Advantages, limitations, and ethical concerns of AI-assisted qualitative research in nursing: insights from a human–AI comparative thematic analysis Authors: Mingjie Wu, Xiaoyang Guo, Shuxian Yang, Ying Li, Ting Yu, Hanjiao Liu Published: 2026 ABSTRACT AbstractBackgroundThe rapid development of large language models has accelerated the use of artificial intelligence (AI) in qualitative research. AI-assisted thematic analysis may improve coding efficiency and data organization, but concerns remain about its ability to interpret emotionally nuanced, contextually embedded, and culturally situated nursing narratives. Existing studies have emphasized coding performance and efficiency, whereas ethical and interpretive concerns in nursing qualitative research remain insufficiently examined.ObjectiveThis secondary qualitative analysis compared human-led reflexive thematic analysis and ChatGPT-assisted thematic analysis of the same nursing interview dataset and explored the advantages, limitations, and ethical concerns of AI-assisted qualitative inquiry in nursing.MethodsSemi-structured interview data originally collected within a broader nursing research platform project were analyzed through two separate pathways: human-led reflexive thematic analysis based on Braun and Clarke’s approach and ChatGPT-assisted thematic analysis. The research team compared the outputs to examine convergence, divergence, contextual simplification, thematic representation, and ethical implications. Ethical concerns were interpreted through comparative and reflexive analysis.ResultsThe research team observed that AI-assisted analysis supported rapid data organization, preliminary coding, and identification of explicit thematic patterns. However, the two analytic pathways differed in their representation of contextual meanings, tacit nursing knowledge, and experience-based nursing narratives. Through comparative and reflexive interpretation, four ethical themes were identified: privacy and data security concerns; loss of contextual and emotional understanding; algorithmic reductionism and cultural bias; and changing roles and responsibilities in nursing research. AI-assisted outputs tended to be semantically coherent but contextually simplified.ConclusionsIn this study, AI-assisted qualitative analysis was useful as a supportive tool for preliminary thematic exploration and data organization. However, human reflexive engagement remained essential for contextual interpretation, ethical judgment, and the representation of nursing experiences. The findings suggest that ethical concerns extend beyond technical performance to contextual representation, interpretive integrity, and researcher responsibility. Future research should develop reflexive, ethically grounded human–AI collaborative approaches to qualitative nursing inquiry.Clinical trial registrationNot applicable.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12912-026-05085-x.Keywords: Artificial intelligence, Qualitative research, Nursing research, Thematic analysis, Reflexive thematic analysis, Ethics, Large language models, ChatGPT, Human–AI collaboration Introduction The rapid development of generative artificial intelligence (AI) has stimulated increasing interest in its application across health care research, education, and clinical practice [1–2]. In particular, large language models (LLMs) such as ChatGPT have demonstrated strong capabilities in natural language processing, text summarization, semantic organization, and pattern recognition, leading researchers to explore their potential use in qualitative data analysis [3–4]. Recent studies have suggested that AI-assisted approaches may improve analytic efficiency, accelerate coding processes, and facilitate rapid synthesis of large qualitative datasets [5–6]. As qualitative research becomes increasingly data-intensive and resource-demanding, AI-assisted thematic analysis has been proposed as a promising strategy for supporting contemporary qualitative inquiry [7]. The transformative potential of LLMs in health-related qualitative research has been widely documented, yet context-specific challenges in nursing remain underexplored [8]. The integration of AI into qualitative research, however, has generated substantial methodological and epistemological debate [9]. Unlike quantitative analysis, qualitative inquiry is fundamentally interpretive and context-dependent. Qualitative research seeks not only to categorize textual information but also to understand how participants construct meanings through emotional experiences, social interactions, and contextual realities [10]. In nursing research, qualitative methods are particularly important because nursing practice is deeply embedded in relational care, emotional labor, contextual decision-making, and tacit professional knowledge [11]. Interviews with nurses frequently contain implicit meanings, emotionally nuanced expressions, and experiential understandings that extend beyond explicit verbal content [12]. Consequently, qualitative nursing analysis depends heavily on researchers’ contextual sensitivity, interpretive engagement, and reflexive understanding of participants’ lived experiences. Among qualitative methodologies, reflexive thematic analysis has become one of the most widely used approaches in nursing and health care research because of its flexibility and emphasis on interpretive meaning-making [13]. Braun and Clarke conceptualize thematic analysis not as a purely technical process of code extraction, but as an active and reflexive engagement between researchers and data [14]. Themes are therefore understood as interpretive constructions shaped through researchers’ positionality, theoretical orientation, clinical experience, and analytic judgment [10]. Although reflexive thematic analysis can produce rich and contextually grounded findings, the process is often labor-intensive, iterative, and highly dependent on researchers’ expertise [15]. These challenges have contributed to growing interest in whether AI-assisted approaches can support qualitative analysis while maintaining methodological rigor and interpretive depth. Emerging studies have explored the use of LLMs in coding, theme generation, rapid qualitative analysis, and qualitative workflow support [16]. Several studies have reported that AI-assisted analysis may improve coding speed and facilitate preliminary thematic organization [16–17]. However, concerns remain regarding the reliability and interpretive validity of AI-generated qualitative findings. Existing evidence suggests that AI-generated analyses may produce oversimplified thematic structures, hallucinated interpretations, shallow contextual understanding, and inconsistent outputs depending on prompt design and analytic conditions [18–19]. Furthermore, LLMs often prioritize semantic regularities rather than contextual or experiential meanings, raising concerns regarding their ability to interpret emotionally complex and socially situated narratives. These concerns may be particularly important in nursing qualitative research. Nursing narratives frequently involve tacit knowledge, emotional communication, relational care practices, and culturally embedded expressions that require interpretive sensitivity beyond literal textual interpretation [20]. Clinical nursing experiences are often communicated indirectly through contextual descriptions, emotional nuance, and experiential reflection rather than through explicit conceptual statements. Consequently, AI-generated thematic analysis may risk reducing complex nursing experiences into generalized or decontextualized categories. Such reduction may not only weaken interpretive depth but also raise broader ethical concerns regarding the representation of participants’ lived experiences in qualitative inquiry. Current discussions surrounding AI-assisted qualitative research have increasingly focused on ethical issues including privacy protection, algorithmic bias, transparency, accountability, and data governance [21–22]. Nevertheless, much of the existing literature approaches AI ethics from a technical or procedural perspective. Comparatively limited attention has been paid to the ethical implications of AI-assisted interpretation within qualitative nursing research contexts [23–25]. In reflexive qualitative inquiry, interpretation itself carries ethical significance because researchers actively shape how participants’ experiences are represented, reconstructed, and communicated. AI-assisted analysis may therefore introduce new ethical concerns related to contextual misrepresentation, interpretive reductionism, emotional flattening, and shifting interpretive authority from human researchers toward algorithmic systems [18]. In addition, existing studies comparing AI-assisted and human qualitative analysis have largely focused on coding agreement, efficiency, or thematic similarity [6]. Few studies have examined how AI may influence the interpretation of emotional, contextual, and experience-based meanings within nursing narratives. Moreover, relatively little research has explored these issues using real nursing interview data and reflexive thematic analysis approaches [8]. Understanding these interpretive and ethical dimensions is increasingly important as AI-assisted qualitative analysis becomes more widely integrated into nursing and health care research workflows. Although previous studies have explored ChatGPT-assisted qualitative analysis, many have focused on coding efficiency, thematic similarity, or general methodological implications. Less attention has been paid to how AI-generated thematic representations differ from human reflexive interpretations when both are applied to the same nursing interview dataset. This issue is particularly important in nursing research, where interpretation depends on clinical experience, emotional sensitivity, tacit professional knowledge, contextual understanding, and ethical judgement. The unique contribution of this study is its human–AI comparative design within a nursing research context. By comparing human-led reflexive thematic analysis with ChatGPT-assisted thematic analysis of the same anonymized interview data, this study provides empirical insight into the possibilities and boundaries of AI-assisted qualitative analysis in nursing. Therefore, this secondary qualitative analysis aimed to compare human-led reflexive thematic analysis and ChatGPT-assisted thematic analysis of the same nursing interview dataset, and to explore the advantages, limitations, and ethical concerns associated with AI-assisted qualitative inquiry in nursing. Methods Study design This study was a secondary qualitative analysis of interview data originally collected within a broader nursing research platform project. The original project was the Shenzhen Bao’an Public Hospital High-quality Development Research Project entitled “Construction and Implementation of a Clinical Research Platform Based on a Specialty-Specific Traditional Chinese Medicine Nursing Intervention Database” (Grant No. BAGZL2024050). The broader project aimed to explore the experiences, needs, expectations, and practical challenges of different categories of potential users of a future nursing research platform, including clinical nursing staff, nursing managers, and nursing researchers. The original interviews mainly focused on nursing research data collection and utilization, data governance, research support needs, platform functions, implementation barriers, and future platform development. During the early stages of platform development, the research team also considered AI-assisted functions as a potential component of the future nursing research platform. Therefore, the dataset included participants’ views on digitalization, nursing research data use, research platform development, AI-assisted tools, and related concerns such as privacy, data security, reliability, ethical responsibility, and algorithmic bias. The present secondary analysis compared human-led reflexive thematic analysis and ChatGPT-assisted thematic analysis of the same anonymized nursing interview dataset. The purpose was not to evaluate the original platform project itself, but to examine how human researchers and ChatGPT organized, represented, and interpreted the same qualitative nursing data, with particular attention to the advantages, limitations, and ethical implications of AI-assisted qualitative analysis. The overall study workflow is presented in Fig. 1. Study workflow of the human–AI comparative thematic analysis Participants and data collection Purposive sampling was used to recruit participants from a tertiary integrated Chinese and Western medicine hospital in Shenzhen, China. The participants were originally recruited for a broader nursing research platform project that aimed to explore the experiences and needs of potential users of a future nursing research platform. Therefore, the sample intentionally included different stakeholder groups related to nursing research data use, digital platform development, and nursing research practice, including clinical nursing staff, nursing managers, nursing researchers, nursing department staff, nursing teachers, and postgraduate nursing students. The inclusion criteria were as follows: (1) being engaged in clinical nursing, nursing management, nursing education, nursing research, nursing department work, or postgraduate nursing training; (2) having experience with or familiarity with nursing documentation, nursing research, digital nursing systems, or research platform-related work; (3) being able to clearly express their experiences and opinions; and (4) agreeing to participate voluntarily. Participants were excluded if they were unable to complete the interview, declined audio recording, or withdrew consent during the study. Eligible participants were identified through nursing department recommendations, department contacts, and the research team’s clinical and academic networks. The primary researcher contacted potential participants, explained the study purpose and interview procedures, and invited them to participate voluntarily. Participants were informed that refusal or withdrawal would not affect their work, study, or professional evaluation. No eligible participants refused to participate or withdrew after providing consent. In this study, the term “nursing personnel” was used broadly to include nursing professionals and nursing research stakeholders. Postgraduate nursing students were included because they were professional master’s nursing students who were undertaking clinical practice and research training in the hospital during their postgraduate education. In China, professional master’s nursing students usually receive hospital-based clinical training and participate in nursing research-related activities during postgraduate study. In addition, these students had completed approximately 10 months of hospital-based clinical internship during their undergraduate nursing education. Therefore, they were considered nursing research stakeholders and potential users of digital nursing research platforms and AI-assisted research support tools. All 19 participants had previous exposure to or experience with AI-assisted tools through personal use, research-related use, or hospital-based AI training. Their reported uses included literature searching, idea organization, writing support, presentation preparation, case presentation preparation, and research-related tasks. Previous AI experience was not used as a formal inclusion criterion, but was considered relevant to the present secondary analysis because AI-assisted functions were planned as a potential component of the future nursing research platform. A total of 19 nursing personnel from the same tertiary integrated Chinese and Western medicine hospital participated in this study. Semi-structured interviews were conducted between May 11 and May 29, 2026, by the primary researcher in quiet and private settings within the hospital environment. Depending on participants’ availability and clinical workflow, interviews were conducted either individually or in small groups. All interviews were conducted in Chinese and audio-recorded with participants’ permission. The interviews were conducted by MW and/or HL. MW was a female nursing postgraduate researcher, and HL was a female chief nurse and master’s supervisor with experience in nursing management and research. Both interviewers were familiar with the study background and had experience in nursing research. No prior personal or work-related relationship was established between the interviewers and participants before recruitment. Participants were informed of the interviewers’ identities, the study purpose, and the voluntary nature of participation before the interviews. No non-participants were present during individual interviews; small-group interviews included only invited participants and the interviewer(s). No repeat interviews were conducted. No field notes were made during or after the interviews. Transcripts were not returned to participants for checking. No qualitative data analysis software such as NVivo was used for the human-led analysis. The interview guide was initially developed for a broader qualitative study exploring stakeholders’ experiences and needs regarding the construction of a specialty-specific traditional Chinese medicine (TCM) nursing research data platform. The guide was informed by relevant literature on nursing informatics, data governance, qualitative research, and digital research support systems. Rather than focusing exclusively on artificial intelligence, the interviews primarily explored participants’ experiences with nursing data collection and management, nursing research practice, research platform construction, and the use of digital technologies in nursing research. Rather than strictly following predetermined questions, the interviews were conducted flexibly to encourage participants to share their practical experiences, reflections, and concerns regarding nursing research and digital technologies. The interviews primarily explored challenges associated with nursing research data acquisition and standardization, expectations regarding research platform support, perceptions of AI-assisted research tools, and reflections on the integration of digital technologies into nursing research practice. Representative questions used in the present secondary analysis included: From your perspective, what are the major difficulties in obtaining, recording, and using nursing research data in current clinical practice? From your perspective, what are the major difficulties in obtaining, recording, and using nursing research data in current clinical practice? What types of nursing data or clinical information do you think are most important to standardize and manage within a nursing research platform? What types of nursing data or clinical information do you think are most important to standardize and manage within a nursing research platform? What kinds of support do you think digital platforms could provide for nursing research? What kinds of support do you think digital platforms could provide for nursing research? What are the major difficulties currently faced by nursing personnel when conducting nursing research? What are the major difficulties currently faced by nursing personnel when conducting nursing research? With the emergence of AI-assisted tools, do you think these technologies have practical value in nursing research? Why or why not? With the emergence of AI-assisted tools, do you think these technologies have practical value in nursing research? Why or why not? The five questions listed above were representative questions used in the present secondary analysis rather than the complete interview guide. Among these five representative questions, one question directly addressed AI-assisted tools, while the remaining questions focused on nursing research data, data standardization, digital platforms, and research support needs. The full stakeholder-specific interview guide included additional AI-related questions and probing questions for clinical nursing staff, nursing managers, and nursing researchers, and is provided in Supplementary File 1. Therefore, the present study is positioned as a secondary qualitative analysis of a broader interview dataset rather than a primary interview study designed exclusively around AI-assisted qualitative analysis. AI-related discussion constituted a focused but not exclusive part of the original dataset. No quantitative word-count or line-count analysis was conducted to calculate the exact proportion of AI-related material, because the present study was designed as a qualitative secondary analysis rather than a content-frequency analysis. Therefore, the ethical themes reported in this study should be understood as comparative and reflexive interpretations based on participants’ AI-related comments, broader digital research platform narratives, and the human–AI analytic comparison, rather than as prevalence-based themes derived solely from direct AI-specific questions. During the interviews, some participants also spontaneously reflected on issues related to emotional interpretation, contextual understanding, data privacy, and the future role of AI-assisted technologies in qualitative nursing research. Additional probing questions were used to further explore these emerging issues in greater depth. Each interview lasted approximately 30–60 min. Interviews were transcribed verbatim shortly after completion. All identifiable information was removed during transcription, and participants were assigned numerical identifiers (P1–P19) to ensure confidentiality and anonymity. Data collection and preliminary review of the interview material were conducted concurrently. Sample adequacy was considered in relation to information power rather than a fixed numerical target. The sample included participants with varied professional roles, educational backgrounds, clinical experience, research involvement, and familiarity with digital or AI-assisted tools. Recruitment was discontinued when the interview data provided sufficient breadth and depth to address the secondary analysis aim and when additional interviews were unlikely to substantially expand the comparative interpretation. Participant characteristics are shown in Table 1. Table 1Participant characteristicsParticipant IDGenderAgeProfessional roleYears of experienceEducationP1Female48Head of TCM Nursing Clinic26Bachelor’s degreeP2Male50Head of Emergency Department28Bachelor’s degreeP3Female46Head of Operating Room24Bachelor’s degreeP4Male47Team Leader of Operating Room25Bachelor’s degreeP5Female45Head of Neurosurgery Department22Bachelor’s degreeP6Female49Head of Encephalopathy Department29Bachelor’s degreeP7Female26Clinical Nurse (Encephalopathy)3Bachelor’s degreeP8Female28Clinical Nurse (Encephalopathy)6Bachelor’s degreeP9Female24Clinical Nurse (Encephalopathy)2Bachelor’s degreeP10Female30Clinical Nurse (Encephalopathy)9Bachelor’s degreeP11Female23Postgraduate Student (Nursing)1Postgraduate studentP12Female24Postgraduate Student (Nursing)1Postgraduate studentP13Female25Postgraduate Student (Nursing)2Postgraduate studentP14Male26Postgraduate Student (Nursing)2Postgraduate studentP15Female33Nursing Teacher4Master’s degreeP16Female29Research Nurse (Encephalopathy)3Master’s degreeP17Female28Nursing Staff (Nursing Department)2Master’s degreeP18Female31Nursing Staff (Nursing Department)4Master’s degreeP19Female27Research Nurse (Cardiology)2Master’s degreeNote. For postgraduate nursing students, years of experience refers to years of postgraduate clinical practice and nursing research training in the hospital during their professional master’s nursing education. These students had also completed approximately 10 months of hospital-based clinical internship during their undergraduate nursing education Participant characteristics Note. For postgraduate nursing students, years of experience refers to years of postgraduate clinical practice and nursing research training in the hospital during their professional master’s nursing education. These students had also completed approximately 10 months of hospital-based clinical internship during their undergraduate nursing education Human reflexive thematic analysis Human-led thematic analysis was conducted following Braun and Clarke’s reflexive thematic analysis approach [26]. Analysis began with repeated reading of the interview transcripts to facilitate familiarization with participants’ narratives and contextual meanings. Initial codes were subsequently developed inductively through close engagement with the data. Rather than treating themes as objective entities emerging directly from the transcripts, the analysis emphasized researchers’ active role in meaning-making and interpretation. Reflexive discussions were conducted throughout the analytic process to examine how researchers’ disciplinary backgrounds, clinical experiences, and interpretive perspectives shaped thematic development. Codes and themes were iteratively reviewed, refined, and reorganized through ongoing comparison across interview narratives. Particular attention was paid to emotionally nuanced experiences, tacit nursing knowledge, contextual meanings, and participants’ reflections regarding digital and AI-assisted technologies in nursing research practice. AI-assisted thematic analysis To explore the role of AI in qualitative nursing inquiry, the same anonymized interview transcripts were analyzed using ChatGPT-assisted thematic analysis. ChatGPT-assisted analysis was conducted using ChatGPT Plus through the ChatGPT web interface, with the model displayed as GPT-5.5. The analysis was conducted between May 30 and June 5, 2026. Chat history and the ChatGPT Memory function were enabled, but the analysis was conducted in study-specific sessions. The original interview transcripts were in Chinese, and the ChatGPT-assisted analysis was also conducted in Chinese to preserve participants’ original meanings and culturally embedded expressions. English translations of representative quotations were prepared only after theme development. Before AI-assisted analysis, the interview transcripts were anonymized to remove direct identifiers and reduce potential indirect identification risks. Potentially identifiable contextual details were reviewed and generalized where necessary before the transcripts were entered into ChatGPT. The anonymized interview transcripts were entered into ChatGPT in batches according to transcript length. Each batch generally contained one to two complete anonymized transcripts. The same general prompt structure was used across transcript batches. The exact number of ChatGPT web sessions was not prospectively recorded, which should be considered a procedural reproducibility limitation. However, the analytic procedure was documented by transcript batch, prompt type, AI-generated output, and researcher review notes. Each transcript batch was analyzed once using the same general prompt structure, and the same batch was not intentionally resubmitted to generate alternative outputs. ChatGPT was asked to summarize the interview content, identify initial codes, and generate preliminary subthemes and themes based only on the provided transcript text. The AI-assisted analysis involved two iterative cycles. In the first cycle, ChatGPT was used to support preliminary coding and thematic organization. In the second cycle, follow-up prompts were used to clarify, organize, and refine the AI-generated outputs. The AI-generated outputs were reviewed by the research team and were not directly accepted as final findings. Instead, they were compared with the themes developed through human reflexive thematic analysis. Examples of human–AI comparative interpretation are presented in Table 2. The ChatGPT-assisted thematic analysis protocol and representative prompt examples are provided in Supplementary File 2. Table 2Complete example of human–AI comparative thematic interpretationAnalytic componentExampleTranscript excerpt from a small-group interviewP9: “If you are writing a research project, you can tell AI your general idea and ask it to help organize your thinking. Sometimes the structure it provides is clearer than what you thought of by yourself.” P10: “But doesn’t it sometimes fabricate data or references?”Human-generated initial codesAI-assisted idea organization; support for early-stage research thinking; perceived usefulness of AI; concern about fabricated data; concern about inaccurate references; need for human verificationHuman-led interpretationHuman-led analysis interpreted this exchange as reflecting participants’ ambivalent understanding of AI-assisted research. AI was viewed as useful for organizing early research ideas, but participants also recognized that AI-generated information may be unreliable and requires human checking. This interpretation was linked to researcher responsibility, verification, and the need to maintain professional judgement when using AI-assisted tools.ChatGPT-assisted representationChatGPT-assisted analysis summarized this exchange as showing that AI tools can support research idea organization, writing, and literature-related tasks, while also presenting risks related to inaccurate or fabricated information.Theme developmentThis comparison contributed to the development of the theme “Changing Roles and Responsibilities in Nursing Research.”Final comparative interpretationBoth analytic pathways identified the usefulness and risks of AI-assisted tools. However, human-led interpretation placed the exchange within the broader issue of changing researcher responsibility: AI may support preliminary thinking, but researchers remain responsible for judging the credibility, relevance, and ethical use of AI-generated outputs. ChatGPT-assisted analysis organized the content into a balanced list of advantages and risks, whereas human-led analysis emphasized the shift in interpretive responsibility and the need for continued researcher reflexivity. Complete example of human–AI comparative thematic interpretation Chronology and human–AI comparative analysis The human-led and ChatGPT-assisted analyses were conducted as two separate analytic pathways using the same anonymized interview dataset. First, the Chinese interview transcripts were transcribed, checked, cleaned, and anonymized. Second, the human-led reflexive thematic analysis was conducted by the research team through repeated reading, coding, theme development, and reflexive discussion. Third, the same anonymized transcripts were analyzed using ChatGPT-assisted thematic analysis. The prompts used in the initial ChatGPT-assisted analysis were based on the transcript text and general thematic analysis instructions, rather than on the final human-generated themes. The chronological workflow of the human–AI comparative thematic analysis is shown in Fig. 1. The comparison between the two analytic pathways began after the human-generated themes and ChatGPT-assisted outputs had been developed. The comparison was conducted at the level of initial codes, preliminary themes, theme descriptions, and interpretations of selected excerpts. Human-led interpretations were not treated as an objective reference standard. Consistent with reflexive thematic analysis, they were understood as situated interpretations constructed through the research team’s clinical, cultural, and methodological engagement with the data. The comparison therefore focused on how the two analytic pathways constructed and represented meaning differently, and why certain representations were considered more contextually grounded in relation to the nursing research setting. The research team examined areas of convergence and divergence, with particular attention to contextual simplification, loss of emotional meaning, culturally embedded interpretation, and researcher responsibility. Through this comparative and reflexive analytic process, the research team identified broader methodological and ethical issues arising from AI-assisted qualitative analysis. These issues were not treated as direct outputs of either the human-led or ChatGPT-assisted analysis alone. Instead, the final themes reported in this study were developed through the research team’s comparative and reflexive interpretation. The ethical themes identified are summarized in Table 3. Table 3Ethical themes emerging from human–AI comparative thematic analysisThemesSubthemesPrivacy and data security concernsConfidentiality risks; Concerns regarding AI data processing and data ownershipLoss of contextual and emotional understandingLoss of experiential nursing knowledge; Reduction of situational and emotional meaningAlgorithmic reductionism and cultural biasMisinterpretation of TCM nursing knowledge and culturally embedded expressions; Linguistic and contextual reductionismChanging roles and responsibilities in nursing researchShifting interpretive authority; Dependence on AI-assisted analysis and reduced researcher reflexivity Ethical themes emerging from human–AI comparative thematic analysis Researcher reflexivity The research team included members with backgrounds in clinical nursing, nursing management, nursing education, nursing research, qualitative research, and traditional Chinese medicine nursing. These professional backgrounds provided important contextual knowledge for interpreting participants’ accounts, particularly those related to nursing documentation, research data use, clinical workflow, TCM nursing practice, digital platforms, and AI-assisted research. Researchers with clinical nursing and TCM nursing backgrounds contributed to the interpretation of tacit nursing knowledge, syndrome differentiation, individualized intervention processes, and culturally embedded expressions. Researchers with nursing research and qualitative research experience contributed to coding, theme development, and reflexive discussion. At the same time, the research team recognized that professional experience and familiarity with the broader nursing research platform project could influence data interpretation. For example, researchers with clinical and TCM nursing backgrounds may have been more sensitive to tacit nursing knowledge, culturally embedded meanings, context-dependent clinical judgement, and the practical complexity of TCM nursing interventions. Researchers with nursing research experience may have paid greater attention to data standardization, research usability, and methodological rigor. To enhance reflexivity, the team held repeated discussions, recorded analytic decisions, and reflected on how their professional experiences, assumptions about AI, and prior understanding of the research context may have shaped coding, theme development, and interpretation. During the human–AI comparative analysis, the team also considered how these positionalities influenced the interpretation of differences between human-led and ChatGPT-assisted outputs. The purpose was not to present the human interpretation as a neutral or objective standard, but to make explicit how clinical, cultural, and methodological knowledge informed the research team’s comparative interpretation. Trustworthiness and rigor Several strategies were used to enhance the trustworthiness and rigor of the study. First, investigator triangulation was applied. At least two researchers independently reviewed the transcripts, preliminary codes, and emerging themes, and differences were discussed within the research team. Second, peer debriefing was conducted through regular team meetings, during which emerging interpretations and comparative findings were critically examined. Third, an audit trail was maintained, including anonymized transcripts, coding records, theme development notes, ChatGPT prompts, AI-generated outputs, and human–AI comparison tables. Fourth, the research team systematically compared human-generated and AI-generated outputs to examine convergence, divergence, contextual simplification, and ethical implications. Finally, representative participant quotations were used to support the themes and improve the transparency of interpretation. Ethical considerations Ethical approval was obtained from the Ethics Committee of Shenzhen Clinical College of Integrated Chinese and Western Medicine, Guangzhou University of Chinese Medicine (Approval No. KY-2026-102-01). All participants provided written informed consent before participation. Participants were informed of the study purpose, interview procedures, confidentiality measures, and their right to withdraw from the study at any time without penalty. Special consideration was given to ethical issues associated with AI-assisted qualitative analysis. All interview transcripts were anonymized before being entered into ChatGPT. Direct identifiers, including participant names, hospital names, departments, contact information, and other personally identifiable information, were removed. Potential indirect identifiers, such as rare professional roles, specific workplace descriptions, unusual clinical situations, and highly specific contextual details, were also reviewed and generalized where necessary. No identifiable information was intentionally included in AI prompts or outputs. AI-assisted interpretations were critically reviewed by the research team to minimize risks of contextual misrepresentation, interpretive distortion, and reduction of participants’ experiences. This study was conducted in accordance with the Declaration of Helsinki. The full interview guide is provided in Supplementary File 1. Reporting was guided by the Consolidated Criteria for Reporting Qualitative Research (COREQ) checklist, which is provided as Supplementary File 3. Results Both human-led and AI-assisted thematic analyses identified several key issues related to nursing research data, data standardization, digital platforms, and research practices. From the research team’s observation, AI-assisted analysis was useful for organizing interview data, identifying explicit semantic patterns, and generating preliminary thematic structures. However, these outputs required human review, contextual interpretation, and ethical evaluation. However, reflexive comparison revealed notable differences in how human researchers and AI systems interpreted contextual meanings, experiential nursing knowledge, and culturally embedded clinical experiences. Rather than simply producing inaccurate interpretations, AI-assisted analysis frequently represented participants’ narratives through semantically coherent but relatively simplified thematic structures.Compared with human researchers, AI-generated interpretations tended to prioritize generalized conceptual categories while providing less attention to contextual, experiential, and relational dimensions of nursing practice. The four themes reported below were developed through the research team’s comparative interpretation of the dataset rather than being treated as themes derived solely from participants’ direct statements about AI. The interpretation drew on three sources: participants’ views on nursing research data, digital platforms, research support needs, and AI-assisted tools; differences between the human-led and ChatGPT-assisted analytic outputs; and the ethical implications identified during the comparison. Through this comparative and reflexive analytic process, four major ethical themes were identified: (1) privacy and data security concerns; (2) loss of contextual and emotional understanding; (3) algorithmic reductionism and cultural bias; and (4) changing roles and responsibilities in nursing research. Privacy and data security concerns This theme was developed from participants’ concerns about data security and confidentiality, together with the research team’s comparison of how human-led and ChatGPT-assisted analyses represented privacy-related risks in nursing research data. Participants expressed substantial concerns regarding data security, privacy protection, and data governance in the context of nursing research and digital platforms. Many participants emphasized that nursing research data often contain clinical information, nursing records, and potentially identifiable patient and staff information. Although digital technologies may facilitate data management and research efficiency, concerns remained regarding how data are stored, managed, accessed, and protected. One participant stated: Research data involve patients, nurses, and clinical records. If the platform is not well managed, there may be risks related to data security and privacy. (P17) Another participant explained: When data are uploaded to a system, we are not always sure who can access them or how they will be used in the future. (P18) Human researchers interpreted these statements as reflecting concerns regarding data ownership, confidentiality, and accountability in digital nursing research environments. Participants emphasized that data governance involves not only technical security measures but also trust, transparency, and clear responsibilities regarding data use. In contrast, AI-assisted thematic analysis primarily categorized these concerns as issues related to “data security” and “information management.” While AI accurately identified the technical dimensions of participants’ concerns, it paid less attention to issues of trust, professional responsibility, and uncertainty surrounding the long-term use of nursing research data. Reflexive comparison suggested that participants’ concerns extended beyond technical confidentiality and reflected broader ethical questions regarding data ownership, transparency, and governance within increasingly digitalized nursing research environments. Loss of contextual and emotional understanding This theme was developed through comparison of participants’ accounts and the two analytic outputs, particularly in relation to how emotional, contextual, and experiential meanings were represented or simplified. Loss of contextual and emotional understanding emerged as one of the most prominent themes during the comparative analysis. Participants frequently discussed challenges related to nursing data standardization, documentation, and the representation of complex nursing practices within digital systems. Human researchers tended to interpret these narratives within their broader clinical and organizational contexts, whereas AI-assisted analysis tended to organize these experiences into broader conceptual categories. One participant stated: Different nurses may record the same intervention in different ways, making it difficult to unify the data. (P17) Human researchers interpreted this statement as reflecting the complexity of nursing knowledge, variations in clinical practice, and the challenges of preserving contextual meaning during standardization. AI-assisted thematic analysis primarily summarized this issue as a challenge in nursing data standardization. Similarly, another participant explained: Many details of nursing care are not reflected in the system records. (P1) Human researchers interpreted this statement as highlighting the limitations of digital systems in representing experiential nursing knowledge, interpersonal communication, and individualized care processes. AI-assisted analysis, however, tended to classify the issue primarily as a documentation problem or system limitation. Participants frequently emphasized that nursing work involves observation, communication, judgment, and experience that may not be fully captured through structured data fields. Human researchers interpreted these narratives through the lens of tacit knowledge, situational understanding, and context-dependent clinical practice, whereas AI-assisted interpretation tended to organize these experiences into broader conceptual categories emphasizing explicit semantic patterns. These findings suggest that AI-assisted thematic analysis may provide less contextual representation of experience-based dimensions of nursing practice within the specific analytic procedure used in this study. Rather than fully capturing the emotional and relational meanings embedded in participants’ narratives, AI-assisted interpretation tended to represent complex nursing experiences through more generalized thematic structures. Such differences in representation may raise ethical considerations regarding how participants’ experiences are interpreted within qualitative nursing research. Examples of interpretive differences are shown in Table 4. Table 4Examples of differences between human reflexive interpretation and AI-assisted thematic representationParticipant narrativeHuman reflexive interpretationAI-assisted thematic representationEthical implicationDifferent nurses may record the same intervention in different ways, making it difficult to unify the data.Reflects challenges in standardizing nursing knowledge and preserving clinical context.Inconsistent nursing documentation.Loss of contextual meaning.Many details of nursing care are not reflected in the system records.Highlights invisible nursing work and limitations of digital representation.Incomplete documentation.Reduction of nursing experiences.Some research data look complete, but they do not necessarily reflect what actually happened in practice.Reveals the gap between recorded data and lived clinical experiences.Data quality issue.Reduction of experiential understanding.Clinical decisions often depend on communication, observation, and accumulated experience.Reflects tacit nursing knowledge and relational clinical judgment.Individualized nursing practice.Oversimplification of relational meaning. Examples of differences between human reflexive interpretation and AI-assisted thematic representation Algorithmic reductionism and cultural bias Participants additionally discussed challenges associated with standardizing nursing knowledge and translating complex clinical experiences into structured research data. Several participants emphasized that many nursing concepts, particularly those related to TCM nursing practice, rely heavily on clinical experience, contextual judgment, and individualized patient care. One participant stated: Some nursing concepts are easy to understand in practice but difficult to describe using standardized terms. (P6) Another participant explained: In TCM nursing, many decisions depend on syndrome differentiation and patient-specific conditions. It is difficult to fully express these situations through fixed categories. (P9) Human researchers interpreted these statements as reflecting the complexity of translating context-dependent nursing knowledge into standardized forms. Participants emphasized that nursing practice often involves implicit understanding, individualized judgment, and culturally embedded professional language that may not be fully captured through structured classifications. In contrast, AI-assisted thematic analysis tended to summarize these concerns using broader categories such as “communication differences,” “terminology inconsistency,” or “standardization challenges.” While these classifications accurately reflected the explicit content of participants’ statements, they provided limited insight into the professional, cultural, and experiential meanings underlying the narratives. Participants also highlighted that many nursing practices, particularly within TCM nursing, are influenced by local clinical traditions, communication styles, and context-specific decision-making processes. Human researchers interpreted these narratives as reflecting concerns regarding the preservation of culturally situated nursing knowledge. AI-assisted analysis, however, often reorganized these experiences into generalized semantic themes that reduced contextual specificity. Additional participants further emphasized the individualized and context-dependent nature of TCM nursing practice. One clinical nurse explained: Even within the same syndrome type, patients may have different physical constitutions. In current documentation, many records are written in a similar way or copied from templates, rather than fully reflecting individual differences. (P8) Another participant stated: It is difficult to prove that a patient’s improvement is caused by one single TCM nursing technique, because there are many influencing factors in clinical practice. (P9) These quotations further show that TCM nursing practice involves individualized judgement, patient-specific conditions, and multiple interacting clinical factors. Human researchers interpreted these accounts as reflecting culturally embedded and clinically situated meanings that may be simplified when AI-assisted analysis reorganizes them into broad categories such as “standardization challenges” or “data quality issues.” These findings suggest that algorithmic reductionism in qualitative nursing research should not be understood solely as a technical issue. Rather, AI-assisted interpretation may privilege standardized and semantically explicit forms of knowledge while underrepresenting context-dependent, culturally embedded, and experience-based dimensions of nursing practice. Such reduction may contribute to the marginalization of important forms of nursing knowledge during qualitative analysis. Changing roles and responsibilities in nursing research Participants frequently discussed the practical challenges associated with conducting nursing research, including data collection, data management, literature review, and research organization. Many participants acknowledged that digital technologies may help improve research efficiency and reduce administrative burden. One participant stated: Researchers spend a lot of time organizing data and reviewing records before they can start analysis. (P15) Another participant commented: Data management and information organization are often very time-consuming parts of the research process. (P16) Human researchers interpreted these statements as reflecting the potential value of digital technologies in supporting routine research activities. Participants recognized that technological tools may facilitate data management, improve efficiency, and support preliminary stages of the research process. Participants also described both the usefulness and risks of AI-assisted tools in nursing research. One participant noted: If I have a research idea, I can tell AI my general thinking and ask it to help me organize the idea. Sometimes the structure it provides is clearer than what I thought of by myself. (P7) Another participant expressed caution regarding AI-generated information: Doesn’t it sometimes fabricate data or references? (P10) These quotations suggest that participants viewed AI as a useful assistant for organizing research ideas and supporting early-stage research work. At the same time, they also recognized the need for human verification and professional judgement. This supports the interpretation that AI-assisted tools may reshape the research process, but researchers remain responsible for checking, interpreting, and ethically using AI-generated outputs. AI-assisted thematic analysis similarly identified themes related to “research efficiency,” “workflow optimization,” and “technology support.” However, reflexive comparison revealed important differences in interpretation. Human researchers emphasized that qualitative inquiry involves more than organizing information; it also requires contextual interpretation, critical reflection, and ethical engagement with participants’ experiences. Several participants additionally suggested that digital tools should function as supportive resources rather than replacements for researchers’ judgment and expertise. Human researchers interpreted these narratives as reflecting concerns regarding professional responsibility and the continued importance of human interpretation in nursing research. AI-assisted analysis, however, tended to frame these concerns primarily as issues of supervision or quality control. Reflexive comparison suggested that increasing reliance on AI-assisted analysis may gradually alter how researchers engage with qualitative data. While AI technologies may effectively support data organization and preliminary thematic exploration, participants emphasized the continued importance of researcher involvement in interpreting meaning, understanding context, and representing participants’ experiences. These findings highlight the need to maintain human interpretive responsibility within AI-assisted qualitative inquiry and support a collaborative rather than replacement-oriented role for AI in nursing research. Discussion This study explored how AI-assisted thematic analysis may influence the interpretation, representation, and ethical understanding of nursing narratives through a human–AI comparative analysis of qualitative nursing interviews. By comparing human reflexive interpretation with ChatGPT-assisted thematic representation, the findings demonstrated that AI-assisted analysis offers substantial advantages in analytic efficiency, rapid data organization, and preliminary thematic exploration. However, important limitations emerged in the interpretation of emotionally situated experiences, tacit nursing knowledge, contextual meanings, and culturally embedded communication. More importantly, the findings suggest that AI-assisted qualitative inquiry is not merely a technical analytic process but also an interpretive and epistemological process that may reshape how nursing experiences are represented during qualitative analysis. Reflexive comparison revealed broader concerns regarding contextual reductionism, interpretive authority, narrative ownership, and the evolving relationship between researchers, technology, and qualitative data in nursing inquiry. Unlike previous studies primarily focusing on coding agreement or analytic efficiency [17], the present study adopted a reflexive and interpretive perspective toward AI-assisted thematic analysis. Rather than evaluating whether AI could “accurately” replicate human coding, this study examined how AI participation may reshape the interpretation and ethical representation of nursing experiences. The findings suggest that although AI can function as a supportive analytic assistant, qualitative nursing research still depends heavily on human contextual understanding, emotional interpretation, and reflexive engagement. Advantages of AI-assisted qualitative research in nursing AI improves analytic efficiency and preliminary data organization Consistent with existing literature, ChatGPT-assisted analysis demonstrated potential advantages in accelerating early-stage qualitative analysis [6, 28]. From the research team’s observation, AI-assisted workflows appeared to reduce the time required for preliminary data organization and initial coding, although no structured measurements of time, cost, or workload were collected. This efficiency may be particularly valuable in nursing research environments where qualitative data are increasingly complex, resource-intensive, and time-sensitive. In nursing research contexts where qualitative datasets are increasingly large and resource-intensive, AI-assisted analysis may provide practical support for researchers managing substantial textual materials. Previous studies have similarly suggested that LLMs may improve analytic efficiency by facilitating rapid coding and preliminary thematic mapping [27–28]. The present findings support these observations and suggest that AI-assisted approaches may be particularly useful during the exploratory stages of nursing qualitative analysis. Importantly, however, the findings also indicate that AI-generated coding structures should not be treated as finalized analytic outputs. Human researchers remained essential for refining codes, reorganizing thematic relationships, and contextualizing participants’ narratives. Similar to findings reported by De Paoli [5], AI-generated outputs frequently required further interpretive revision and reflexive evaluation by researchers. Thus, AI-assisted thematic analysis may best function as a collaborative and iterative analytic process rather than an autonomous replacement for qualitative interpretation. AI assists in identifying explicit thematic patterns Another advantage observed in this study was AI’s ability to rapidly identify explicit semantic patterns across multiple interview transcripts. ChatGPT effectively recognized recurring topics related to privacy concerns, system limitations, digital workload, and AI-related uncertainty. AI-generated summaries were often coherent at the descriptive level and demonstrated strong capacity for organizing large amounts of textual information into structured categories. These findings align with prior studies suggesting that LLMs may support thematic clustering and pattern recognition in qualitative datasets [31]. In particular, AI-assisted thematic analysis may help researchers identify broad thematic directions that can later be refined through reflexive interpretation. Such capabilities may be valuable in implementation research, nursing informatics, and rapid qualitative inquiry where timely synthesis is necessary. However, the findings also demonstrated that AI-generated themes were generally more descriptive than interpretive. AI frequently summarized what participants explicitly said but struggled to capture why participants expressed certain emotions or how contextual meanings shaped their narratives. Consequently, AI-assisted thematic exploration may support descriptive pattern recognition, whereas interpretive meaning-making in nursing qualitative research still depends heavily on human contextual engagement and reflexive understanding. Limitations of AI-Assisted Qualitative Analysis AI demonstrates limited contextual and emotional interpretation One of the most prominent findings of this study was the limited capacity of AI to interpret emotionally nuanced and contextually embedded nursing narratives. Nursing interviews often contained implicit meanings, emotionally layered experiences, and tacit professional knowledge that extended beyond explicit semantic content. Human researchers were more likely to identify emotional undertones, situational meanings, and relational dynamics embedded within participants’ narratives within the context of this study. For example, when participants described nursing experiences as “difficult to standardize” or “hard to explain clearly,” human researchers interpreted these statements as reflecting tacit nursing knowledge, emotional labor, and experiential clinical reasoning. In contrast, AI-generated interpretations frequently reduced such narratives to generalized descriptions of communication difficulties or system limitations. These findings are consistent with previous concerns that LLMs may oversimplify qualitative narratives and struggle with interpretive meaning-making [29]. This limitation is particularly important in nursing research because nursing practice is inherently relational, emotional, and context-dependent. Many participants described nursing knowledge as embodied, experiential, and relational, suggesting that important dimensions of nursing practice may not be fully reducible to standardized textual interpretation. Unlike highly structured textual data, nursing narratives often rely on implicit communication, emotional expression, and situational understanding developed through clinical experience. AI-generated analysis may therefore risk simplifying emotional complexity and representing nuanced experiences through broader conceptual categories. AI may oversimplify tacit nursing knowledge The present findings additionally demonstrated that AI-assisted thematic analysis may inadequately represent tacit forms of nursing knowledge. Participants frequently described experiential judgments, intuitive communication strategies, and relational care practices that were difficult to articulate through standardized language. Human researchers interpreted these accounts as reflecting embodied clinical knowledge and contextual professional expertise. AI-generated analysis, however, often translated these experiences into generalized procedural or organizational issues. Previous qualitative scholars have emphasized that meaning in qualitative inquiry is not merely located in explicit textual content but also in silence, emotional tone, contextual relationships, and experiential understanding [30–31]. AI systems trained primarily through probabilistic language prediction may struggle to capture these relational and experiential dimensions of nursing practice. Consequently, excessive reliance on AI-generated thematic summaries may risk reconstructing complex nursing experiences into analytically organized yet epistemologically simplified forms of interpretation. AI-generated outputs remain inconsistent and prompt-dependent Another limitation observed in this study involved the variability and inconsistency of AI-generated outputs. Different prompts occasionally produced different coding structures and thematic summaries despite using the same interview data. Some AI-generated outputs prioritized concise summaries, whereas others produced broader descriptive categories or repetitive themes. Multiple rounds of prompting and refinement were often required before AI-generated outputs became sufficiently coherent for comparative analysis. These findings align with previous research demonstrating that LLM outputs are highly sensitive to prompt design and contextual instructions [32–33]. Without clear prompting structures and continuous researcher supervision, AI-generated analyses may become inconsistent, overly generalized, or disconnected from participants’ original meanings. Thus, methodological rigor in AI-assisted qualitative analysis requires transparent documentation of prompts, iterative reflexive review, and continuous human oversight throughout the analytic process. Ethical concerns in AI-assisted nursing qualitative research AI raises concerns regarding contextual representation and interpretive integrity Beyond methodological limitations, the findings of this study suggest that AI-assisted thematic analysis raises broader ethical concerns regarding the representation of participants’ lived experiences. Existing AI ethics discussions in health research have primarily focused on technical concerns such as privacy, transparency, and algorithmic bias [24]. While these issues remain important, the present study indicates that ethical concerns in qualitative nursing research also involve interpretive integrity and contextual representation. In qualitative inquiry, participants’ narratives are not merely informational data but contextual representations of emotional, relational, and lived experiences [9]. When AI-generated analysis simplifies emotionally nuanced narratives or removes contextual complexity, there is a risk that participants’ experiences may be inaccurately represented within research findings. This issue extends beyond technical coding accuracy and relates fundamentally to how qualitative meaning is ethically reconstructed during analysis. Participants’ concerns therefore extended beyond technical privacy and reflected broader uncertainty regarding who ultimately controls, interprets, and represents nursing narratives once qualitative data enter AI-mediated analytic systems. The findings therefore suggest that contextual misrepresentation should be understood as an important ethical issue in AI-assisted qualitative nursing research. Ethical rigor in qualitative inquiry requires preserving the complexity, ambiguity, and contextual depth of participants’ narratives rather than reducing them into simplified semantic categories. The advantages and limitations of AI-assisted qualitative analysis are summarized in Table 5. Table 5Advantages and limitations of AI-assisted qualitative analysis in nursingDimensionAdvantagesLimitationsData organizationRapid organization and management of large volumes of qualitative dataPotential loss of contextual details during data reductionPreliminary codingEfficient identification of explicit concepts and recurring patternsOverlapping, generalized, or decontextualized codesTheme generationRapid generation of initial thematic structuresLimited interpretive depth and reflexive engagementInterpretation of nursing experiencesRecognition of explicit meanings within narrativesDifficulty capturing tacit knowledge and experiential understandingContextual understandingEfficient summarization of textual contentOversimplification of situational and relational meaningsResearch workflowReduced workload and improved analytic efficiencyDependence on prompt design and researcher oversightCultural and professional understandingBasic identification of culturally related contentDifficulty interpreting culturally embedded and context-dependent nursing knowledge Advantages and limitations of AI-assisted qualitative analysis in nursing AI may reinforce algorithmic and cultural biases Another important finding of this study involved concerns regarding algorithmic bias and cultural limitations in AI-assisted qualitative analysis. Participants frequently emphasized that nursing communication in Chinese clinical contexts often relies on indirect language, contextual implication, and emotionally embedded expression. Human researchers interpreted these culturally situated meanings through their professional and clinical experiences within the study context, whereas AI-generated analyses tended to emphasize more explicit semantic patterns. These findings are consistent with previous studies suggesting that LLMs trained predominantly on English-language datasets may demonstrate limitations when analyzing culturally specific or non-English qualitative data [10–11]. AI systems are shaped by the linguistic structures, sociocultural assumptions, and textual patterns embedded within their training data. Consequently, AI-generated interpretations may unintentionally reproduce dominant linguistic norms while inadequately recognizing context-specific expressions used in localized nursing practice. In the present study, this limitation became particularly apparent when participants discussed experiential nursing judgment, emotional communication, and context-dependent clinical interactions. Human researchers frequently interpreted these narratives as reflecting relational care practices and tacit professional knowledge, whereas AI-generated analyses often translated them into generalized themes related to communication challenges or workflow issues. In this sense, AI systems may privilege semantically recognizable forms of expression while underrepresenting relational, ambiguous, and culturally situated dimensions of nursing communication. The findings therefore raise broader concerns regarding epistemic bias in AI-assisted qualitative research, particularly when working with non-Western nursing contexts and culturally embedded narratives [34]. Ethical concerns surrounding AI-assisted qualitative research should therefore include not only technical bias but also interpretive and cultural bias related to meaning representation. AI may alter the reflexive role of qualitative researchers The findings additionally suggest that AI-assisted qualitative analysis may influence the traditional role of researchers in qualitative inquiry. Reflexive thematic analysis emphasizes that researchers actively co-construct meaning through ongoing interaction with the data [8]. Interpretation in qualitative research is shaped not only by textual analysis but also by researchers’ positionality, professional experience, emotional engagement, and reflexive reflection [18]. Participants in this study expressed concern that excessive reliance on AI-generated themes could reduce researchers’ immersion in qualitative data and weaken reflexive engagement with participants’ narratives. Human researchers in this study repeatedly revisited interview transcripts, reflected on contextual meanings, and modified thematic interpretations throughout the analytic process. In contrast, AI-generated outputs frequently appeared definitive and structurally coherent despite lacking deeper interpretive grounding. This issue reflects broader methodological concerns regarding the outsourcing of qualitative interpretation to algorithmic systems [35]. Unlike traditional CAQDAS software, which primarily supports data organization, LLMs actively generate interpretations and thematic narratives. Consequently, researchers may become increasingly dependent on AI-generated conceptual structures during analysis. Such dependence may unintentionally shift interpretive authority away from human reflexivity toward algorithmically generated outputs. Such shifts may gradually transform the relationship between researchers and qualitative data from immersive reflexive engagement toward technologically mediated thematic reconstruction. Importantly, the findings of this study do not suggest that AI should be excluded from qualitative nursing research. Rather, the results indicate that reflexive human engagement remains essential when integrating AI into qualitative inquiry. AI-generated outputs should be critically reviewed, contextually interpreted, and reflexively examined rather than accepted as objective analytic findings. The proposed reflexive relationship framework is illustrated in Fig. 2. Reflexive relationship between human researchers and AI-assisted analysis Implications for AI-assisted qualitative nursing inquiry The findings of this study have several implications for future nursing qualitative research. First, AI-assisted qualitative analysis should be understood as a collaborative rather than autonomous analytic approach. Human researchers remain essential for contextual interpretation, emotional understanding, and ethical representation of participants’ narratives. AI may support preliminary coding and thematic organization, but interpretive responsibility should remain grounded in human reflexive engagement. This interpretation is consistent with recent nursing-specific evidence on the educational and professional implications of ChatGPT. Sarıkahya et al. reported that ChatGPT may support nursing education by improving efficiency, facilitating educational support, and assisting with content generation. However, they also highlighted concerns regarding critical thinking, academic integrity, and overreliance on AI technologies. These findings support the need to position ChatGPT as a supportive tool rather than a replacement for human expertise, interpretation, and professional judgement [36]. Second, future nursing qualitative studies using AI-assisted approaches should prioritize methodological transparency. Clear documentation of prompts, analytic workflows, iterative refinements, and researcher involvement is necessary to improve trustworthiness and analytic rigor in AI-assisted qualitative inquiry. Similar to audit trails in traditional qualitative research, AI-assisted studies should preserve records of prompt development, thematic revisions, and reflexive decision-making processes. Third, nursing researchers should remain attentive to the ethical implications of contextual reductionism during AI-assisted analysis. Nursing narratives frequently involve emotional complexity, relational care, and tacit professional judgment that may not be fully captured through automated semantic interpretation. Reflexive consideration of how AI-generated interpretations reshape participants’ narratives is therefore essential for maintaining ethical integrity in nursing qualitative research. Finally, future approaches should prioritize collaborative human–AI frameworks that preserve reflexive interpretation, contextual sensitivity, and ethical representation while benefiting from the analytic efficiency of AI-assisted thematic exploration. These findings should be understood within the specific procedure used in this study, including one AI tool, one prompt protocol, one nursing interview dataset, and one cultural context. They should therefore not be interpreted as universal evidence regarding all large language models or all applications of AI-assisted qualitative analysis. Limitations This study has several limitations. First, the study focused on ChatGPT-assisted thematic analysis and did not compare multiple AI models. The AI-assisted analysis was also based on one prompt protocol. AI-generated outputs are prompt-dependent, and different prompt structures, levels of contextual information, or iterative strategies may produce different coding and thematic results. Therefore, the findings should be interpreted within the specific model, dataset, prompt protocol, and cultural context used in this study. Future studies may compare different LLMs, including Chinese-language models such as DeepSeek, particularly in relation to culturally embedded nursing narratives and non-English qualitative analysis. Second, this study did not collect structured data on time, cost, or workload during the human-led and ChatGPT-assisted analytic processes. Therefore, efficiency-related observations should be interpreted as research team observations rather than measured outcomes. Although ChatGPT-assisted analysis appeared to support preliminary data organization and thematic structuring, it also required additional work related to transcript anonymization, output checking, contextual verification, and reflexive interpretation. Third, the interview data were collected within a nursing research context in China, which may limit transferability to other cultural or clinical settings. Nursing communication styles, emotional expression, and contextual meanings may differ across health care systems and sociocultural environments. Fourth, AI technologies continue to evolve rapidly, and the interpretive capabilities of future LLMs may differ from those examined in the present study. Consequently, findings regarding AI-assisted qualitative analysis should be interpreted within the context of the tool and procedure used in this study. Finally, although reflexive thematic analysis was used to enhance interpretive rigor, qualitative analysis inevitably remains shaped by researchers’ positionality, assumptions, and contextual understanding. The findings should therefore be understood as reflexive interpretations rather than objective representations of AI performance. Conclusion This study explored the opportunities, limitations, and ethical implications of AI-assisted qualitative analysis in nursing through a human–AI comparative thematic analysis of nursing interviews. The findings suggest that ChatGPT-assisted thematic analysis may support preliminary coding, thematic exploration, and organization of qualitative data. However, within the procedure used in this study, important limitations remained in its ability to represent emotionally nuanced, contextually embedded, and culturally situated nursing narratives. Reflexive comparison showed that ChatGPT-assisted interpretation often organized nursing experiences into semantically coherent but contextually simplified thematic representations. Ethical concerns therefore extended beyond technical accuracy and included contextual reductionism, interpretive integrity, algorithmic bias, and the changing relationship between researchers and qualitative data. Overall, within the procedure used in this study, ChatGPT-assisted thematic analysis appeared useful as a supportive tool for preliminary thematic exploration and data organization. However, it did not provide interpretations equivalent to those developed through human reflexive engagement. Human researchers remained important for contextual interpretation, ethical judgement, and the representation of nursing experiences. Given that this study involved one AI tool, one prompt protocol, one dataset, and one specific cultural context, the findings should not be generalized to all large language models or all applications of AI in qualitative research. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 2