Nurses Reveal Hopes and Fears Over Generative AI in Clinical Research

Generative artificial intelligence has swept into hospitals, newsrooms, and laboratories with astonishing speed, but one of the most revealing portraits of how the technology is actually being used at the bedside comes from a new qualitative study published in BMC Nursing. Researchers Wenbo Qiao and Xinyue Xiang, both of the First Affiliated Hospital of Zhejiang University School of Medicine in Hangzhou, China, set out to capture what frontline clinical nurses genuinely experience when they turn to generative AI tools to support clinical research. Their findings paint a picture that is neither utopian nor dystopian, but something far more practical: a workforce that sees real efficiency gains in data-heavy tasks while remaining deeply wary of technical failures, privacy risks, and the murky ethics of machine-assisted scholarship.

The study adopted an exploratory qualitative descriptive design, a method chosen precisely because the researchers wanted rich, contextualized accounts rather than numeric satisfaction scores. Through purposive and snowball sampling, the team recruited twelve frontline clinical nurses from multiple tertiary hospitals in Zhejiang Province. Crucially, every participant had hands-on experience with both clinical research and generative AI applications, ensuring that the interviews captured informed users rather than curious outsiders. To maximize diversity, the sample deliberately spanned different hospital departments, professional titles, years of clinical work, levels of research experience, and habitual patterns of AI use, a design decision that strengthens the credibility of the themes that ultimately emerged.

Data collection took the form of semi-structured online interviews, a format that allowed participants to speak freely while ensuring that key domains such as role perception, workflow impact, and support needs were consistently explored. The researchers analyzed transcripts inductively using thematic content analysis supported by NVivo 15.0 software, and the study followed the COREQ checklist, the widely accepted reporting standard for qualitative research. Saturation was assessed dynamically during repeated coding cycles: no new codes or themes appeared after the tenth interview, and two additional interviews confirmed that the dataset had reached its interpretive limits. That kind of methodological transparency matters, because qualitative findings live or die on the rigor with which themes are derived from raw testimony.

From this analysis, five interrelated themes emerged, which the authors summarize as an interpretive model of a collaborative practice ecology. The first theme describes a spectrum of attitudes stretching from efficiency-driven acceptance to ethical skepticism. Some nurses had embraced generative AI enthusiastically, praising its ability to accelerate literature review, questionnaire drafting, and the mundane mechanical work that often bogs down research projects. Others viewed the same capabilities through a more cautious lens, questioning whether speed obtained at the cost of verification and accountability is genuinely a gain for science. The study’s refusal to flatten this diversity into a single sentiment is one of its most valuable contributions, since most prior work has either focused on nursing education or treated nurse researchers as a homogeneous block.

The second theme concerns dual application scenarios. Participants reported that generative AI genuinely empowers data-oriented tasks, from cleaning and structuring datasets to generating code snippets and summarizing text. Yet the tools proved conspicuously limited when it came to understanding clinical context. Nurses described situations in which AI outputs were technically fluent but clinically naive, missing the subtleties of patient populations, departmental workflows, and the lived realities behind a data point. This gap between statistical plausibility and clinical validity is a recurring concern in health AI, and the study documents how frontline staff, who occupy the interface between data and patients, feel it most acutely.

The third theme identifies what the authors call the core challenges: technical reliability, data security, and ambiguities around academic integrity. Reliability worries centered on hallucinations and subtle errors that could propagate into research outputs if unchecked. Data security loomed even larger, given that clinical research often involves identifiable patient information subject to strict confidentiality obligations. Nurses questioned whether entering study-related content into third-party AI platforms could expose sensitive data. Meanwhile, academic integrity emerged as a gray zone: participants were uncertain about when AI assistance crosses the line from acceptable support into misconduct, noting the absence of clear institutional rules to guide them. The paradox is striking: nurses are using tools faster than the norms governing their use can be written.

The fourth theme tracks evolving role perceptions. Over time, participants began to reconceptualize generative AI from a basic tool, something akin to an advanced search engine or spell-checker, into a potential intelligent data-analysis assistant capable of more substantive collaboration. This perceptual shift carries practical consequences. A tool framing invites casual, unexamined use; an assistant framing invites delegation, oversight, and questions about responsibility. As nurses reposition AI within their professional hierarchy of collaborators, institutions will need to decide what levels of autonomy are appropriate and who bears accountability when an AI-assisted analysis goes wrong.

The fifth and final theme captures expectations for the future. Nurses in the study want three things: profession-adapted technology that understands nursing-specific terminology and contexts, targeted AI literacy training that goes beyond generic tutorials, and clear institutional norms that define acceptable use, protect patient data, and resolve integrity questions. The authors argue that collaboration between nurses and generative AI requires a deliberate balancing of efficiency against risk, and they call for a systematic strategy encompassing context-adapted tools, enhanced AI literacy, and explicit ethical and organizational guidelines. In other words, the responsibility for safe and effective adoption does not rest on individual nurses alone; it belongs to hospitals, educators, and technology developers as well.

The significance of this research extends well beyond Zhejiang Province. Clinical nurses are increasingly expected to contribute to research output as part of professional advancement, yet they typically juggle research with demanding clinical schedules, making efficiency tools especially attractive. At the same time, nursing research deals with some of the most sensitive data in medicine. The tension the study documents, between the productivity that generative AI promises and the vigilance that patient privacy and scientific rigor demand, is likely to play out in every health system adopting these technologies. By grounding the debate in the concrete experiences of actual users, the study offers policymakers a template for what guidance must address: verification practices, data-handling boundaries, integrity definitions, and training curricula.

The authors are candid about their limitations. The sample comprised only twelve GenAI-experienced nurses drawn from tertiary hospitals in a single Chinese province, so the findings should be applied cautiously to other settings, particularly primary care environments or institutions at earlier stages of AI adoption. Still, the interpretive model they propose, a collaborative practice ecology in which attitudes, applications, challenges, roles, and expectations interlock, provides a framework that future quantitative and intervention studies can test and refine. As generative AI continues its rapid diffusion into healthcare, this study stands as an early, careful record of how the people closest to patients are negotiating the technology’s promise and peril, and a reminder that the success of AI in medicine will be determined not by the sophistication of the algorithms but by the trust, competence, and protections afforded to the professionals who use them.

Beyond its substantive findings, the study offers a useful illustration of how qualitative evidence can complement the growing body of quantitative surveys on AI adoption in healthcare. Numbers can reveal how many nurses use generative AI or how frequently, but they cannot explain why a nurse hesitates to paste a patient dataset into a chatbot, or how professional identity shifts when a machine becomes a working partner. By following the COREQ reporting standard and documenting saturation explicitly, the authors provide a level of procedural detail that allows other researchers to appraise the trustworthiness of the themes and to replicate the approach in different health systems.

The institutional setting of the research is also worth noting. Tertiary hospitals in China are typically academic medical centers where research participation is woven into professional expectations for nursing staff, and where ethics oversight structures such as the institutional review board that approved this study are well established. That environment helps explain why participants were both experienced users of AI and acutely aware of governance gaps: they work in organizations that simultaneously demand research productivity and enforce strict data confidentiality, leaving them to navigate the tension largely on their own.

The study’s transparency extends to its own relationship with the technology it examines. The authors disclose that a generative AI tool was used solely to improve the readability and language of the manuscript, with full human review and accountability, and that no AI was involved in the design, data collection, analysis, or interpretation of the research. This kind of declaration is becoming an expected feature of credible publications, and its presence here models the very norm clarity that participants said they wanted from their own institutions.

For readers considering how such findings might translate into practice, the most actionable thread is the call for AI literacy training tailored to nursing. Generic digital skills courses rarely address the specific failure modes of generative models, such as fabricated citations or plausible but incorrect clinical reasoning, and they seldom cover the data-protection calculus nurses must perform before using a third-party platform. Profession-specific curricula, paired with written institutional policies defining acceptable use, would directly address the ambiguities participants described.

Finally, the interpretive model of a collaborative practice ecology invites empirical testing. Future work could quantify the attitude spectrum, compare nurses across hospital tiers and regions, or evaluate whether targeted training and clear guidelines measurably reduce the risks participants identified while preserving the efficiency gains they value.

Subject of Research: Frontline clinical nurses' experiences and challenges using generative AI to support clinical research

Article Title: Experiences and challenges of clinical nursing staff using generative AI to support clinical research: a qualitative study

Article References: Qiao, W., & Xiang, X. (2026). Experiences and challenges of clinical nursing staff using generative AI to support clinical research: a qualitative study. BMC Nursing. https://doi.org/10.1186/s12912-026-05167-w

Image Credits: AI Generated

DOI: 10.1186/s12912-026-05167-w

Keywords: generative artificial intelligence, clinical nursing, clinical research, qualitative study, nursing research, data security, academic integrity, AI literacy, frontline nurses, thematic content analysis, tertiary hospitals, China