The global community is facing an unprecedented demographic challenge: rapid population aging. According to the Ministry of Civil Affairs and the National Working Commission on Aging of China, by the end of 2023, the number of individuals aged 60 and above in China had reached 296.97 million—21.1% of the total population—with those aged 65 and over accounting for 216.76 million (15.4%). Over the next decade, the country is projected to add more than 20 million older adults individuals annually, and the population of the oldest old (aged 80 and above) is expected to surpass 50 million by 2035 (1). This demographic shift, coupled with a rising burden of chronic disease, places immense strain on healthcare and caregiving systems.

In the economically vibrant and densely populated Guangdong–Hong Kong–Macao Greater Bay Area (GBA), population aging presents similarly urgent challenges. Ensuring access to high-quality and sustainable eldercare services has become a critical regional priority. According to a recent industry report by KPMG, the rapid advancement of artificial intelligence (AI) is reshaping opportunities in healthcare and eldercare. Over the coming decade, AI technologies are expected to drive transformative innovations—from AI-assisted diagnostic systems and personalized treatment plans to autonomous caregiving robots—potentially alleviating workforce shortages and enhancing service delivery (2).

The GBA has emerged as a testbed for AI-driven healthcare innovations. As reported by China Daily, public hospitals in Hong Kong are deploying AI systems to identify high-risk patients and enable early intervention. Locally developed models, such as the ophthalmic diagnostic system VisionFM, now demonstrate accuracy on par with or superior to expert clinicians. In Shenzhen, AI-assisted surgical planning platforms like “CARES Copilot” are being integrated into clinical practice. Collectively, these advancements signal a growing regional commitment to leveraging AI for more efficient, intelligent healthcare systems (3).

To provide readers with a clearer overview of the diversity of AI-driven healthcare systems in the region, we have summarized the main categories of current applications in Figure 1, including remote monitoring, care coordination, telehealth, documentation, and decision support.

However, the realization of AI’s full potential hinges not only on technical capability, but also on end-user acceptance. In clinical and caregiving contexts—where safety, reliability, and ethics are paramount—trust plays a pivotal role. Despite AI’s promise, widespread skepticism persists among patients and their families. Studies indicate that 65.8% of surveyed adults express low trust in the use of AI in healthcare, and 57.7% doubt the healthcare system’s ability to prevent AI-related harm (4). Even as general health and AI literacy improve, trust deficits remain, suggesting that building acceptance will require deeper engagement and transparency (5).

Mistrust is often exacerbated by negative user experiences with overhyped or underperforming AI caregiving products. Such setbacks have made trust and perceived risk key barriers to adoption. Notably, when it comes to eldercare decisions, family members frequently serve as surrogate decision-makers. Many older adults rely on family due to cognitive limitations or unfamiliarity with digital tools, depending on them to install health apps, interpret electronic health records, or facilitate telehealth consultations (6, 7).

In this context, family caregivers often become de facto gatekeepers for AI adoption. If they harbor doubts or perceive high risk, they may withhold consent for AI-assisted care, regardless of its potential benefits (8). Therefore, understanding the perspectives of patients’ families—their willingness to adopt AI caregiving technologies and the factors shaping their trust—is crucial to unlocking the transformative value of AI in eldercare. As the GBA continues to accelerate AI integration into healthcare systems, addressing the needs, concerns, and behavioral intentions of these key stakeholders is essential for achieving meaningful, user-centered innovation.

Although earlier research highlighted the persistent resistance to innovation in care services (9, 10), more recent evidence indicates that the COVID-19 pandemic has accelerated the adoption of digital health and also reshaped the dynamics of trust. For instance, Park et al.’s research found that the emotions of family caregivers strongly influence their willingness to adopt medical AI (6); After surveying 2,039 respondents, Nong and at found that although health literacy has improved, respondents’ trust in the use of AI in the healthcare system remains fragile (5). Similarly, a quantitative survey and analysis by Janne Kauttonen and other researchers on trust and acceptance of artificial intelligence applications in the healthcare sector revealed that doubts about AI-assisted healthcare persisted after the pandemic, mainly stemming from emotional and privacy concerns. These findings highlight the necessity of examining the acceptance of AI care in the post-pandemic care environment (11). In the context of sensor-based health monitoring, studies by T.F. Kummer have shown that hospitals in Germany and Australia deploying intelligent surveillance systems frequently encounter user anxiety triggered by a complex interplay of factors. These include uncertainty about the system’s effectiveness, concerns over privacy and security, and emotionally driven intuitive rejection (10). Such findings underscore that, in healthcare settings, technology adoption is shaped not only by rational cost–benefit evaluations but also by psychological and affective dimensions.

Located at the crossroads of Eastern and Western medical traditions and cultural frameworks, the Guangdong–Hong Kong–Macao Greater Bay Area (GBA) presents a distinctive context in which public perceptions of medical AI are shaped by a unique interplay of influences. The bases for trust and the ways users perceive risk in this region are likely to diverge significantly from those observed elsewhere. Nevertheless, existing research on public acceptance of AI-driven healthcare solutions within the GBA remains limited, with even fewer studies concentrating on the perspectives of family caregivers—a stakeholder group of considerable importance.

Conventional frameworks for studying technology acceptance, such as the Technology Acceptance Model (TAM) and the Unified Theory of Acceptance and Use of Technology (UTAUT), are commonly utilized to evaluate factors influencing the adoption of AI in healthcare and eldercare environments (12). Despite their broad application, these models frequently overlook essential psychological components, particularly trust and perceived risk (8). Although the UTAUT’s core elements—performance expectancy, effort expectancy, social influence, and facilitating conditions—offer explanatory power in general information systems research, their ability to predict user intentions diminishes in healthcare contexts, where the stakes of decision-making and potential errors are much higher.

In such settings, trust in AI’s reliability and safety, as well as perceived risks of unintended harm, become pivotal in shaping decisions. Overlooking these factors may obscure the reasons why some family members remain unwilling to authorize the use of AI-based care technologies, even when their potential benefits are clearly demonstrated (4, 8). Therefore, there is a compelling need to extend existing theoretical models by incorporating trust as a central antecedent and perceived risk as a moderating factor in explaining AI adoption behaviors in healthcare.

Based on these insights, this study seeks to answer the following core research question:

What factors influence family members’ willingness to adopt AI caregiving technologies in the Greater Bay Area, and how do trust and perceived risk shape this decision-making process?

To tackle this research gap, we construct an extended conceptual framework grounded in UTAUT, incorporating trust and perceived risk as pivotal psychological factors. Specifically, the model maintains the core UTAUT predictors—performance expectancy (perceived AI utility), effort expectancy (ease of use), social influence (social encouragement/pressure), and facilitating conditions (supportive resources/infrastructure)—as foundational determinants of behavioral intention. Trust is integrated as an additional direct antecedent, defined as family members’ assurance in the safety and competence of AI-assisted caregiving. Furthermore, we propose that perceived risk—conceptualized as the subjective assessment of potential adverse consequences linked to AI use—exerts a moderating effect on the trust-intention relationship. This synthesized framework seeks to provide a more holistic explanation of the psychological drivers shaping AI adoption choices in eldercare, particularly within the Greater Bay Area’s unique socio-technical environment.

Performance expectancy, as defined by Venkatesh et al. (26), denotes an individual’s perception that adopting a specific technology will enhance their task performance or goal attainment. Empirical evidence consistently links this expectancy to stronger intentions to adopt technologies (36). Conversely, effort expectancy captures the perceived ease of use and low effort required when utilizing a technology (26). This dimension similarly predicts adoption intention (37). To enhance clarity for nursing participants, effort expectancy was assessed using reverse-scored items in this study.

Social influence is conceptualized as an individual’s belief about whether significant others (e.g., society, hospitals, physicians) expect them to use a particular technology (26). Within AI-based nursing, this translates to perceptions of AI acceptance by these key stakeholders. Facilitating conditions represent the perceived availability of resources and support enabling effective technology use (26).

Building on UTAUT’s theoretical foundation, the following hypotheses are advanced:

Considering the specific context of AI in nursing, this study further proposes that trust in AI shapes nurses’ acceptance willingness (38). Therefore, an additional hypothesis is formulated:

As AI-based nursing is an emerging technology, nurses’ decision-making may involve more complex psychological and organizational factors. For instance, under high perceived risk, nurses may question whether AI can make accurate diagnoses; perceived risk may suppress social normative pressure, rendering social influence less effective; and there may be cognitive dissonance between perceived risk and trust (e.g., concerns about algorithmic transparency). Therefore, perceived risk is introduced as a moderating variable:

Age is also introduced as a moderating variable. Older individuals tend to experience a decline in working memory capacity, making them more cognitively burdened by multi-step AI operations and more susceptible to frustration (39). Furthermore, older users are more emotionally motivated and less susceptible to social approval or authoritative opinions (40). Glikson and Woolley (38) also suggest that a high level of trust can reduce perceived risk in older adults. Based on this, the following hypotheses are proposed:

Venkatesh and Morris (26) found that gender moderates the effect of facilitating conditions on technology acceptance, with women being more influenced by the availability of support. Additionally, research in healthcare settings has shown that female nurses tend to have greater needs for team-based resources and collaboration. Therefore, the following hypothesis is proposed (Figure 2):

This study aimed to explore the factors influencing family caregivers’ acceptance of AI nursing technology and the underlying mechanisms, constructing and empirically testing a theoretical model integrating the Unified Theory of Acceptance and Use of Technology (UTAUT) with additional variables of trust and perceived risk. The study focused specifically on the Guangdong-Hong Kong-Macao Greater Bay Area—a region characterized by both international and local features—to provide valuable insights for promoting effective application of AI in nursing services within this region and beyond. The findings identified performance expectancy (β = 0.235, p < 0.01), facilitating conditions (β = 0.349, p < 0.01), and caregiver trust (β = 0.334, p < 0.001) as the three primary factors driving behavioral intention to use AI nursing. These results indicate that caregivers prioritize whether AI effectively reduces caregiving burdens (e.g., automated nursing reports), sufficient technical support from hospitals (e.g., 24-h equipment maintenance), and trust in algorithm reliability. Although social influence significantly enhanced usage intention (β = 0.198, p < 0.05), its effect was notably attenuated by perceived risk (moderation effect β = −0.339, p < 0.05). Furthermore, older caregivers exhibited significantly higher sensitivity to operational complexity (interaction effect of age and effort expectancy β = 0.254, p < 0.05), suggesting a need for simplified interaction designs. However, the moderating effect of age on social influence was not significant (p > 0.05), highlighting the rigorous and autonomous nature of caregivers’ decision-making processes regarding AI nursing.

Although AI systems in the Greater Bay Area span a wide spectrum—from monitoring sensors and scheduling tools to telehealth platforms and clinical decision support—their very heterogeneity contributes to uncertainty among caregivers. Even when certain systems demonstrate strong technical performance in pilot programs, caregivers often generalize concerns about malfunction, privacy, or operational complexity across all AI systems. This explains why our study found persistently low trust (β = 0.334, p < 0.001) despite evidence of effectiveness in similar contexts, and why perceived risk significantly weakened the effect of social influence (β = −0.339, p < 0.05). Building trust therefore requires not only improving algorithmic performance but also enhancing system transparency, caregiver training, and human-in-the-loop oversight mechanisms.

By focusing on caregivers as decision-makers and employing literature review and empirical analysis, this research advances social understanding of caregivers’ decision logic concerning AI nursing acceptance. It facilitates theoretical innovation in AI nursing research, shifting focus from technological efficacy to family acceptance mechanisms, and highlights how risk perception reshapes technology acceptance in safety-sensitive contexts.

Within the unique social-technological and healthcare context of the Guangdong-Hong Kong-Macao Greater Bay Area, this study holds particular research value. On the one hand, the region boasts advanced medical technologies and high digital literacy, potentially leading to overall greater acceptance of AI nursing. Conversely, the differences of institutional among cities (e.g., Hong Kong, Guangzhou, Macao) add valuable complexity to regional AI adoption research. This study aimed to identify regional variations in caregiver trust and risk perception and their effects on technology adoption intentions, thus informing region-specific policy development.

This study has several limitations. First, the sample size was relatively small. Our target population—primary family members actively involved in caregiving and medical decision-making—was inherently difficult to recruit, given substantial psychological and time pressures during patient care; strict adherence to ethical standards and informed consent without inducements further reduced response rates. Second, the sample was concentrated in the Guangdong–Hong Kong–Macao Greater Bay Area and skewed younger (ages 18–25 constituted 60.12%), which limits generalizability to other age groups and regions. Model- and design-related constraints should also be noted. Although the theoretical framework included trust and perceived risk, it did not fully capture deeper ethical constructs such as algorithmic transparency and accountability mechanisms, which may partly explain why certain moderation effects (e.g., H6a, H6c) were not significant. Moreover, the cross-sectional survey design cannot reflect dynamic shifts in technology acceptance—especially adaptations in caregiver trust following real-world AI exposure. Our ethical–legal discussion remains partly conceptual and jurisdiction-specific.

While the EU Artificial Intelligence Act (Regulation (EU) 2024/1689) establishes a risk-based regime, key implementation details and timelines are still evolving; by contrast, China currently regulates AI through sector-specific instruments rather than a single comprehensive act. Likewise, patient co-ownership models (e.g., GPOC) show early feasibility in the literature but have not yet been operationalized or evaluated in our context (47–49). To address these limitations, future work will:

In parallel, we will expand sample diversity—especially older decision-makers—strengthen cross-region comparisons, integrate ethical and institutional context variables, and adopt longitudinal designs to better illuminate trust-building mechanisms and risk-mitigation pathways in high-risk healthcare scenarios, thereby providing more precise decision-support for the sustainable implementation of smart nursing in aging societies.

Our findings also highlight the lack of formal training programs for caregivers in the use of AI-driven healthcare systems. Despite receiving some technical support, caregivers often struggled with system complexity and malfunction, which negatively affected their trust and willingness to adopt these technologies. Future studies should explore the development of structured, accessible training programs to enhance caregiver competency in using AI healthcare tools, especially for older caregivers who may face additional technological barriers. Moreover, it is essential that technical support be available, comprehensive, and responsive to caregivers’ needs to improve their experience with AI systems.

Finally, our study did not differentiate in detail between the multiple categories of AI-driven healthcare systems (e.g., monitoring, coordination, documentation), which may limit comparability across heterogeneous applications. Future research should stratify system types to better capture context-specific trust dynamics.