This longitudinal descriptive qualitative study used a phenomenological approach to systematically track changes in senior leaders’ perspectives on GenAI adoption over 6 months [14]. The study design incorporated two waves of data collection (August 2024 and February 2025) to capture evolving awareness and attitudes. The interval was selected to allow sufficient time for changes while minimizing recall bias. The phenomenological approach was chosen to prioritize participants’ lived experiences of navigating GenAI adoption in their professional roles, which aligned with established methodologies for studying technology assimilation in health care leadership contexts [15].

Participants were purposively selected through the professional network of the first author (ZZ) from three tertiary hospitals located in Hebei, Guangdong, and Sichuan provinces to represent regional diversity in China’s health care infrastructure [16,17]. The sample comprised senior leaders holding executive positions with direct authority over institutional technology decisions, including hospital directors/deputies and department heads/deputies responsible for digital technology adoption initiatives. Participants were contacted by phone and informed of the purpose and process of the study and the rights of participants before being invited. The first interviews were conducted immediately or during a scheduled appointment, at the end of which an appointment was made for the second interview. Recruitment continued until theoretical saturation was achieved, with two consecutive interviews yielding no novel analytic information [18,19]. All interviews in both waves were conducted with the same participant cohort. Notably, we enrolled traditional Chinese medicine (TCM) hospital leaders to enrich sample diversity and ensure our findings apply beyond Western medicine contexts.

Data were collected through semistructured telephone interviews guided by a protocol organized into three investigative domains [20,21]. The first domain focused on overall perceptions of GenAI, probing participants’ awareness (“Have you heard of GenAI?”) and self-assessed knowledge levels (“Can you describe how GenAI works in your own words?”). The second domain explored professional involvement in GenAI adoption, including current or past engagement with artificial intelligence (AI)–related projects, training programs, or policy development. The third domain examined attitudes and decision-making processes, investigating factors influencing institutional adoption choices such as resource allocation priorities, staff readiness assessments, and perceived utility. Notably, the protocol included two interview guides for the first and second interview waves. The core components remained consistent, with modifications made to questions in the second interview guide to reflect changes over time (Multimedia Appendix 1).

The interview protocol underwent a 3-stage development process. Initial questions were derived from a systematic review of GenAI adoption literature [6-13], which were refined through pilot testing with two TCM administrators not included in the study sample. The pilot testing identified ambiguous terminology requiring simplification. Final revisions emerged from iterative discussions among the research team, involving three rounds of role-played interviews to standardize probing techniques [22-25].

ZZ, JZ, MX, and YZ conducted the interviews. All participants knew ZZ professionally. To minimize bias arising from these relationships, each interviewer was assigned only participants at or above their organizational level. This arrangement reduced potential power differentials and social desirability effects.

Interviews were recorded with consent. Audio recordings were transcribed verbatim within 48 hours of collection, with transcriptions cross-verified against original recordings by an independent researcher to ensure accuracy [26,27].

The analysis followed Colaizzi’s data analysis method [28], implemented manually through collaborative team discussions. Researchers first immersed themselves in the data by independently reading all transcripts twice, once for holistic understanding and again to identify significant statements. Through iterative discussion, the team clustered these meanings into preliminary themes, which were further condensed into core thematic categories through constant comparative analysis. The final analytic phase involved member checking, where findings were shared with 10 participants to confirm interpretive accuracy [29,30].

Analytical rigor was ensured through multiple verification strategies. Peer debriefing involved reviews with an external qualitative researcher who critiqued emerging themes and challenged interpretive assumptions, refining conclusions about decision-making dynamics [31]. An audit trail documented the key decisions of team discussions [32]. Member checking engaged participants in reviewing anonymized theme summaries for clarification. Reflexivity was maintained via researcher discussions on preconceptions about GenAI’s role, enabling explicit bracketing during interpretation [33]. The strategies collectively addressed credibility through external scrutiny, dependability via process transparency, confirmability by tracing analytical decisions, and transferability through contextual descriptions of institutional settings. Rigor was further strengthened by cross-verifying transcripts against recordings to ensure consistency between raw data and interpreted findings.

We received an ethics exemption from the Academic Ethics Committee of Hebei University of Chinese Medicine/Hebei Province Traditional Chinese Medicine Hospital according to its institutional policy, as the study involved no patients, patient data, or interventions and posed minimal risk for participants. Participants provided verbal informed consent before the first interview and verbal reaffirmation at the 6-month follow-up. We did not provide compensation to participants in this study. All data were anonymized using numeric identifiers (P1-P32, followed by a role code of H for hospital director/deputy and D for department head/deputy, and a code W1/2 for interview wave; eg, P1-HW2 means participant 1 who is a hospital director/deputy and the quote is from the second interview wave). All identifying information was removed from both the demographic data and transcriptions.

Audio files and transcripts were stored on two password-protected flash drives, with one for backup, which were kept separately by the first and the corresponding authors, along with the hard copies such as field notes and discussion meeting minutes. The data were accessible only to the research team. Participants retained the right to withdraw data until the final analysis phase [34,35].

A total of 32 participants were interviewed, all of whom completed both interview waves, resulting in a final dataset of 64 interviews. Interviews averaged 39.3 (range 21-47) minutes during the first wave and 44.7 (range 29-52) minutes during the second wave. A total of 32 leaders (12 women, 20 men) from 3 tertiary hospitals (1 TCM hospital in Hebei and 2 Western medicine hospitals in Sichuan and Guangdong) participated. Ages ranged from 39 to 53 (mean 46.4) years, and leadership tenure spanned 4-17 (median 11.3) years. Participants included 8 (25%) hospital directors/deputies and 24 (75%) department heads/deputies, 2 of whom were promoted to deputy director by the second interview. The cohort was a good representation of the demographic characteristics of Chinese hospital leaders (Table 1).

Our analysis of interview transcriptions revealed three main themes, as follows.

Our interviews revealed a marked improvement in participants’ understanding of GenAI over the 6-month period. Compared with the first wave of interviews, a substantially higher number of participants in the second wave demonstrated awareness and knowledge about GenAI. While only 17 participants (13 department heads/deputies, 4 hospital directors/deputies) initially were able to correctly describe GenAI in their own language and name examples of GenAI technology and possible professional scenarios for application, it increased to 29 participants in the second wave (22 department heads/deputies, 7 hospital directors/deputies). Participants attributed this enhanced understanding to a variety of knowledge sources, including media coverage, discussions with peers and staff, professional conventions, and direct involvement in GenAI projects:

Our interviews reveal a clear changing process from early uncertainty to later enthusiasm as leaders moved past initial obstacles and embraced specific catalysts for GenAI adoption.

This longitudinal qualitative study investigated the evolving perceptions and attitudes of senior Chinese hospital leaders toward the adoption of GenAI in professional settings over a 6‐month period. Three key themes emerged: a significant improvement in the leaders’ understanding of GenAI, a generally supportive attitude toward its adoption that evolved over time and varied by leadership role, and a dynamic shift in the perceived obstacles and catalysts influencing GenAI integration. Our findings provide valuable insights into the complex process of technology adoption within the unique context of Chinese hospitals, where government policies and competitive pressures drive rapid digital transformation.

The first key finding was a marked increase in participants’ awareness and knowledge of GenAI over time. Initially, fewer than half of the participants could correctly describe the concept and potential applications of GenAI. However, by the second interview wave, this number increased significantly, indicating that as leaders were exposed to diverse sources of information, including media reports, peer discussions, professional conventions, and hands-on involvement in GenAI projects, their understanding deepened. This finding supports existing research that identifies a lack of knowledge as a major barrier to AI adoption in multiple industries [36-38]. In our study, as leaders’ awareness improved, their ability to envision practical applications and assess potential benefits increased, thereby aligning with the diffusion of innovation theory, which emphasizes that early exposure and accessible knowledge are critical to moving along the adoption curve [39]. This suggests that, in promoting GenAI adoption, hospitals and policy makers should invest in targeted educational initiatives to further bridge the knowledge gap among health care leaders.

The second major theme was the supportive attitude toward GenAI adoption that was expressed across all participants, with notable evolution regarding formal institutional adoption. In the first wave, many senior leaders, particularly those in higher administrative roles, favored individual experimentation with GenAI rather than committing significant institutional resources. This cautious approach is commonly seen in early technology adoption stages, where concerns about budget allocation, uncertain return on investment, and the risks inherent in new technologies are often prevalent. By the second wave, however, hospital directors and deputies exhibited increased willingness to move beyond individual use and actively plan for institutional implementation, including formal budgeting and strategic integration. This transition reflects a growing confidence in the technology’s potential, a progression that echoes patterns observed in health care technology adoption literature, where initial caution gradually gives way to institutional endorsement as practical benefits become clearer [40]. Our findings suggest that showcasing successful pilot projects and clear return on investment pathways could accelerate the formal adoption process.

The third theme highlighted the evolving landscape of obstacles and drivers influencing GenAI adoption. In the initial phase, obstacles such as limited technical literacy, uncertainty about practical applications, skepticism regarding benefits, and a lack of dedicated budgets were predominant. These challenges resonate with previous studies that identify similar barriers in AI integration within health care [41,42]. However, as the study progressed, many of these obstacles diminished, likely as a result of improved knowledge, increased hands-on experience with GenAI, and changes in the policy and technological environment, and new drivers emerged. For instance, peer pressure from other hospitals that had begun integrating GenAI, a clearer vision of its potential applications, and the increasing availability of budgetary resources all served as catalysts. Notably, our study period coincided with DeepSeek’s launch in China (July 2024) and its global debut (August 2024). Although participants did not discuss the driving impact of DeepSeek explicitly, several leaders noted that its technical breakthrough accelerated both awareness and institutional interest in GenAI, which reinforces the notion that breakthroughs in technology can rapidly shift the adoption landscape [43,44]. Furthermore, institutional isomorphic pressures may help explain why hospital leaders converged on similar GenAI strategies [45,46], especially in the context of the high-profile debut of DeepSeek. Under coercive pressures, national policy endorsements and regulatory expectations compelled leaders to align with government-backed technology agendas. Simultaneously, normative pressures emerged as peer institutions shared GenAI successes at professional forums, creating industry norms that legitimized investment [42,47]. The dual forces reduced uncertainty, reinforced trust in GenAI’s value, and accelerated formal adoption across hospitals, demonstrating how external mandates and interorganizational learning jointly shape technology uptake. The observations align with the broader literature on technology diffusion, which posits that overcoming initial barriers through positive experiences and external validation can lead to accelerated adoption [43-45]. Leveraging coercive and normative pressures can normalize innovation, streamline budget approvals, and overcome resistance rooted in perceived risk [46,47]. Therefore, fostering a culture of knowledge and experience sharing, networking to share and showcase peer hospital GenAI implementations, and aligning institutional plans with emerging mandates can be strategies to leverage peer influence and emerging technological advancements.

Notably, the three themes are closely intertwined. Improved understanding of GenAI appears to have played a crucial role in transforming supportive attitudes toward formal adoption, as increased knowledge alleviated initial uncertainties and facilitated resource commitment. Concurrently, the mitigation of early obstacles allowed emerging catalysts, such as the influence of DeepSeek and competitive peer adoption, to drive a shift from cautious exploration to proactive institutional integration. In comparison with existing literature, our findings extend traditional technology adoption models by emphasizing the critical role of longitudinal exposure and experiential learning in overcoming barriers and activating drivers. These insights suggest that hospitals should consider implementing ongoing education and pilot initiatives, while policy makers should create frameworks that support early experimentation and gradual scaling of GenAI applications in health care.

Our findings offer several important implications for hospital administrators and policy makers seeking to facilitate GenAI adoption. First, there is a clear need for ongoing educational initiatives aimed at senior hospital leaders to enhance their understanding of GenAI’s capabilities, limitations, and potential applications within health care. These initiatives should leverage various channels, including media, peer-to-peer learning, and professional conventions. Second, demonstrating the practical value and return on investment of GenAI through well-defined pilot projects is crucial to encourage formal institutional adoption, particularly among budget-conscious leaders. Showcasing successful use cases and addressing concerns related to resource allocation can build confidence and pave the way for broader implementation. Third, fostering a culture of innovation and knowledge sharing among hospitals can leverage the power of peer influence and accelerate the adoption process. Creating platforms for hospitals to share their experiences and best practices with GenAI implementation can be highly beneficial. Finally, continuous evaluation and strategic consideration of emerging AI technologies, such as DeepSeek, are essential to capitalize on advancements that offer improved accessibility, efficiency, and cost-effectiveness. Addressing remaining obstacles such as data privacy and ethical considerations through clear guidelines and robust security measures will also be critical for building trust and ensuring the responsible integration of GenAI in Chinese hospitals. Additionally, creating interhospital networks, showcasing peer GenAI implementations, and then integrating the insights into institutional planning in accordance with emerging policy mandates can be a potential strategy to combine normative peer influence with coercive endorsement, helping leaders normalize GenAI use, justify budget requests, and accelerate formal adoption.

This study has several limitations. First, our sample was drawn from three tertiary hospitals. This limited the diversity of sampling, excluding other types of hospitals in China. Second, data collection relied on telephone interviews, which could limit the depth of information compared to face-to-face interactions. Future research should consider incorporating a larger, more diverse sample and a quantitative or mixed methods design to validate and extend our findings and provide a further understanding of GenAI adoption dynamics in health care settings. Furthermore, we did not collect direct evidence to assess how DeepSeek’s launch and national-level endorsement drove GenAI uptake, which was a likely driver of participants’ changes in perception and attitudes over the study period. Future research should investigate the operational impacts of major technology rollouts, such as DeepSeek’s debut, and accompanying policy endorsements on adoption dynamics in hospital settings.

Senior Chinese hospital leaders’ perceptions and attitudes toward GenAI have evolved significantly over time. Improved knowledge through diverse information channels has enhanced their comprehension and shifted their stance from initial reluctance toward support for formal institutional adoption. The dynamic interplay between diminishing obstacles and emerging catalysts, notably the influence of peer practices and technological breakthroughs such as DeepSeek, underscores the potential for transformative change in health care management. Targeted educational initiatives, pilot projects, and robust policy frameworks are needed to facilitate GenAI integration.