Exploring Factors Influencing Patients’ Intention to Adopt Generative AI on Online Healthcare Platforms

Abstract

The development of generative AI has disrupted various fields, and the field of online healthcare is no exception. However, there is a lack of research on patients’ intention to adopt generative AI on online healthcare platforms. Therefore, the aim of this study is to investigate the factors influencing patients’ intention to adopt generative AI. Employing a questionnaire-based survey, we explore the factors influencing patients’ intention to adopt generative AI through the UTAUT2 model, considering the moderating effects of construal level, health literacy, and AI literacy. We find that performance expectancy, social influence, and facilitating conditions are positively associated with patients’ intention. Surprisingly, effort expectancy and hedonic motivation do not have a significant impact on patients’ intention. Construal level positively moderates the relationship between performance expectancy and patients’ intention; health literacy negatively moderates the relationship between social influence and patients’ intention. AI literacy positively moderates the relationship between effort expectancy and patients’ intention but negatively moderates the relationship between social influence and patients’ intention. This study enriches UTAUT2 theory and provides practical insights for the development and promotion of generative AI on online healthcare platforms.

1. Introduction

Generative artificial intelligence (AI) is reshaping and disrupting various industries [1,2], and the online healthcare is no exception. Embedded within online healthcare platforms, generative AI trains language models using real consultation data from these platforms, enabling rapid responses and answers to patient inquiries through advanced learning and reasoning capabilities. Generative AI holds significant value on online healthcare platforms [3]. Unlike automated response systems that leverage AI technologies to assist doctors in making quick diagnoses and responses [3,4], the generative AI developed for online healthcare platforms functions as an intelligent medical assistant for patients. By learning from extensive medical data and knowledge, it can answer health-related questions, provide medical consultations, and explain disease information in a patient-friendly manner [5]. Moreover, it can assist patients in making appointments, selecting doctors, and querying information, thus serving as a convenient, efficient, and friendly tool for online healthcare service that is accessible anytime and anywhere. Additionally, generative AI can enhance patient engagement by personalizing interactions based on individual health needs and preferences, ultimately improving the overall health awareness and health literacy of patients [6,7].

Despite the many benefits of generative AI for online healthcare services, their widespread adoption of online healthcare platforms and patients remains limited, and there is scant literature exploring patients’ intention to adapt generative AI online healthcare platforms. Previous research on generative AI in the field of healthcare has primarily focused on both its benefits and challenges. Existing research has extensively explored the role of generative AI in healthcare, including its support for diagnostic decision-making by physicians [8], assistance in clinical practice and research [9], enhancement of healthcare quality and efficiency [10], optimization of operations and management [5], and reinforcement of patient education and physician training [11]. There are also concerns regarding generative AI, such as the potential for generating inaccurate content, making inappropriate diagnoses [12], issues related to privacy and security [13,14], ethical dilemmas regarding patient understanding [15], and deficiencies in legal regulation. However, research on generative AI in online healthcare remains limited, and it is still unclear which factors influence patients’ adoption of generative AI on such platforms. Undoubtedly, the value that generative AI brings to healthcare cannot be denied; however, this value can only be realized if patients adopt generative AI. Therefore, exploring the factors influencing patients’ intention to adopt generative AI on online healthcare platforms is essential for promoting its broader application and maximizing its value. UTAUT2 is a well-established model for understanding technology acceptance and usage, and it has been widely applied in research on user adoption of new technologies [16,17]. In this study, generative AI is an emerging technology, patients constitute a new user group, and online healthcare represents a novel usage context. Therefore, this study employs the UTAUT2 model to investigate the factors influencing patients’ intention to adopt generative AI on online healthcare platforms.

Venkatesh et al. [18] points out that moderators are critical for meaningful extensions of theories of technology acceptance. For example, individual differences in age, gender, and experience moderate the effects of UTAUT2 on behavioral intention [18]. Therefore, examining boundary conditions in patients’ intention to adopt generative AI on online healthcare platforms is crucial, as it contributes to the extension of technology adoption theories and deepens the understanding of this specific context. Performance expectancy and effort expectancy correspond to attributes situated at different construal levels [19]. Performance expectancy, as an outcome-oriented variable, aligns with a higher construal level, in which individuals adopt an abstract mode of thinking that emphasizes causes and consequences. In contrast, effort expectancy, as a process-oriented variable, is associated with a lower construal level, wherein individuals engage in a concrete mode of thinking that highlights execution and processes [20]. Accordingly, the influence of performance expectancy and effort expectancy on the adoption of generative AI may differ across individuals with different construal levels. Social influence and facilitating conditions represent the key social and technological factors through which individual access and use health-related information. Health literacy determines both the motivation and the ability to acquire and utilize such information [21,22]. Therefore, the impact of social influence and facilitating conditions on the adoption of generative AI for health-related information is likely to vary across individuals with different levels of health literacy. Additionally, individuals with different levels of AI literacy demonstrate distinct AI-related behaviors. Specifically, variations in AI literacy shape individuals’ trust in and ability to use AI technologies [23]. Consequently, differences in AI literacy may lead to divergent perspectives on the adoption of generative AI. Taken together, this study aims to enhance the understanding of patients’ intention to adopt generative AI on online healthcare platforms by examining the moderating roles of construal level, health literacy, and AI literacy from the patient’s perspective.

In summary, this study aims to examine the factors influencing patients’ intention to adopt generative AI on online healthcare platforms and the moderating effect of construal level, health literacy, and AI literacy. We collect data through a questionnaire via online channels. The analysis reveals that performance expectancy, social influence, and facilitating conditions are positively associated with patients’ intention to adopt generative AI on online healthcare platforms, whereas effort expectancy and hedonic motivation do not exhibit a significant effect. Furthermore, the results show that the construal level positively moderates the relationship between performance expectancy and patients’ intention. Health literacy negatively moderates the relationship between social influence and patients’ intention. AI literacy positively moderates the relationship between effort expectancy and patients’ intention but negatively moderates the relationship between social influence and patients’ intention. These findings contribute to a more nuanced understanding of patients’ intention to adopt generative AI and offer practical implications for the effective design, implementation, and dissemination of such technologies in online healthcare contexts.

2. Literature Review

2.1. UTAUT2

The adoption of technology is a well-researched topic in the field of information systems. Numerous theories and models have been developed to identify the factors that influence technology adoption, such as the theory of reasoned action, the technology acceptance model, the motivational model, the theory of planned behavior, the model of PC utilization, the innovation diffusion theory, and the social cognitive theory. Venkatesh et al. [24] synthesized these theories and models to create a unified model, known as the unified theory of acceptance and use of technology (UTAUT), which encompasses four primary influences of performance expectancy, effort expectancy, social influence, and facilitating conditions. Later, Venkatesh et al. [18] extended the UTAUT model to consumer behavior research by incorporating three additional influences of hedonic motivation, price value, and habit and proposed the unified theory of acceptance and use of technology 2 (UTAUT2). In general, the factors that influence technology adoption include technology-related variables, user characteristics, and usage environment, which together constitute the information system scenario [19,25]. These aspects are also critical for the wide-ranging application of UTAUT2 to new technologies (e.g., generative AI), new users (e.g., patients), and new usage contexts (e.g., healthcare settings). Additionally, the incorporation of new elements and structures in the UTAUT2 model can enhance its explanatory power and expand its theoretical boundaries.

Using generative AI represents a unique form of consumption behavior and falls within the UTAUT2 research scenario. Given that generative AI involves new technologies, patients, and usage environments, we aim to explore the factors that influence patients’ adoption through the UTAUT2 model. Due to the limited widespread use of generative AI, patients have not yet developed usage habits, and generative AI is provided as a free feature. Therefore, habit and price value are not considered in the model. Additionally, this paper examines the impact of patient characteristics, such as the construal level and health literacy, and AI literacy on the adoption of generative AI to explore the boundary conditions of this technology adoption model.

2.2. Construal Level

The perspective of the construal level theory posits that individuals represent information at different levels of abstraction and embodiment [26]. Individuals with a high construal level tend to focus on abstract, superordinate, decontextualized, and non-textual features that convey the “why” aspects of things and desirability. Conversely, individuals with a low construal level tend to focus on concrete, subordinate, contextualized, and textual features that convey the “how” aspects of things and feasibility [26]. In other words, individuals with a high construal level are more concerned with the causes and consequences, whereas those with a low construal level are more concerned with the execution and processes [27].

The construal level theory posits that individuals represent things at varying levels of embodiment or abstraction, which is influenced by both situational and individual factors [28]. On the one hand, construal level is related to psychological distance, which includes temporal distance, spatial distance, social distance, and hypothetical distance [29,30,31]. When the mental distance is farther, people tend to focus more on abstract features, whereas when it is closer, people tend to focus more on concrete features [20]. Several studies have demonstrated that situational factors such as mental distance significantly affect the construal level. On the other hand, Vallacher and Wegner [32] proposed the action identification theory, suggesting individual differences in action identification levels, with lower levels of identities focusing on how things are performed and more concrete representations and higher levels of identities focusing on the causes or consequences of things and more abstract representations. Vallacher and Wegner developed the Behavior Identification Form (BIF) to measure the individual action identification level, which many studies have used to assess the action identification level [28,32]. The notion of construal level aligns with the concept of action identification level and is a personal attribute varied across individuals, except that construal level has a broader meaning scope and stronger explanatory power [30].

Based on the construal level theory, distinct attributes characterize the construal levels of performance expectancy and effort expectancy. Furthermore, individuals with varying construal levels exhibit differential impacts of their respective performance expectancy and effort expectancy on their acceptance intention to a given product or service. Therefore, the present study incorporates the patient’s construal level as a moderator variable to investigate the boundary conditions that delimit the relationship between performance expectancy and effort expectancy expectations and their effect on patients’ intention of use.

2.3. Health Literacy

Health literacy is a multifaceted concept that refers to an individual’s or group’s competencies in maintaining and promoting their health [33]. The World Health Organization (WHO) definition of health literacy is widely recognized and asserts that it comprises an individual’s cognitive and social skills that determine their motivation and capacity to access, comprehend, and use health-related information and to enhance and maintain their well-being [21]. Individuals with high levels of health literacy exhibit rational judgment and decision-making abilities as well as cognitive and social skills that enable them to solve problems and enhance their health status [34]. On the contrary, low-health-literate individuals tend to exhibit poorer health status and higher healthcare costs due to their lack of these skills [35], resulting in increased hospitalizations, emergency room visits, and mortality risks [36].

In the field of healthcare, there is evidence that individuals with varying levels of health literacy exhibit different health-related behaviors, and the moderating effect of health literacy has been explored. For example, Virlée et al. [22] found that online health community members with low health literacy do not effectively integrate and utilize online health resources compared to those with high health literacy. Similarly, Chen et al. [37] noted that health literacy can moderate the social and knowledge relationship between doctors and patients in online physician choices. Mackert et al. [21] also reported that individuals with high health literacy tend to have higher perceived usefulness and perceived ease of use of health information technology as well as lower privacy perceptions and higher trust in such technology. Therefore, in the investigation of patient behavioral intentions, it is pertinent to take into account the individual factor of health literacy, which can serve as a significant boundary condition.

2.4. AI Literacy

AI literacy denotes an individual’s proficiency in AI, encompassing a grasp of its fundamental concepts, technical intricacies, effective methodologies for its creation and deployment, and awareness of the governing legal frameworks [23]. An increasing number of scholars recognize the significance of AI literacy [38]. In the education sector, AI literacy enhances learners’ ability to utilize generative AI for problem-solving, underscoring the importance of fostering AI literacy [39]. Similarly, in the healthcare industry, improving AI literacy facilitates the effective application of AI by medical professionals [40]. Numerous researchers investigate measurement methods of AI literacy. Celik [41] argues that digital divide, cognitive absorption, and computational thinking significantly impact AI literacy. Some scholars measure AI literacy based on subjective understanding of AI technologies [42], while others assess it through objective knowledge of AI [23]. Existing studies indicate that AI literacy positively influences perceptions of ease of use and usefulness, thereby enhancing the overall acceptance of AI technologies. However, there is a divergence in findings: some research suggests that patients with higher AI literacy exhibit greater acceptance of AI [43], while others propose that patients with lower AI literacy demonstrate higher acceptance [23]. Nevertheless, there is a notable gap in research exploring the moderating role of AI literacy, particularly in relation to AI acceptance. Therefore, this study aims to investigate the moderating effect of AI literacy to address this research gap.

3. Research Hypothesis

3.1. Influencing Factors

In the realm of healthcare service, performance expectancy refers to the degree to which patients perceive that utilizing a particular healthcare system will help them achieve their overall health goals [24,44]. Generative AI can explain medical conditions to patients, answer health-related questions, assist in health management, and provide medical consultations, thereby supporting patients in achieving their health goals [14]. Effort expectancy denotes the ease with which patients can employ healthcare technologies [44]. Generative AI are designed to be patient-friendly and easily operable. It possesses communication and interaction capabilities similar to those of a physician, allowing patients to easily engage with it in a conversational manner to address their concerns [45]. Social influence pertains to the extent to which important individuals in one’s social network direct individuals towards utilizing technology to keep healthy [24]. In particular, family members and friends who express concerns regarding the well-being of individuals recommend and propose the use of generative AI [46,47]. Additionally, patient groups present on social media platforms possess informational and normative effects on individuals’ adoption of generative AI [48]. Facilitating conditions, which relate to patients’ perceptions of the availability of resources and support for technology use [24], are bolstered by the popularity of mobile Internet and Internet applications, which ensure access to generative AI. Finally, hedonic motivation refers to the pleasure or enjoyment derived from using generative AI [18]. Using generative AI, patients have an opportunity to attain more social support by obtaining timely and no-cost online healthcare service [49]. Additionally, the novelty of generative AI could enhance patient engagement and enjoyment. Summarizing the above discussion resulted in the following hypotheses:

H1: 

Performance expectancy is positively associated with patients’ intention to adopt generative AI.

H2: 

Effort expectancy is positively associated with patients’ intention to adopt generative AI.

H3: 

Social influence is positively associated with patients’ intention to adopt generative AI.

H4: 

Facilitating conditions are positively associated with patients’ intention to adopt generative AI.

H5: 

Hedonic motivation is positively associated with patients’ intention to adopt generative AI.

3.2. Moderation Effects

According to construal level theory, different individuals can represent an object or event at different construal levels [50]. Individuals with a higher construal level tend to adopt an abstract cognitive style that emphasizes outcomes and desirability, whereas those with a lower construal level rely on a concrete cognitive style that focuses on processes and feasibility [28]. When patients consider adopting generative AI, performance expectancy refers to the extent to which using such technology is perceived to provide beneficial outcomes, an outcome-focused evaluation aligned with purpose and desirability. This type of evaluation is consistent with the cognition of individuals with a higher construal level [30,51]. Therefore, performance expectancy is expected to have a stronger impact on adoption intention for individuals with a higher construal level compared with those at a lower construal level [20]. In contrast, effort expectancy pertains to the perceived ease of use of generative AI and reflects a feasibility-oriented evaluation centered on task execution, which aligns with the cognition of individuals with a lower construal level [18,24]. Accordingly, for individuals with a higher construal level, effort expectancy is expected to have a weaker impact on adoption intention compared with those with a lower construal level. Based on these theoretical insights, the following hypotheses are proposed:

H6a: 

The construal level strengthens the relationship between performance expectancy and patients’ intention so that the relationship is stronger when the construal level is higher.

H6b: 

The construal level weakens the relationship between effort expectancy and patients’ intention so that the relationship is weaker when the construal level is higher.

According to the definition of health literacy provided by the World Health Organization (WHO) [21], individuals with low health literacy that have a reduced ability to access, understand, and apply basic health information and services and make informed decisions related to their health [33] rely heavily on medical professionals with specialized knowledge and training [37]. The literature has established a significant association between low health literacy and unfavorable consequences, including but not limited to suboptimal health status, escalated hospital admissions, and elevated healthcare expenditure, arising from the inability to proficiently utilize health-related information [22,35]. Conversely, individuals who exhibit high health literacy exhibit a comprehensive understanding of health-related knowledge, possess requisite skills, and demonstrate the necessary consciousness to effectively manage their health and overall well-being [22,52]. Therefore, individuals with lower health literacy may be more susceptible to social influence when considering generative AI due to their limited healthcare knowledge and self-awareness compared with those with higher health literacy. Additionally, individuals with lower health literacy may have a decreased ability to access, comprehend, and integrate healthcare-related resources and technologies compared with those with higher health literacy, resulting in a lower role of facilitating factors in promoting the use of generative AI. Therefore, the following hypotheses are proposed:

H7a: 

Health literacy weakens the positive relationship between social influence and patients’ intention so that the relationship is weaker when health literacy is higher.

H7b: 

Health literacy strengthens the positive relationship between facilitating conditions and patients’ intention so that the relationship is stronger when health literacy is higher.

AI literacy refers to an individual’s comprehensive understanding of the knowledge and technology of AI, the methods of its development and application, and the rules that need to be adhered to [23]. Consequently, compared to patients with lower AI literacy, those with higher AI literacy possess a deeper understanding of AI technology, a greater awareness of its value, and stronger operational capabilities. Therefore, patients with high AI literacy have a better grasp of the capabilities of generative AI, thereby enhancing the effect of performance expectancy on their intention to adopt; they are more familiar with the operational methods of generative AI, enabling them to use it with more ease, which in turn boosts the impact of effort expectancy on their intention to use it. AI literacy allows patients to better utilize technical support resources, thereby strengthening the influence of facilitating conditions on their intention to adopt it; patients with higher AI literacy may enjoy the interaction process with generative AI more, deriving pleasure from the exploration process, thus enhancing the effect of hedonic motivation on their intention to adopt it. Patients with higher levels of AI literacy are more capable of independently evaluating the utility and reliability of generative AI. As such, they are less likely to be influenced by doctors, friends, and social media [53]. This suggests that social influence exerts a weaker effect on adoption intention among patients with higher AI literacy. Based on this, we propose the following hypotheses:

H8a: 

AI literacy positively moderates the relationship between performance expectancy and patients’ intention.

H8b: 

AI literacy positively moderates the relationship between effort expectancy and patients’ intention.

H8c: 

AI literacy positively moderates the relationship between facilitating conditions and patients’ intention.

H8d: 

AI literacy positively moderates the relationship between hedonic motivation and patients’ intention.

H8e: 

AI literacy negatively moderates the relationship between social influence and patients’ intention.

The conceptual framework in this study is presented in Figure 1.

4. Research Methodology

4.1. Research Sample and Data Collection

To test the research hypotheses, this study employed a questionnaire as the primary instrument for data collection, administered through online channels via the Credamo platform. Given that generative AI is predominantly accessed and utilized via the Internet, the use of online distribution was considered appropriate. To obtain a random sample, the questionnaire was randomly distributed through the data marketplace of the Credamo platform, which allowed access to a broad and diverse pool of potential respondents. Therefore, data collected through these channels are regarded as broadly representative and reasonably random for the context of this study. The data collection was conducted in China during September 2023.

The research questionnaire encompassed three components. Firstly, the introduction section explained the purpose of the questionnaire and provided information about informed consent for participants. As shown in Appendix A. It also included a filter question, “Do you know generative AI on online healthcare platforms?”, to exclude respondents who are unaware of this feature. Secondly, a survey was conducted to examine the influencing factors, patients’ intention, and boundary conditions. The questions employed a 7-point Likert scale, ranging from “strongly disagree” to “strongly agree,” and were adapted from established scales in the literature. Lastly, demographic variables, including gender, age, education, and income, were measured and used as control variables. Details are provided in Appendix A.12. Moreover, we set the attention check in Appendix A.6 and Appendix A.9. A total of 462 questionnaires were collected online. After excluding invalid responses, 305 valid questionnaires remained, resulting in a response rate of 66%. To ensure a random sample and control for the influence of demographic characteristics on the results, demographic information was measured. The demographic characteristics of the sample are presented in

Table 1. Demographic characteristics.

Name	Category	Freq	Percentage	Name	Category	Freq	Percentage
Gender	Male	122	40	Education	Less than undergraduate	36	11.8
	Female	183	60		Undergraduate	232	76.1
Age	18~25	104	34.1		Postgraduate or above	37	12.1
	26~30	124	40.7	Income	Less than $412	83	27.2
	31~40	57	18.7		USD 413–687	46	15.1
	41~50	14	4.6		USD 688–962	51	16.7
	51~60	6	2		USD 963–1237	32	10.5
	Above 60	0	0		Above $1237	93	30.5

.

4.2. Measures

(1) UTAUT2. This study utilized the measurement scale developed by Venkatesh et al. to assess the variables of performance expectancy, effort expectancy, social influence, facilitating conditions, and hedonic motivation within the UTAUT2 model as well as patients’ intention to adopt generative AI [18]. To ensure the applicability of scale to the current context, it was appropriately adapted and translated. For example, performance expectancy was measured using items such as “I find generative AI on online healthcare platforms useful in my daily life”, “Using generative AI on online healthcare platforms helps me handle health tasks more quickly”, and “Using generative AI on online healthcare platforms helps me manage my health and medical consultations more efficiently.” Participants rated their answers on a 7-point Likert scale, where 1 indicates strongly disagree and 7 indicates strongly agree. Details are provided in Appendix A.1, Appendix A.2, Appendix A.3, Appendix A.4, Appendix A.5 and Appendix A.11.

(2) Health literacy. To examine health literacy, this study employed the validated scale developed by Heijmans et al., which comprises nine items describing various scenarios, such as “When reading health-related instructions or brochures from hospitals, pharmacies, or community centers, you find words or phrases that you don’t understand” [54]. Participants rated their agreement with the items on a 7-point Likert scale, where 1 indicates strongly disagree and 7 indicates strongly agree. Details are provided in Appendix A.7.

(3) Construal level. To evaluate the construal level, this study utilized the Behavior Identification Form (BIF) developed by Vallacher and Wegner [32]. The BIF consists of 25 items, each presenting two interpretations of a behavior: one abstract interpretation representing the outcome of the behavior and one concrete interpretation representing the process of the behavior [32], as shown in Appendix A.8. For instance, one interpretation of “joining the army” is “helping the Nation’s defense”, while the other is “signing up”. The scale employs a scoring system wherein “1” signifies an abstract interpretation, which is high construal level, and “2” indicates a concrete interpretation, which is low construal level. Following Rosen et al., the total score of 25 items for each participant was averaged to derive their construal level [55]; the higher average score indicated a lower construal level.

(4) AI literacy. To measure AI literacy, this study employed the scale developed by Tully et al., which includes a comprehensive list of 17 multiple-choice questions, such as recognizing AI, “Which of the following is NOT powered by AI”; the interdisciplinarity of AI, “Which of the following fields contributes to the development of AI”; and understanding intelligence, “Which form of intelligence involves emotional understanding and social skills”; order of answers within each question were randomized [23]. Details are provided in Appendix A.10. Each question was scored as 1 point, and participants’ AI literacy was measured by calculating their total scores.

5. Results

5.1. Reliability and Validity

This study utilized SmartPLS 4.0 to assess the reliability and validity of the constructs. The results show that factor loadings for each construct exceed 0.7, Cronbach’s alpha values range from 0.708 to 0.924, composite reliability (CR) values fall between 0.726 and 0.997, and average variance extracted (AVE) values surpass the recommended threshold of 0.5. These results collectively indicate strong internal reliability and convergent validity of the scale, as shown in

Table 2. Reliability and validity.

Construct	Item	Outer-Loading	Cronbach’s Alpha	CR	AVE
Effort expectancy	EE1	0.825	0.812	0.818	0.639
	EE2	0.783
	EE3	0.814
	EE4	0.774
Facilitating conditions	FC1	0.798	0.708	0.726	0.630
	FC2	0.735
	FC3	0.844
Health literacy	HL1	0.700	0.924	0.997	0.607
	HL2	0.884
	HL3	0.826
	HL4	0.787
	HL5	0.754
	HL6	0.748
	HL7	0.739
	HL8	0.773
	HL9	0.787
Hedonic motivation	HM1	0.870	0.841	0.842	0.759
	HM2	0.859
	HM3	0.884
Performance expectancy	PE1	0.828	0.722	0.728	0.641
	PE2	0.784
	PE3	0.790
Social influence	SI1	0.885	0.865	0.866	0.787
	SI2	0.876
	SI3	0.901
Patients’ intention	PI1	0.887	0.756	0.775	0.675
	PI2	0.720
	PI3	0.849

. Furthermore, as presented in

Table 3. Correlation matrix.

	EE	FC	HL	HM	PE	SI	PI
EE	0.799
FC	0.528	0.794
HL	−0.244	−0.172	0.779
HM	0.553	0.397	−0.199	0.871
PE	0.586	0.509	−0.171	0.514	0.801
SI	0.475	0.320	−0.228	0.443	0.514	0.887
PI	0.506	0.486	−0.133	0.491	0.603	0.497	0.822

Note: The diagonal is the square root of the AVE. All abbreviations are defined in Table 1.

and

Table 4. Heterotrait–Monotrait ratio (HTMT).

	EE	FC	HL	HM	PE	SI	PI
EE
FC	0.699
HL	0.261	0.216
HM	0.667	0.509	0.214
PE	0.756	0.708	0.205	0.652
SI	0.562	0.418	0.246	0.519	0.645
PI	0.639	0.652	0.187	0.608	0.810	0.615

Note: All abbreviations are defined in Table 1.

, the square root of the AVE for each construct exceeds the corresponding inter-construct correlations, and the HTMT ratios remain below the recommended threshold of 0.85, demonstrating good discriminant validity [56]. Consequently, the measurement model is deemed credible and valid [57,58].

5.2. Common Method Bias

As all data in this study are self-reported, the potential for common method bias (CMB) is carefully considered. To assess this, an unrotated exploratory factor analysis was conducted. Therefore, an unrotated exploratory factor analysis was performed, and seven factors with eigenvalues greater than 1 were extracted [59]. The maximum factor variance explained was found to be 29.851%, which is less than the 40% threshold typically considered indicative of serious common method bias [60]. The results suggest that common method bias was not a significant concern in this study.

5.3. Hypothesis Testing

The hypotheses in this study were tested using SmartPLS 4.0. The results of the main effect, as shown in

Table 5. Result of main effect.

Path	Path Coefficients	Standard Deviation	T Statistics	Supported
EE → PI	0.036	0.079	0.455	No
FC → PI	0.170 *	0.071	2.393	Yes
HM → PI	0.140	0.077	1.815	No
PE → PI	0.307 ***	0.075	4.086	Yes
SI → PI	0.194 **	0.069	2.798	Yes
age → PI	−0.069	0.051	1.347	No
edu → PI	0.041	0.044	0.925	No
gender → PI	0.037	0.089	0.415	No
income → PI	0.126 *	0.054	2.328	Yes

* p < 0.05, ** p < 0.01, *** p < 0.001.

, indicate that performance expectancy positively associates with patients’ intention (β = 0.307, p < 0.001), thus supporting hypothesis 1. However, effort expectancy does not have a significant effect on patients’ intention (β = 0.036, p > 0.05), thus not supporting hypothesis 2. Social influence is found to be positively associated with patients’ intention (β = 0.194, p < 0.01), supporting hypothesis 3. Additionally, facilitating conditions positively associate with patients’ intention (β = 0.170, p < 0.05), providing support for hypothesis 4. Finally, the impact of hedonic motivation on patients’ intention is not significant (β = 0.140, p > 0.1), indicating that hypothesis 5 was not supported.

Table 6. Result of moderating effect.

Path	Path Coefficients	Standard Deviation	T Statistics	Supported
CL x PE → PI	−0.159 **	0.079	2.012	Yes
CL x EE → PI	0.058	0.065	0.889	No
HL x FC → PI	0.022	0.061	0.362	No
HL x SI → PI	−0.127 **	0.062	2.059	Yes
AIL x EE → PI	0.224 ***	0.080	2.788	Yes
AIL x PE → PI	−0.025	0.080	0.309	No
AIL x SI → PI	−0.131 **	0.061	2.140	Yes
AIL x HM → PI	−0.051	0.084	0.609	No
AIL x FC → PI	−0.116	0.071	1.631	No

** p < 0.05, *** p < 0.01.

presents the results of the moderation effects. The findings show that the construal level positively moderates the relationship between performance expectancy and patients’ intention. The construal level strengthens the positive effect of performance expectancy on patients’ intention (β = −0.159, p < 0.05), supporting hypothesis 6a. However, the moderating effect of the construal level on the relationship between effort expectancy and patients’ intention is not significant (β = 0.058, p > 0.05), failing to support hypothesis 6b. Health literacy, on the one hand, negatively moderates the relationship between social influence and patients’ intention (β = −0.127, p < 0.05), supporting hypothesis 7a. On the other hand, the moderating effect of health literacy on the relationship between facilitating conditions and patients’ intention is not significant (β = 0.022, p > 0.05), failing to support hypothesis 7b. The moderating effects of AI literacy on the relationships between performance expectancy, facilitating conditions, and hedonic motivation and patients’ intention are not significant (β = −0.025, p > 0.05; β = −0.116, p > 0.05; β = −0.051, p > 0.05), thus failing to support hypotheses 8a, 8c, and 8d. However, AI literacy significantly and positively moderated the relationship between effort expectancy and patients’ intention (β = 0.224, p < 0.01) and negatively moderated the relationship between social influence and patients’ intention (β = −0.131, p < 0.05), supporting hypotheses 8b and 8e, respectively.

6. Discussion

6.1. Principal Findings

This study aims to investigate the factors influencing patients’ intention to adopt generative AI on online healthcare platforms. First, the findings suggest that performance expectancy exerts a positive influence on patients’ intention to adopt generative AI. This implies that generative AI is a promising and effective model and can enhance health management efficiency and healthcare utility on online healthcare platforms. Second, effort expectancy does not exhibit a significant effect on patients’ intention, suggesting the difficulty that patients may face in trusting and operating the unique features and functionalities of generative AI. Third, social influences are identified as a positive determinant of patients’ intention, as individuals’ choices and inclinations are inevitably shaped by significant others, who not only impact patients’ decisions to adopt generative AI but also care about their physical well-being. Four, the research results reveal that facilitating conditions have a significant effect on patients’ intention to adopt generative AI. This finding can be attributed to the widespread accessibility of information and knowledge and the presence of appropriate hardware and software infrastructure, which supports the utilization of generative AI. Five, notably, while generative AI is a novel technology, its primary function is health management and healthcare service. Therefore, hedonic motivation is not the primary motivator for adoption of such technology.

Additionally, this study examines three moderators that influence the relationship between influencing factors and patients’ intention. Firstly, the research results demonstrate that the construal level plays a positive moderating role in the relationship between performance expectancy and patients’ intention. This effect is attributed to the tendency of individuals with a higher construal level to prioritize outcome-related factors that embody desirability over those with a lower construal level. However, no significant effect of the construal level is found on the relationship between effort performance and patients’ intention, possibly due to evolving societal values and contemporary changes, wherein even individuals with a lower construal level may not place adequate attention to the process aspect that could impact their assessment of objects and behaviors.

Secondly, the results of this study suggest that health literacy plays a negative moderating role in the relationship between social influence and patients’ intention while having no significant effect on the relationship between facilitating conditions and patients’ intention. This indicates that individuals with lower health literacy are more vulnerable to the influence of significant others on their inclination to adopt generative AI. Nevertheless, the pervasiveness of the Internet, the widespread use of hardware and software, and the abundance of knowledge and information online have made the impact of facilitating conditions on intention independent of patients’ health literacy.

Thirdly, the study findings reveal that AI literacy exerts a significant and positive influence on the relationship between effort expectancy and patients’ intention. This is because individuals with higher levels of AI literacy, compared with those with lower AI literacy, are better able to understand the functions and operational processes of AI, thereby strengthening the influence of perceived ease of use on their adoption intention. Additionally, AI literacy negatively affects the relationship between social influence and patients’ intention. This is attributed to patients with higher AI literacy possessing a more profound understanding of AI technologies, enabling them to independently assess the value of AI, making them less susceptible to external influences. However, no significant effect of AI literacy is detected on the relationship between performance expectancy and patients’ intention, which may be attributed to the powerful functionalities of generative AI. This leads to the influence of performance expectancy on patients’ intention to adopt generative AI remaining stable regardless of their level of AI literacy. Similarly, AI literacy does not significantly moderate the relationship between facilitating conditions and patients’ intention. This may be possibly because the facilitating conditions for adopting generative AI are sufficiently consistent across individuals with varying levels of AI literacy. Furthermore, no significant moderating effect of AI literacy is found on the relationship between hedonic motivation and patients’ intention, suggesting that patients’ perceived enjoyment in adopting generative AI is not contingent upon their level of AI literacy [61].

6.2. Theoretical Implications

This study offers three primary contributions to the study of generative AI and UTAUT2 models. Firstly, while prior research has examined factors such as social influence, novelty value, and anthropomorphism in relation to the acceptance of ChatGPT-3.5 [62], limited attention has been given to the adoption of generative AI by patients within online healthcare settings. This study seeks to address this gap by investigating the factors that influence patients’ intention to adopt generative AI on online healthcare platforms, thereby advancing theoretical understanding and offering insights into the key determinants shaping patient behavior in this emerging context. Secondly, this study expands the UTAUT2 model to apply to the adoption of generative AI. Our analysis reveals that effort expectancy and hedonic motivation do not significantly affect the intention of using generative AI. This finding highlights the need for a tailored approach to comprehending patients’ intention regarding generative AI on online healthcare platforms, as traditional models may not fully fit new research scenarios. Thirdly, this study extends the theoretical boundaries and explanatory power of the UTAUT2 model in healthcare by introducing moderators such as the construal level, health literacy, and AI literacy. These variables demonstrate influence on the relationship between factors and patients’ intention. By integrating these moderating variables into the UTAUT2 model, we offer a more comprehensive understanding of the factors influencing patients’ intention to adopt generative AI in online healthcare scenarios.

6.3. Implications for Practice

This study has practical implications for the development and promotion of generative AI on online healthcare platforms. First, performance expectancy is positively associated with patients’ intention to adopt generative AI, but effort expectancy has no significant influence on patients’ intention. Thus, platform managers should underscore the need for effective educational and promotional efforts to enhance patients’ understanding of the benefits of generative AI and their ability to use it. This is further supported by the confirmation that AI literacy positively moderates the relationship between effort expectancy and patients’ intention. Second, the findings indicate that social influence is positively associated with patients’ intention, especially among individuals with lower health literacy and AI literacy [63]. Platform managers should encourage patients to recommend the use of generative AI to others by providing rewards and should particularly encourage young patients to recommend it to older adults and to those living in remote areas with low health literacy and AI literacy [64]. Third, facilitating conditions are positively associated with patients’ intention to adopt generative AI, whereas hedonic motivation has no significant influence on patients’ intention. Therefore, platforms should enhance the compatibility of generative AI to facilitate patient use, while investment in entertainment functionalities may not be necessary. These strategies are likely to attract more patients and expand the value of generative AI on online healthcare platforms.

6.4. Limitations and Future Research

This study has several limitations that warrant consideration and provide directions for future research. First, patients may engage with generative AI on online healthcare platforms for varying purposes, such as seeking diagnostic support, managing chronic conditions, or accessing general health information. These differentiated purposes reflect heterogeneous expectations and value perceptions [65], which may systematically influence the relationships between antecedent factors and patients’ intention to adopt generative AI. However, this study does not account for such heterogeneity. Future research should examine the moderating role of patients’ usage purposes, thereby unpacking how different user goals shape the strength or direction of influencing factors. Secondly, this study does not account for additional constructs that may influence adoption behavior, such as price value and usage habits [66]. As generative AI technologies become more prevalent in online healthcare contexts, future research could incorporate these variables to capture evolving patients’ behavior. Furthermore, exploring other potentially relevant factors, such as novelty value, perceived risk, and privacy concerns, may offer a more comprehensive understanding of the facilitators and inhibitors shaping patients’ intention to adopt generative AI in an online health context [14,67]. Lastly, the data are collected in China without regional distinction and within a specific time frame, which may limit the generalizability of the findings across diverse cultural contexts and temporal settings. Future studies should collect data across multiple regions and time points to improve external validity and capture potential cultural, regional (rural vs. urban), and temporal variations in patients’ perceptions of generative AI on online healthcare platforms [68].

7. Conclusions

This study employs the extended model of the unified theory of acceptance and use of technology 2 (UTAUT2) to investigate the factors that influence patients’ intention to adopt generative AI on online healthcare platform as well as the moderating effects of the construal level, health literacy, and AI literacy. The questionnaire-based data analysis revealed that performance expectancy, social influence, and facilitating conditions positively associated with patients’ intention, while effort expectancy and hedonic motivation do not significantly affect patients’ intention. Moreover, the inclusion of moderators reveals that the construal level strengthens the relationship between performance expectancy and patients’ intention, while health literacy weakens the relationship between social influence and patients’ intention. Additionally, AI literacy strengthens the relationship between effort expectancy and patients’ intention but weakens the relationship between social influence and patients’ intention. These findings contribute to a more nuanced understanding of the factors that drive patients’ intention to adopt generative AI on an online healthcare platform and highlight the importance of considering the role of a boundary condition, which may strengthen or weaken the factors on patients’ intention.