Master | 49 | 23.0 |
| doctor | 17 | 8.0 | |
| general works | 25 | 11.7 | |
| science
and engineering |
26 | 12.2 | |
| normal category | 23 | 10.8 | |
| Finance | 30 | 14.1 | |
| Medical category | 26 | 12.2 | |
|
Types of major |
Agriculture and forestry | 26 | 12.2 |
| arts | 22 | 10.3 | |
| Sports | 19 | 8.9 | |
| Language-based | 16 | 7.5 | |
| philosophy | 14 | 6.6 | |
| economics | 23 | 10.8 | |
| law | 8 | 3.8 | |
| education | 20 | 9.4 | |
| literature | 19 | 8.9 | |
| computer science | 16 | 7.5 | |
| history | 14 | 6.6 | |
| science | 15 | 7.0 | |
| Engineering(except computer science) | 16 | 7.5 | |
| agronomy | 11 | 5.2 | |
| medical science | 16 | 7.5 | |
| management | 24 | 11.3 | |
| art | 17 | 8.0 |
3.2. Data analysis
3.2.1. Reliability test The Cronbach’s alpha (Cronbach’s alpha) is a commonly used method for internalconsistency testing, assessing the correlation between various indicators in a questionnaire.ts value ranges from 0 to 1; the closer it is to 1, the higher the internal consistency and reli-ability of the questionnaire. Generally, an α value greater than 0.7 is considered acceptable for [29]. The reliability test results for the questionnaire are shown in Table 3.
Table 3. Reliability analysis.
| Perceive | Perceive ease | adopti on | behavior Actual use Trust | Purpose | using | Overall | |
| usefulness | of use | attitude | disposition behavior and risk | of use | effect | reliability | |
| Cloningbach coefficienta | 0.880 | 0.871 | 0.877 | 0.835 0.786 0.87 8 | 0.885 | 0.86 9 | 0.932 |
It can be seen that the coefficient α of clonebach is 0.932, and the reliability of 8 potential variables is greater than 0.7, which can be regarded as the questionnaire has a high degree
|
of trust,and the reliability meets the requirements.
3.2.2. Validity test
Validity tests are usually carried out by KMO measurement and Bartlett spherical test.
|
These two methods can be used to evaluate the factor structure validity of questionnaires.
|
KMO measurement: KMO measurement measures the correlation between variables
|
in factor analysis. The KMO value ranges from 0 to 1, and it is generally considered that a
|
KMO value greater than 0.7 indicates that factor analysis is feasible.
|
Bartlett Correlation Test: The Bartlett correlation test is used to examine whether
|
there is a relationship between variables. This testis based on the assumption that there
|
is a correlation among the variables understudy. If the test results in rejecting the nullhypothesis (p-valueless than the set significance level of 0.005), it can be concluded that 1 there is a correlation between the variables, making them suitable for factor analysis.The questionnaire data were tested by KMO and Bartlett standard test with SPSS 26.The KMO value was 0.894, which was greater than 0.7, so the questionnaire data were very 4 suitable. The significance probability of Bartlett spherical test was 0.000, which was far less than 0.005. Therefore, the correlation between variables was good, so exploratory factor
|
analysis could be carried out. The validity test results are shown in Table 4.
Table 4. KMO measurement and Bartlett’s spherical test.
| KMO and Bartlett test | ||
| KMO sample appropriateness measure. | .894 | |
| Bartlett’s
test of sphericity |
Approximate chi-square free degree
conspicuousn ess |
3396.236 325
.000 |
Table 5. Model fit test.
| metric | CMIN | DF | CMIN/DF | GFI | RMSEA | CFI | TLI | IFI |
| ideal value | – | – | < 3 | >0.9 | <0.08 | >0.9 | >0.9 | >0.9 |
| Target value | – | – | < 5 | >0.8 | <0.10 | >0.8 | >0.8 | >0.8 |
| fitted value | 824.983 | 420 | 1.964 | 0.821 | 0.067 | 0.883 | 0.871 | 0.885 |
- Model test
Adaptability testis a method to evaluate whether the model is suitable for the data.Amos is used to fit the model, and the fitting parameters and fitting values of the model are obtained, as well as the standardized fitting index (see Table 5). The table above presents the various fit indices of the model. The results show that: the CMIN is 824.983 , the DF is 420, and the CMIN/DF ratio is 1.964<3, which is relatively ideal. The RMSEA is 0.067<0.08, and the indicators of GFI, CFI, TLI, and IFI are all greaterthan 0.8. In summary, all indicators meet the standard requirements, indicating a good fit of he model. The model established in this paper on the factors influencing the effectivenessof generative AI tools used by college students has a high degree of alignment with the survey sample, making it suitable for evaluating and analyzing the factors affecting the effectiveness of generative AI tools used by college students.
- Discussion and analysis of results
5.1. Result analysis
The standardized path coefficient is an indicator used to measure the strength of the
|
relationship between latent and observed variables. Its magnitude indicates the degree of
correlation between two variables. If the standardized path coefficient is large, it suggests a strong correlation between the two variables. The standard error represents the samplingerror of the coefficient estimate; the smaller the standard error, the more precise the estimate.
|
The results of structural equation modeling for the effect of generative AI tool useby college students are shown in Figure 3. The numbers in the figure are standardizedregression coefficients between variables, indicatingH the degree of influence between
|
variables.
Table 6. Standardized path analysis results.
| argument | dependent variable | Standardized path coefficient | standard error | z (C.R.) | p |
| Intention to use | Perceive usefulness | 0.33 | 0.098 | 4.525 | *** |
| Trust and risk | Perceive usefulness | 0.361 | 0.071 | 5.122 | *** |
| Trust and risk | Perceive ease of use | 0.554 | 0.064 | 7.606 | *** |
| Gender | Perceive usefulness | -0.056 | 0.156 | -0.964 | 0.335 |
| Age | Perceive usefulness | -0.223 | 0.161 | -3.835 | *** |
| Educational background | Perceive usefulness | 0.179 | 0.083 | 3.081 | 0.002 |
| Types of school | Perceive usefulness | -0.018 | 0.032 | -0.313 | 0.754 |
| Types of major | Perceive usefulness | 0.012 | 0.02 | 0.2 | 0.841 |
| Gender | Perceive ease of use | -0.045 | 0.146 | -0.724 | 0.469 |
| Age | Perceive ease of use | 0.019 | 0.149 | 0.31 | 0.756 |
| Educational background | Perceive ease of use | -0.05 | 0.077 | -0.81 | 0.418 |
| Types of school | Perceive ease of use | -0.067 | 0.03 | -1.075 | 0.282 |
| Types of major | Perceive ease of use | -0.043 | 0.019 | -0.694 | 0.488 |
| Perceive usefulness | Attitude towards Using | 0.39 | 0.061 | 5.569 | *** |
| Perceive ease of use | Attitude towards Using | 0.39 | 0.071 | 5.493 | *** |
| Attitude
towards Using |
Behavioral intention | 0.496 | 0.091 | 6.263 | *** |
| Behavioral intention | Actual
system use |
0.481 | 0.055 | 5.423 | *** |
| Actual
system use |
Effect of use | 0.525 | 0.116 | 6.282 | *** |
*** p < 0.001
Figure 3. Estimation results of model parameters.
According to the results of Amos software (see Table 6), the standardized path coeffi- cient from usage purpose to perceived usefulness is 0.33 (z=4.525, p<0.05),indicating that usage purpose has a significant positive impact on perceived usefulness; the higher the usage purpose, the higher the perceived usefulness. When college students have clearergoal orientation and stronger intrinsic motivation when using generative AI tools, they are more likely to evaluate the practical value of these tools in improving work and studyefficiency and meeting task objectives from their own needs, thereby enhancing “perceived usefulness.” The standardized path coefficient from trust and risk to perceived usefulness is 0.361 (z=5.122, p<0.05), indicating that trust and risk have a significant positive impact on perceived usefulness; the higher the trust in generative AI tools and the lower the percep- tion of potential risks during use, the higher the subjective evaluation of the actual practical value of these tools. The standardized path coefficient from trust and risk to perceived ease of use is 0.554(z=7.606, p<0.05), indicating that trust and risk have a significant positive impact on perceived ease of use. From a behavioral psychology perspective, trust works by reducing users’ psychological defenses against technical complexity: when users trust the system, they are more likely to perceive the interface design as reasonable and the operation 336 process intuitive, thus reducing their subjective judgment of operational difficulty. Thecontrollability of risk perception is achieved by reducing cognitive load: if users believe the system can effectively avoid risks, they do not need to invest extra attention in monitoring the operation process, thus perceiving higher usability. For example, when using intelligent office software, if users trust their data security and believe that accidental operations can be recovered, they are more likely to focus on the task itself rather than technical details, thereby experiencing a subjective sense of “smooth operation.” The standardizedpath coefficient from age to perceived usefulness is -0.223 (z=-3.835, p<0.05),indicating a significant negative impact of age on perceived usefulness; that is,the older the age, the lower the perceived usefulness. Although college students are generally in their youth, age differences still correspond to varying depths of technology exposure. For instance, freshmen aged 18 are mostly “digital natives,” having been exposed to smart devices and the internet since childhood, with a clearer understanding of the functional boundaries of technology; whereas senior students over 25 (such as graduate students) may have weaker intentions to explore new technologies due to limited early access to technical educational resources. This generational difference leads older students to rely more on traditionallearning methods, thus underestimating the practical value of digital tools. Additionally, earning new technologies requires time and effort, and older students may reduce their patience for technological exploration due to academic pressure or family responsibilities.For example, when faced with complex data analysis tools, younger students might view them aschallenges and actively learn, while older students might directly dismiss their value out of concern for operational errors or high time costs. The standardized pathcoefficient from education to perceived usefulness is 0.179 (z=3.081, p<0.05),indicating a significant positive impact of education on perceived usefulness. This relationship stemsrom the deep reliance on technology in higher education: the increased complexity of academic tasks compels students to actively master professional tools, with technology directly serving research output and career development. Meanwhile, systematic technicaltraining provided by universities to high-education groups further reinforces their under-standing of tool effectiveness, forming a cyclical mechanism of “education advancement → upgraded technical needs → enhanced perceived usefulness.” The standardized path coefficient from perceived usefulness to usage attitude is 0.39(z=5.569, p<0.05),indicating a significant positive impact of perceived usefulness on usage attitude, meaning the higher the perceived usefulness, the higher the usage attitude. This relationship is rooted in the core logic of TAM: when university students believe that generative AI tools can effectively solve problems, theirrational cognition transforms into emotional recognition, forming a psychological drive of “useful → want to use,” which is then reinforced through actual 372 usage experience, enhancing the persistence of attitudes. The standardized path coefficient rom perceived ease of use to usage attitude is 0.39(z=5.493, p<0.05), indicating a significant 4 positive impact of perceived ease of use on usage attitude, meaning the higher the perceived ease of use, the higher the usage attitude. According to TAM, when university students 376 believe that generative AI tools can be easily mastered without complex learning, their 377 psychological cognitive burden decreases, reducing frustration, thus forming a positive 378 attitude of “willing to use.” The standardized path coefficient from usage attitude to be- havioral intention is 0.496(z=6.263, p<0.05), indicating that usage attitude has a significant positive impact on behavioral intention; the higher the usage attitude, the higher the be- havioral intention. According to TAM, attitude, as a concentrated manifestation of affective cognition, drives individuals to convert their intrinsic preferences into action tendencies by reducing decision hesitation and reinforcing the rationality of behavior. The standardized path coefficient from behavioral intention to actual usage behavior is 0.481(z=5.423, p<0.05), indicating that behavioral intention has a significant positive impact on actual usage be- havior; the higher the behavioral intention, the higher the actual usage behavior. The standardized path coefficient from actual usage behavior to usage effect is 0.525(z=6.282, p<0.05), indicating that actual usage behavior has a significant positive impact on usage effect. Through continuous use, students gradually master the tool’scharacteristics and deeply integrate them into academic settings, forming a closed loop of “usage behavior reinforcement → skill proficiency improvement → task result optimization.” Gender has no significant impact on perceived usefulness. Possible reasons include: qual opportunities for technical exposure between men and women in moderneduca- tional environments, tool design trending towards neutrality; the functional positioning of generative AI is not directly related to gender; gender differences in technological cognition among college students have weakened with the popularization of digital literacy. The type of school has no significant impact on perceived usefulness. This may be because, regardless of the school type, the widespread adoption and promotion of generative AI tools have made access and usage similar among students, thus the type of school is no longer a critical factor. The direct correlation between students’ usage purposes and technical functions may obscure the environmental differences brought about by the type of school. The professional category had no significant effect on perceived usefulness. This result suggests that the general characteristics of generative AI tools may transcend professional barriers and meet the common needs of students in various disciplines. Gender had no significant effect on perceived usability. This result maybe due to the neutral design of generative AI tools and the gender equality trend in technology literacy cultivation in modern educational environment, which makes gender no longer a key factor affecting perceived usability. Age has no significant impact on perceived usability. This result may stem from the common technological literacy among college students: despite age differences, students generally receive systematic digital skills education, such as foundational computer courses, and the interface design of generative AI tools inherently supports intergenerational friend- liness, making age no longer a key factor influencing perceived usability. Educational background has no significant impact on perceived usability. This result may stem from the intelligent design of generative AI tools, which significantly lowers the learning threshold, preventing the technological advantages of highly educated individuals
|
from translating into differences in perceived usability. Additionally, college students generally have higher technical literacy and can quickly adapt to tool operations regardless of their educational level. There was no significant effect of school type on perceived usability. This suggests that 2 the interface design and interaction logic of generative AI tools may have been universal, and that students with different educational backgrounds tended to perceive the difficulty of mastering technology in a similar way. The professional category had no significant effect on perceived usability. This result suggests that the interaction design of generative AI tools may have broken through the disciplinary barriers, and its intuitive interface logic and low learning cost features enable students from all majors to quickly master basic operations.
5.2. Suggestions to improve the effectiveness of generative AI tools used by college students Based on the above research, suggestions are put forward for college students to use generative AI tools to improve the effectiveness of their use:
- Make clear the purpose and functional compatibility of use: develop AI tool modules with strong adaptability according to different professional needs, so as to help students achieve academic goals more accurately. 2. Strengthen trust and risk management: enhance students ’trust in the reliability of technology by transparently operating the logic of AI tools; optimize the data security protocol of AI tools, clarify the user privacy protection policy, and reduce students’ 438 perceived risk of privacy leakage through regular security certification.
- Optimize perceptual usefulness and ease of use: Embed AI tools into practical opera- tion links in traditional classrooms to let students intuitively experience the efficiency improvement of tools and enhance perceptual usefulness. Optimize the interface friendliness of AI tools to reduce learning costs, especially pay attention to the use experience of senior students or students with non-technical background.
- Promote attitude and behavior transformation: through the sharing of successful cases, stimulate students’ interest and confidence in tools, and form a positive cycle of “useful means willing to use”.
|
- Focus on group differences and ethics education: Provide special technical guidance
|
for graduate students or senior students to alleviate their negative attitudes caused by time pressure or technical anxiety. Integrate AI ethics courses into technical training to guide students to use tools reasonably and balance efficiency improvement with academic integrity.
- Conclusion
This study combines empirical data from college students, using questionnaire surveysand relevant statistical analysis methods to construct a theoretical model and verify its
|
reliability and validity. The study analyzes the actual effects and influencing factors ofcollege students when using generative AI tools from multiple dimensions. Through this
|
research, we aim to provide targeted recommendations for educators in higher education,helping students better utilize generative AI tools for learning and innovation in the age of artificial intelligence, while avoiding potential negative impacts.
Author Contributions: Yihan Zhang was responsible for the overall design of this study, includingthe planning of the research ideas and the formulation of the experimental protocol. Yihan Zhangindependently completed the data collection work to ensure the accuracy and integrity of the data.Meanwhile, Yihan Zhang undertook the task of writing the paper, from the initial draft conception to 4the content refinement, all of which were independently accomplished by her
Funding: This research did not receive any specific grant from funding agencies in the public,commercial, or not-for-profit sectors. 46Institutional Review Board Statement: This study does not involve human subjects, so ethical approval from an Institutional Review Board was not required.Informed Consent Statement: For the study, data were collected anonymously, and informed consent was obtained from all participants regarding the publication of this study.Data Availability Statement: The data supporting the results of this study are included within thearticle and its Supplementary Materials.Conflicts of Interest: The authors declare no conflict of interest.
References
- Lan, B.; Yang,J. The mechanism, risks and prevention of new quality productivity enabled by generative artificial intelligence.
Party Building and Ideological Education in Schools 2025, pp. 15–19.
|
- Zhao, L.; Chen, Y. Where is the boundary of student use of AI?, 2025. Legal Daily, (004)
Wang, Q. Application and Prevention of Generative AI in Graduate Research Tasks. Software Guide 2025, 24, 1–7.
Lin, S.; Jiang, Y. Theory of structural equation modeling and its application in management research. Science of Science and
Technology Management 2006, pp. 38–41.
- Liu, Q.; Zhang, Z.; Yang, P. A Study on Users’ Willingness to Use Generative AI Tools in Higher Education. Journal of Hubei
Second Normal University 2024, 41, 85–93.
- Wu, B. Preliminary study on structural equation modeling. Master’s thesis, Tianjin University, 2006.
- Chen, Y.; Yin, X.; Zhang, C. Research on the Formation Mechanism of Digital Resilience in College Students’ Online Learning —
Based on the Mediating Role of Teacher Support. Educational Review 2024, pp. 60–70.
- Venkatesh, V.; Morris, M.G.; Davis, G.B.; Davis, F.D. User acceptance of information technology: Toward a unified view. MIS
- Li, Y.; Xu,J.; Jia, C.; et al.. The Current Situation and Thinking of Generation AI Application among College Students – Based on
the Survey of Zhejiang University. Open Education Research 2024, 30, 89–98.
- Yang, X.; Peng, D.; Xie, F. Research on the Impact of Perception Risk Cognition Based on TAM/TPB on User Trust and Behavior:
A Case Study of Yu ’ebao, a Value-added Payment Product. Management Review 2016, 28, 229–240.
- Qi,J.; Xu, Y.; Liu, J.;et al.. The Impact of Generative AI Tool Use on College Students’ Critical Thinking and Autonomous
Learning Ability. Research on Electro-Education 2024, 45, 67–74.
- Li, Y.; Ma, S.; Chen, X. Analysis on the Adaptability of Ethnic Products to Live Streaming E-commerce Platforms Based on
Technology Acceptance Model. Modernization of Shopping Mall 2021, pp. 47–49.
Technology 2024, 21, 94–98. 498
- Smith,J.K.; Lee, H. Teen and Young Adult Perspectives on Generative AI: Patterns of Use, Excitements, and Concerns. Journal of 499
Adolescent Technology Studies 2024, 15, 123–145. 500
- Zhang, Y.; Ke, H. Research on the Acceptance of Virtual Simulation Technology and Teaching Innovation under the Perspective 501
of TAM and TTF Integration. Modern Information Technology 2023, 7, 177–183. 502
- Sun, N.; Tao, Y. Digital Intelligence Resilience: Understanding the Impact of Generative AI on Adolescents–Based on an 503
Examination of the Education and Publishing Field. Digital Publishing Research 2024, 3, 84–93. 504
- Luhmann, N. Trust: A Simplifying Mechanism of Social Complexity; Shanghai People’s Publishing House, 2005; pp. 6–7. 505
- Chen,H.; Zhu,Z.; Jin,J. A Composite Framework, Mechanism and Future Approaches of Trust in Generative Artificial Intelligence 506
from the Perspective of Knowledge Production. China Editing 2024, pp. 60–69. 507
- Hu, B.; Wang,J.; Yan, S. Study on the Influence of Explainability on Users’ Acceptance of AI for Knowledge Creation. Scientific 508
Research Management 2025, pp. 1–15. Online ahead of print, accessed: 2025-04-12. 509
- Venkatesh,V.; Davis,F.D. A theoretical extension of the technology acceptance model: Four longitudinal field studies. Management 510
science 2000, 46, 186–204. 511
- Davis, F.D. Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quarterly 1989, 512
13, 319–340. 513
A mixed-method study. Education and Information Technologies 2019, 24, 2127–2144. 515
- Luo, F.; Ma, Y.; Jiao, L. Iterative Drive Transformation: How the Technical Characteristics of DeepSeek Empower the Digital 516
Transformation of Education. Journal of Soochow University (Education Science Edition) 2025, pp. 1–12. Online ahead of print, 517
accessed: 2025-04-12. 518
PhD thesis, Massachusetts Institute of Technology, 1986. 520
- Douglass, R.B. Belief, attitude, intention, and behavior: an introduction to theory and research, 1977. 521
- Mayer, R.C.; Davis,J.H.; Schoorman, F.D. An integrative model of organizational trust. Academy of management review 1995, 522
20, 709–734. 523
- Katz, E.; Blumler, J.G.; Gurevitch, M. Uses and gratifications research. The public opinion quarterly 1973, 37, 509–523. 524
- Deci, E.L.; Ryan, R.M. Intrinsic motivation and self-determination in human behavior; Springer Science & Business Media, 2013. 525
- Wu, W. “People-Media-City”: Agglomeration Characteristics and Classification Governance Strategies of Internet Celebrity City 526
Development Model — Data Analysis Based on Top30 Longred Cities. E-Government 2025, pp. 1–21. Online ahead of print, 527
accessed: 2025-04-12. 528