The technology acceptance model (TAM) was first proposed by Davis in 1985 based on the theory of reasoned action (TRA), aiming to explain and predict users' acceptance behaviors of information technologies. Its core constructs include user motivation variables—perceived usefulness and perceived ease of use—as well as outcome variables, namely behavioral intention and actual system use (Davis, 1985). Perceived ease of use and perceived usefulness are the key core variables in the TAM model that explain users' usage intentions and behaviors, together forming the psychological basis for users' acceptance of new technologies (Marangunić and Granić, 2015). Specifically, perceived ease of use refers to the level of effort users perceive when using a specific new technology, i.e., the perceived difficulty in learning and operating the technology; perceived usefulness reflects users' subjective belief that using a specific new technology can improve their work performance or bring specific benefits, especially the perceived advantages in terms of technical performance, such as improving production efficiency, reducing costs, or enhancing decision-making quality.

Since its proposal, the TAM model has been widely applied and validated in the field of information systems due to its simplicity and strong explanatory power, and it has spawned extended models such as TAM2 (Venkatesh and Davis, 2000) and TAM3 (Venkatesh and Bala, 2008). A large body of existing literature has verified the applicability and explanatory power of TAM and its extended models in the field of agricultural AI usage intentions. For example, a study by Caffaro et al. (2020) on the influencing factors of Italian farmers' adoption intentions toward smart farming technologies showed that a TAM-based mediation model indicated perceived usefulness as the core determinant driving farmers' AI adoption intentions, with its direct positive impact significantly higher than that of perceived ease of use. Similarly, when exploring the influencing factors of Chinese farmers' adoption intentions toward live-streaming e-commerce, Chen et al. (2024) found that external variables such as government support and social learning indirectly affect adoption intentions through the mediating role of perceived ease of use, while perceived usefulness directly and positively influences farmers' adoption decisions. These studies all demonstrate that the TAM model can effectively capture the psychological perception process of agricultural users toward emerging technologies and provide a solid theoretical foundation for understanding their usage intentions.

The unified theory of acceptance and use of technology (UTAUT) was proposed by Venkatesh et al. (2003). This theory integrates the core findings and key constructs of eight mainstream theoretical models in previous technology acceptance research (including TRA, TAM, MM, IDT, SCT, etc.), aiming to more comprehensively and universally predict and explain the acceptance behaviors and usage of new technologies. A notable feature of UTAUT is its universality across different types of new technologies and good transferability across research contexts, enabling it to effectively address differences in technical characteristics and user groups. Therefore, its theoretical framework can provide a more comprehensive perspective for research on the adoption intentions of complex emerging technologies such as agricultural AI.

To better align with the specific context of agricultural AI adoption intentions in this study and improve the model's explanatory power and predictive accuracy, we targeted the adjustment and extension of the original core constructs of UTAUT (performance expectancy, effort expectancy, social influence, and facilitating conditions) by integrating the unique characteristics of agricultural AI and the behavioral traits of agricultural users. Specifically, the adjusted exogenous variables mainly include three core variables related to technology commitment (technological interest, technological competence belief, technological control belief) (Heerink et al., 2010; Kulviwat et al., 2007; Venkatesh et al., 2011) and two key variables related to technological innovation characteristics (mobility, autonomy) (Issa et al., 2022; Soma et al., 2019). These newly added constructs aim to more accurately capture users' intrinsic psychological tendencies toward agricultural AI and the unique attributes of the technologies themselves, thereby supplementing and improving the theoretical framework of the UTAUT model in explaining the adoption intentions of agricultural AI (Neyer et al., 2012).

Among the technology commitment variables, technological interest is a representative psychological attribute reflecting individuals' positive subjective evaluations and curious exploratory desires toward technological progress and its applications, embodying users' intrinsic preferences and willingness to actively approach technologies. Technological competence belief refers to users' subjective expectations and self-efficacy regarding their ability to effectively understand, learn, and apply a specific technology to solve practical problems and seize potential opportunities in a given technical context. As the third core dimension of technology commitment, technological control belief defines users' subjective expectations regarding the extent to which they can control the technology usage process, address potential problems during usage, and achieve expected outcomes, concentrating on users' perceived controllability of technology usage (Neyer et al., 2012).

Among the technological innovation characteristics variables, mobility endows agricultural AI equipment with the ability to move autonomously and perform human-like tasks in various complex farmland environments, such as field walking, water navigation, and aerial flight, which greatly expands the application scope and flexibility of the technology. Autonomy refers to the ability of agricultural AI systems or robots to independently perceive the environment, make decisions, and execute specific agricultural tasks without (or with minimal) human intervention, instructions, or continuous control (Bekey, 2012). The organic combination of these technological innovation characteristics and technology commitment variables can more comprehensively reflect the unique value of agricultural AI and users' subjective perceptions of them, thereby further improving the theoretical explanatory power of the UTAUT model for testing the impact of innovative technological characteristics on the acceptance and usage intentions of new technologies.

Based on the aforementioned review and analysis of the technology acceptance model (TAM) and the unified theory of acceptance and use of technology (UTAUT), and taking full account of the characteristics of agricultural artificial intelligence technology and the actual circumstances of agricultural users, this study aims to construct an integrated theoretical framework to more systematically and in-depth examine the adoption intention of agricultural AI among agricultural users in Shandong Province. Specifically (as seen in Figure 1), this study plans to integrate the core mediating variables of perceived usefulness and perceived ease of use from the TAM, three core technology commitment-related variables (technology interest, technology competence belief, and technology control belief) derived from the extended UTAUT model, as well as two key technology innovation characteristic-related variables (mobility and autonomy) as the main exogenous variables of the research framework. Among these, perceived ease of use and perceived usefulness are designated as core mediating variables, responsible for transmitting the impacts of the aforementioned exogenous variables on the final outcome variable. Adoption intention, as the core outcome variable of this study, is used to measure the intensity of users' subjective intention to accept and use agricultural AI.

Furthermore, considering the heterogeneity of the agricultural user group, users with different demographic characteristics (such as educational background, farming styles, work experience, and usage frequency) may exhibit significant differences in their cognitive levels, learning abilities, risk preferences, and resource accessibility regarding new technologies. These differences may affect their evaluations of the perceived ease of use and perceived usefulness of agricultural AI, thereby moderating the intensity of the impacts of exogenous variables on mediating variables or the influence paths of mediating variables on the outcome variable. Therefore, this study further introduces demographic characteristics as potential moderating variables to test their moderating effects in the model, aiming to more elaborately reveal the formation mechanisms of adoption intention among different user groups.

Through the integration of the aforementioned variables and the setting of influence paths, this study ultimately constructs a more comprehensive and context-adaptive theoretical framework. It is expected to more comprehensively and in-depth analyze the complex influence mechanisms underlying the adoption intention of agricultural AI among agricultural users in Shandong Province, thereby providing a theoretical basis and practical implications for relevant policy formulation and technology promotion.

As a crucial enabling characteristic of agricultural AI, the core value of mobility lies in breaking the spatiotemporal constraints of traditional agricultural production. It allows farmers to access, learn, and apply AI anytime and anywhere through mobile terminals such as smartphones and wearable devices, thereby significantly reducing the physical barriers and initial obstacles to AI use (Elliott and Boyd, 2020). The inherent portability and contextual adaptability of mobile applications often imply more intuitive user interfaces and simplified operational processes, which directly facilitate the reduction of farmers' learning costs and operational complexity (Nah et al., 2005). This flexibility in technology access and convenience in operation can effectively enhance farmers' evaluations of perceived ease of use by improving the overall accessibility of agricultural AI. Meanwhile, mobility greatly expands the application boundaries of AI in agricultural scenarios, such as supporting remote expert collaborative diagnosis of plant diseases and pests and providing location-based real-time agricultural services (Paiva et al., 2021). These extended applications can strengthen farmers' recognition of the practical value of technical tools, making it easier for them to perceive the potential capabilities of AI in improving production efficiency, optimizing resource allocation, and meeting personalized needs (Ye et al., 2020), thereby enhancing their perceived usefulness of agricultural AI (Chinomona, 2013). Based on this, this study proposes the following hypotheses:

H1a: Mobility has a positive impact on perceived ease of use;

H1b: Mobility has a positive impact on perceived usefulness.

The concept of autonomy originally originated in the field of personality psychology to describe the degree of independence in an individual's thoughts and behaviors (Bublitz and Merkel, 2009). In recent years, it has been gradually extended to the fields of work tasks and organizational management and integrated into the theoretical frameworks of technological innovation and user behavior research (De Spiegelaere et al., 2016). As a key antecedent factor affecting job satisfaction, high performance, and service innovation capabilities (Demircioglu, 2021), autonomy in the application context of agricultural AI is mainly reflected in the independent decision-making and execution capabilities of technical systems, such as intelligent agricultural assistants and autonomous agricultural machinery. These autonomous technologies can significantly reduce users' manual operation steps and cognitive load by automating core operational processes (Beloev et al., 2021). For example, intelligent greenhouse temperature control systems can automatically adjust temperature and humidity according to preset algorithms and real-time environmental data, reducing the frequency and intensity of manual intervention by farmers, thereby enhancing their perceived ease of use of the technology; the autonomous inspection and precision spraying functions of agricultural drones improve operational accuracy and resource utilization efficiency by optimizing operational processes and reducing human operational errors (Moreno-Jacobo et al., 2021). This improvement in practical utility further enhances users' perceived usefulness of the technology (Yoon and Hwang, 2024). Therefore, this study hypothesizes:

H2a: Autonomy has a positive impact on perceived ease of use;

H2b: Autonomy has a positive impact on perceived usefulness.

As a key intrinsic motivational factor, technological interest profoundly drives individuals' willingness and behavioral tendencies to explore and adopt new technologies. When farmers hold a strong curiosity and exploratory desire toward agricultural AI itself, they are more likely to proactively invest time and energy in learning technical knowledge and overcoming potential learning barriers and usage anxiety (Hang, 2024). This intrinsic motivation triggered by interest can accelerate users' understanding and adaptation to the operational logic of technical interfaces and reduce their subjective perception of technical complexity, thereby enhancing the perceived ease of use of agricultural AI. Furthermore, farmers with strong technological interest often exhibit greater initiative in exploring and expanding the potential application value of technology: Technology enthusiasts may discover the efficiency improvement space or innovative application models of AI tools in specific agricultural scenarios earlier and actively integrate them into existing workflows, or even explore new application scenarios. This process of in-depth use and value excavation will significantly strengthen their perceived usefulness of the technology (Kim et al., 2021). Therefore, this study proposes

H3a: Technological interest has a positive impact on perceived ease of use;

H3b: Technological interest has a positive impact on perceived usefulness.

Technological competence belief includes users' subjective evaluation and confidence in their own digital skills, relevant prior knowledge, and ability to learn new technologies, which directly determines users' learning efficiency and usage effectiveness when facing new technologies. Generally speaking, users with higher technological competence belief, due to their existing knowledge reserves and learning abilities, can more quickly understand and master the operational processes and core functions of agricultural AI tools, thereby effectively reducing cognitive barriers and frustration during the learning process and enhancing the perceived ease of use of the technology (Ardiyanti and Susilowati, 2024a). Moreover, such users are more inclined to explore and utilize advanced functions of the technology or conduct personalized settings. This proactive exploration and in-depth use behavior, in turn, further optimizes their perceived ease of use of the technology (Adenekan et al., 2023). In addition, technological competence belief indirectly enhances users' perceived usefulness of agricultural AI in increasing yields and optimizing management by strengthening their sense of control and self-efficacy during technology use. Thus, this study hypothesizes

H4a: Technological competence belief has a positive impact on perceived ease of use;

H4b: Technological competence belief has a positive impact on perceived usefulness.

Technological control belief specifically refers to farmers' subjective perception and confidence in their ability to effectively control the operational processes, outcomes, and inherent settings of agricultural AI during its use. When users perceive that they can exercise effective control over the technical system—such as customizing the operation interface according to personal habits, flexibly adjusting functional parameters, or modifying permission settings—their sense of uncertainty and frustration during use is significantly reduced, thereby directly enhancing the perceived ease of use of the technology (Lee et al., 2007). For instance, allowing farmers to adjust the sensitivity threshold of AI pest and disease identification models based on their own experience and plot characteristics, or to customize the presentation format of data reports, can both strengthen their sense of control and improve their ease-of-use experience. More importantly, users with a higher technological control belief can more confidently and flexibly adjust and configure technical functions according to the specific needs of actual agricultural production scenarios, enabling technical tools to better align with their unique production processes and management objectives, thereby maximizing the practical benefits brought by technology application. This customized application and benefit improvement will significantly enhance their perceived usefulness of the technology (Renko and Druzijanic, 2014). For example, experienced farmers can obtain decision support information more tailored to local realities by independently configuring key parameters of AI farmland data analysis tools, which undoubtedly strengthens their recognition of the decision support value of AI. Therefore, this study proposes

H5a: Technological control belief has a positive impact on perceived ease of use;

H5b: Technological control belief has a positive impact on perceived usefulness.

Perceived ease of use is one of the core constructs in the technology acceptance model. In the context of this study, it reflects farmers' subjective evaluation of the cognitive and physical efforts, such as time and energy, required to learn and use specific agricultural AI (Kai et al., 2025). Its connotation extensively involves multiple dimensions, including the complexity of the operational logic of AI tools, the steepness of the learning curve, and the compatibility with users' existing knowledge systems and workflows. When farmers generally believe that the learning process of a certain agricultural AI is simple and understandable, the operational steps are intuitive and clear, and the required time and energy costs are within an acceptable range, their willingness to try and ultimately adopt the technology will be significantly enhanced (Okai et al., 2024). For example, a simplified operation interface integrated with voice control and dialect recognition functions designed for elderly or low-education farmer groups, as well as supporting simplified technical training courses targeting specific AI tools, can effectively lower the threshold of perceived ease of use, thereby improving their adoption intention. Therefore, this study hypothesizes

H6: Perceived ease of use has a positive impact on adoption intention.

Perceived usefulness is also a core concept in technology acceptance research, referring to users' subjective recognition of the extent to which the use of agricultural AI can help them improve work efficiency, reduce production costs, increase economic benefits, or achieve other important work objectives (Yousra and Khalid, 2021). When farmers truly perceive the multi-dimensional benefits brought by agricultural AI through direct experience or indirect information acquisition—such as yield improvement achieved through precision fertilization AI systems (Adewusi et al., 2024), reduction in pesticide waste and environmental damage based on intelligent pest and disease early warning models (Li and Wang, 2024), avoidance of extreme climate losses using AI-driven refined weather forecasts (Zidan and Febriyanti, 2024), and reduction in labor intensity and savings in labor costs achieved with agricultural picking robots (Gupta, 2024)—their recognition of the value of the technology will be strengthened, which in turn translates into a stronger adoption intention. Based on this, this study proposes

H7: Perceived usefulness has a positive impact on adoption intention.

As a core contextual identifier of agricultural production, differences in agricultural types directly determine the demand orientation and adaptation logic of agricultural AI application (Guo et al., 2023). Distinct agricultural types exhibit significant divergences in their core demands for technological functions (Spanaki et al., 2022): the needs of crop farming for technologies such as precision irrigation and pest monitoring (Bayar et al., 2025); the reliance of fruit and forestry on growth cycle regulation and quality inspection tools (Taneja et al., 2023); the special requirements of animal husbandry and aquaculture for environmental monitoring and automated feeding (Mandal and Ghosh, 2024); and the urgent demand of agricultural enterprises for large-scale data management and intelligent decision-making systems (Mishra and Mishra, 2024). All these factors influence practitioners' value judgments on the innovative characteristics of agricultural AI, such as autonomy and mobility. Such contextual demand differences further moderate the strength of the relationship between technological innovation characteristics, and perceived usefulness/ease of use, resulting in differentiated impact effects of the same technological attribute across different agricultural types.

Usage frequency reflects the depth of user interaction with technology and directly embodies technological familiarity and usage inertia (Topsakal, 2025). High-frequency users accumulate rich operational experience through long-term practice, gain a clearer understanding of the functional boundaries and optimization potential of technology (Nizamani et al., 2024), form more rational perceptions of perceived ease of use, and make judgments on perceived usefulness that are more aligned with actual production needs—these factors may weaken or strengthen the impact of technological innovation characteristics on perceptual variables. In contrast, low-frequency users are still in the initial stage of technological cognition, and their perceptions are easily influenced by initial experiences and surface-level functions (Nizamani et al., 2024), making the driving effect of technological commitment factors on perceptual variables potentially more significant. Meanwhile, path dependence formed by usage frequency also moderates the efficiency of converting perceived usefulness and perceived ease of use into adoption intention.

Educational background and work experience, as core dimensions of human capital, profoundly affect users' technological learning capabilities and contextual application capabilities (Kai et al., 2025; Zhao et al., 2024). Users with higher educational levels possess stronger information processing and abstract thinking abilities, enabling them to understand technological logic more efficiently, thereby amplifying the positive impact of technological interest on perceived ease of use. Rich work experience helps users integrate technological capability beliefs with specific production scenarios, accurately identify technological adaptation points, and strengthen the promoting effect of technological control beliefs on perceived usefulness. Both factors also moderate the relationship between perceptual variables and adoption intention by shaping users' value judgment criteria (Abaddi, 2025; Jain et al., 2025). Therefore, we propose

H8: Agricultural type, usage frequency, educational background, and work experience partially moderate the relationships between mobility, autonomy, technological interest, technological capability belief, technological control belief and perceived usefulness/perceived ease of use.

H9: Agricultural type, usage frequency, educational background, and work experience partially moderate the relationships between perceived usefulness/perceived ease of use and adoption intention.

This study adopts a three-stage progressive statistical analysis framework to systematically test the theoretical model and research hypotheses. The specific process is as follows. Stage 1: measurement model testing–confirmatory factor analysis is employed to comprehensively assess the convergent validity and discriminant validity of the scale, ensuring the rationality and stability of the measurement relationship between latent variables and observed variables, thereby laying a measurement foundation for subsequent analyses. Stage 2: structural model analysis–partial least squares structural equation modeling (PLS-SEM) is used to estimate the path relationships among core variables. The Bootstrap sampling method (5,000 repetitions) is applied to test the significance of path coefficients, further verifying research hypotheses and revealing the driving mechanism of farmers' adoption intention toward agricultural artificial intelligence technology. Stage 3: heterogeneity analysis–the multi-group analysis (MGA) method is adopted to focus on examining the differences in technology adoption paths between farmer and non-farmer groups, deeply exploring the impact of individual heterogeneity on the technology acceptance process and providing empirical evidence for formulating differentiated technology promotion strategies.

In structural equation modeling analysis, measurement model testing is a critical step to ensure the rationality and theoretical consistency of the relationships between latent variables and observed variables. This study conducts empirical analysis using partial least squares structural equation modeling (PLS-SEM), a method particularly suitable for handling complex models and small-sample data. It effectively controls for measurement errors and verifies theoretical hypotheses. The measurement model testing in this study focuses on three dimensions. Reliability testing: Internal consistency indicators [e.g., Cronbach's α and composite reliability (CR)] are used to assess the stability of observed variables in measuring latent variables. Validity testing: This includes convergent validity [measured by indicators such as factor loadings and average variance extracted (AVE)] and content validity, ensuring that observed variables fully reflect the theoretical connotations of latent variables. Discriminant validity: This is verified by comparing the correlation coefficients between latent variables and the square root of AVE, confirming the independence of measurements for different latent variables. This systematic testing process consolidates the measurement foundation for subsequent structural model analysis and enhances the credibility of the research conclusions.

After completing the reliability and validity tests of the measurement model, this study employs partial least squares structural equation modeling (PLS-SEM) to conduct empirical analysis on the path relationships among latent variables in the theoretical model. The core objectives of structural model assessment include hypothesis verification, relationship analysis, and effectiveness evaluation. Hypothesis verification assesses the statistical rationality of research hypotheses through significance tests of path coefficients; relationship analysis reveals the mechanism of action between variables by quantifying the direct effect among latent variables; effectiveness evaluation comprehensively judges the explanatory power and predictive power of the model based on R² and predictive relevance indicators (e.g., Q²). Compared with traditional structural equation modeling (SEM), PLS-SEM adopts a variance maximization-oriented iterative calculation method, which can more flexibly handle complex models and non-normal data, and is particularly suitable for predictive analysis with small samples or multicollinearity.

To mitigate potential sampling bias arising from online surveys, this study adopts the inverse probability weighting (IPW) method for robustness testing. Taking educational level, age, and planting scale as covariates, the inverse of the sample inclusion probability estimated by the logistic regression model is used as the weight to re-test the path coefficients of the structural equation model. The results are presented in Table 12.

The results of the IPW analysis indicate that the difference between the weighted path coefficients and the original sample coefficients is small, and the significance levels of all significant paths remain stable. This suggests that the impact of sampling bias in this study on the core conclusions is limited, and the research results have good robustness. The bias is mainly reflected in the slightly higher proportion of farmers with higher educational levels and medium-to-large planting scales in the sample compared to the overall population, but their interference with the model's path relationships has not changed the core research conclusions.

To further reveal the heterogeneous characteristics of different professional role groups in the adoption path of agricultural AI, this study divided the samples into three groups based on professional roles: farmers (375), agricultural material dealers (35), and agricultural experts (15). A multi-group analysis (MGA) was used to test the significant differences in path coefficients between groups, and the results are shown in Table 13. The core differences are mainly reflected in the influence path of perceived ease of use on perceived usefulness: the path coefficient of this path in the agricultural material dealer group (0.875, p < 0.001) was significantly higher than that in the farmer group (0.491, p < 0.001; p = 0.007). This indicates that dealers' perception of technology ease of use is more likely to be converted into recognition of usefulness, which may be related to their professional attribute of paying more attention to technology promotion efficiency. In addition, there was a marginal inter-group difference in the path of technology interest on perceived usefulness: The difference between the farmer group (0.155, p < 0.05) and the dealer group (−0.114, ns) was significant (p = 0.042). This reflects that farmers' technology interest is more likely to be converted into practical value perception, while dealers may be affected by market promotion pressure, resulting in a weak correlation between interest and value perception. Other core paths (such as the path of mobility on perceived usefulness, and perceived usefulness on adoption intention) showed no significant differences among the three groups (p > 0.05), indicating that the value of technology mobility and the driving effect of core perceived variables on adoption intention are universally applicable across groups.

The moderating effect analysis aims to reveal how individual heterogeneous variables influence the path relationships between core constructs. This study adopted a two-stage method to examine the moderating roles of educational background, work experience, agricultural type, and usage frequency (see Table 14). An extended model with interaction terms was constructed, and a Bootstrap sampling test (5,000 repeated samplings) was used to verify the significance of the moderating effects. After incorporating the moderating variables, the explanatory power of the model for perceived ease of use and perceived usefulness increased from 0.595 and 0.573 to 0.641 and 0.638, respectively. This indicates that these individual characteristic variables can effectively supplement the explanatory dimensions of the model, further confirming the important impact of the heterogeneity of agricultural user groups on the technology acceptance path.

Educational background showed a differentiated moderating effect on the relationship between perceived variables and adoption intention. The interaction term coefficient between educational background and perceived usefulness was −0.193 and significant (p < 0.01), meaning that farmers with a higher educational level have stronger rational screening capabilities in the conversion process from perceived usefulness to adoption intention. They not only focus on the direct benefits of technology but also comprehensively consider the long-term adaptability and potential risks of technology, thereby weakening the driving effect of pure perceived usefulness on adoption intention. However, the interaction term coefficient between educational background and perceived ease of use was 0.174 and significant (p < 0.05), indicating that farmers with a higher educational level, relying on better learning abilities, can more efficiently convert the positive perception of technology ease of use into actual adoption intention. The convenience of technology operation has a more prominent impact on their decision-making. This difference reflects the multi-dimensional consideration characteristics of farmers with higher educational levels in technology adoption decisions, and their value judgment system is more complete.

Work experience had a significant positive moderating effect on the path between technological competence belief and perceived ease of use, with an interaction term coefficient of 0.058 (p < 0.05). For farmers with rich work experience, the long-accumulated production practice experience enables them to combine abstract technological competence beliefs with specific field scenarios, accurately identify the fit between technology operation and actual production, and thus more effectively convert competence beliefs into positive perceptions of technology ease of use. In contrast, farmers with insufficient experience, due to the lack of practical references, find that their technological competence beliefs often remain at the theoretical level and are difficult to convert into convenient experiences in actual operations. This result highlights the key bridging role of practical experience in the implementation of technological capabilities and provides a basis for formulating differentiated training strategies for farmers with different experience levels.

Agricultural type had a significant moderating effect on the relationship between autonomy and perceived usefulness, with an interaction term coefficient of 0.125 (p < 0.05). There are essential differences in the demand for technological autonomy among different agricultural production types. Large-scale agricultural enterprises or contiguous farmers have a more urgent demand for autonomous technology, and its autonomous operation function can effectively improve production efficiency. However, small-scale farmers with scattered operations, as well as practitioners in the fruit forestry and aquaculture industries, face insufficient adaptability of existing autonomous technologies due to fragmented production scenarios and diversified operation processes, resulting in a limited role of autonomy in improving perceived usefulness. This finding reveals the important impact of the adaptability between technological innovation characteristics and agricultural production scenarios on users' value perception and provides a direction for the scenario-based optimization of technology.

Usage frequency exhibited a significant negative moderating effect on the path between mobility and perceived usefulness, with an interaction term coefficient of −0.189 (p < 0.01). High-frequency users have formed a more comprehensive understanding of technology functions through long-term practice. Their judgment on the value of mobility is no longer limited to superficial convenience but focuses more on the stability and in-depth application value of technology in complex scenarios, thereby weakening the positive impact of mobility on perceived usefulness. In contrast, low-frequency users are in the initial stage of technology cognition, and the basic convenience experience brought by mobility has a more significant impact on their value perception. This moderating mechanism reflects the dynamic impact of users' technology familiarity on value judgment, suggesting that technology promotion should formulate differentiated value communication strategies according to the stages of users' usage frequency.

To further accurately identify the boundary conditions of moderating effects, this study employed the Johnson–Neyman technique to conduct supplementary analysis on the moderating effects of five hypotheses. The Johnson–Neyman technique can identify the range of values of the moderating variable within which the impact of the independent variable on the dependent variable reaches statistical significance. Compared with traditional simple slope analysis, this method provides more precise conditional boundary information, which is helpful for understanding the mechanism and practical application value of moderating effects.

Technological innovation characteristics (mobility, autonomy) and technology commitment (technology interest, technology control belief) exerted significant positive impacts on perceived ease of use, with technology interest being the most prominent driver. This highlights that farmers' intrinsic motivation to explore technology and sense of operational control are core to lowering usage barriers. Mobility breaks agricultural production's spatiotemporal constraints, enabling on-demand access via mobile terminals; autonomy reduces manual intervention through automation—both jointly simplifying operations. Technological interest stimulates learning motivation, while technological control belief alleviates adoption anxiety.

This finding is partially consistent with international studies on agricultural AI adoption. For instance, Caffaro et al. (2020) found that technological convenience and intrinsic interest are key drivers of perceived ease of use among Italian farmers adopting smart farming technologies, which aligns with the positive impact of mobility and technological interest identified in this study. However, a notable difference exists: Italian farmers, who are more focused on large-scale commercial agriculture, prioritize the compatibility of autonomous technologies with industrialized production processes, while Chinese smallholder farmers in Shandong emphasize the flexibility of autonomous functions to adapt to scattered plots. This divergence reflects the influence of agricultural production models on technology perception across regions. These results uphold the foundational role of perceived ease of use in the technology acceptance model (TAM) and, by integrating agricultural AI's unique attributes, address the contextual gap of traditional technology acceptance theories in agricultural scenarios, reflecting the sector's specific demands for technology adaptability and convenience.

Mobility, technology interest, and perceived ease of use also significantly drove perceived usefulness. Mobility expands technology application boundaries, reinforcing farmers' recognition of practical value; technological interest prompts active exploration of efficiency gains and innovative use cases; and the strong correlation between perceived ease of use and perceived usefulness confirms TAM's classic proposition. As a core mediator, perceived ease of use directly influences adoption intention and indirectly shapes it by enhancing perceived usefulness. Notably, perceived usefulness had a slightly stronger driving effect on adoption intention than perceived ease of use, underscoring farmers' focus on production efficiency in adoption decisions.

This conclusion is in line with research on U.S. agricultural producers by Chen et al., who found that perceived usefulness related to production efficiency is the primary determinant of AI adoption intention among American farmers (Chen et al., 2023). However, U.S. farmers pay more attention to the long-term economic returns of AI technologies, while Shandong farmers, with a higher proportion of small-scale operations, focus more on short-term practical benefits. Additionally, Mohr and Kühl (2021) noted that European farmers' perceived usefulness of agricultural AI is often intertwined with environmental protection goals, whereas Chinese farmers' perceived usefulness is more economically oriented, which also reflects differences in policy guidance and production concepts across regions. This further verifies TAM's applicability to agricultural AI while clarifying the unique paths of technological attributes and individual psychology in agricultural contexts.

Several unsupported hypotheses revealed critical practical dilemmas. Autonomy's non-significant impact on perceived usefulness stems from a misalignment between current autonomous technologies (designed for large-scale, standardized production) and Shandong's dominant small-scale farmers, who face scattered plots and diverse operations. Rigid processes and high adjustment costs of existing autonomous systems even burden smallholders, reducing perceived value.

This phenomenon differs from findings in countries with mature large-scale agriculture. For example, Ardiyanti and Susilowati (2024b) found that autonomy has a significant positive impact on perceived usefulness among Australian grain farmers, as their large-scale, contiguous farmland is highly compatible with autonomous agricultural machinery. In contrast, the fragmented production model of Chinese smallholders limits the practical value of autonomous technologies, which is a key contextual factor that international studies often overlook.

Technology competence belief's non-significant negative impact on perceived ease of use and usefulness reflects two issues: Farmers overestimate their digital skills, leading to negative perceptions when actual performance falls short; and training systems prioritize theory over field-based practical training, hindering the conversion of competence beliefs into operational capabilities. Additionally, technology control belief's non-significant effect on perceived usefulness indicates that farmers' operational control does not translate to value recognition—rooted in a poor matching between technology functions and real production needs. These findings refine the understanding of agricultural AI adoption barriers and extend TAM-UTAUT's application boundary by identifying contextual constraints.

Individual heterogeneity further differentiated technology acceptance paths. Higher education enabled farmers to convert perceived ease of use into adoption intention more effectively but induced rational screening of perceived usefulness, reflecting a more comprehensive value judgment system. Rich work experience facilitated the conversion of technology competence belief into perceived ease of use, highlighting practical experience as a bridge for translating technical capabilities into usability perceptions.

The moderating effect of educational background is consistent with cross-national research by Gikunda (2024), who found that higher education strengthens the conversion of perceived ease of use to adoption intention among farmers in both developing and developed countries. However, Gikunda (2024) noted that in developed countries such as Germany, higher education also enhances farmers' risk perception of AI technologies, leading to more cautious adoption decisions, whereas Shandong farmers with higher education mainly strengthen rational screening of practical value, with less emphasis on technological risks. This difference may be attributed to the relatively immature risk prevention system of agricultural AI in China compared to developed countries.

Agricultural type moderated autonomy's impact on perceived usefulness: large-scale operations valued autonomous technology more, emphasizing the importance of aligning innovation with production scenarios. Usage frequency had a negative moderating effect on mobility's impact on perceived usefulness—high-frequency users judged mobility more rationally (focusing on stability over surface convenience), while low-frequency users prioritized basic convenience. These results supplement boundary condition research for agricultural technology adoption models, enriching the theoretical understanding of heterogeneous user behavior.

For technology R&D, efforts should prioritize small-scale farmers' needs by developing modular, scenario-specific intelligent systems—e.g., flexible autonomous functions adaptable to scattered plots and diverse operations—to resolve adaptability issues. For training, an integrated “theory + practice” system with regular field support is essential, replacing pure theoretical instruction with hands-on, production-linked training to help farmers translate competence beliefs into practical skills. For promotion, differentiated strategies are needed: target low-education farmers with simplified operational guidance to enhance perceived ease of use; provide experienced farmers with advanced skill training to leverage their competence beliefs; emphasize autonomous technology's value for large-scale operations; and highlight mobility's convenience for low-frequency users while addressing high-frequency users' demands for stability and in-depth functions.

This study's key theoretical contribution lies in constructing an agricultural AI adoption intention model that integrates technological innovation characteristics and technology commitment, expanding TAM-UTAUT's application scope. By revealing the complex moderating mechanism of individual heterogeneity, it addresses the contextualization deficit of traditional technology acceptance theories in agriculture, providing a more nuanced framework for understanding agricultural AI adoption intention. Moreover, by comparing with international studies, this study clarifies the contextual differences in agricultural AI adoption mechanisms between China and other countries, enriching the cross-cultural application of technology acceptance theories.