Clinical nurses' practices and roles when patients use artificial intelligence for health information: A scoping review
Year: 2026
Type: preprint
Abstract: Background Patients now routinely consult artificial intelligence (AI) tools, particularly conversational agents and large language models, to interpret symptoms, laboratory results, and treatment options. A growing share of this information is carried into clinical encounters, where it may complement, complicate, or openly contradict the advice given by clinicians. Nurses are frequently the first point of contact for these patients and the professionals who spend most time at the bedside, so they are the ones who must recognise that AI-generated content is in play, judge its credibility and clinical applicability, discuss it without undermining the therapeutic relationship, and decide what responsibility they bear when a patient acts on it. Evidence on the adjacent phenomenon in medicine is already synthesised: a scoping review of symptom checkers used by laypersons summarised their impacts on primary care physicians. In nursing, by contrast, the literature remains fragmented. A recent case study described a patient who uploaded laboratory images to a conversational agent and received advice that conflicted with the clinical evidence, and proposed communication and handover protocols in response; several small qualitative studies and commentary pieces address related situations. To our knowledge, no evidence synthesis has mapped how clinical nurses actually respond to patient-initiated AI use, which practices they adopt, which roles they assume, and where the ethical and accountability gaps lie. Because the phenomenon is recent, conceptually unsettled, and studied through heterogeneous designs, a scoping review is the appropriate design at this stage. Review questions 1.In what forms does patient-generated AI health information appear in clinical nursing settings (types of tools, content, patients' motives, and the ways it is presented to nurses)? 2.How do nurses recognise and appraise such information, and what communication and relationship-maintaining strategies do they use? 3.What roles do nurses assume in these situations, how prepared do they feel, and what ethical and accountability issues arise? 4.How is the available evidence distributed across populations, settings, and study designs, and where are the gaps? Eligibility criteria (Population, Concept, Context) Population. Registered nurses, advanced practice and specialist nurses, community and public health nurses, and nursing staff in direct patient care, including nurse managers who still hold clinical duties. Nurse students, nursing educators without clinical responsibilities, and non-nursing professionals are excluded. Concept. Nurses' practices and roles when patients or family members use AI to obtain health information, operationalised in four domains: (1) recognition and judgement, that is, noticing that AI-generated information is being used, appraising its credibility and applicability, and deciding whether to adopt, partly adopt, reject, or refer for confirmation; (2) communication and trust, that is, strategies such as explaining, shared decision making, avoiding, or responding authoritatively, together with changes in the nurse-patient relationship and the emergence or resolution of disagreement; (3) role and preparedness, that is, perceived role boundaries, knowledge and skills, training needs, and organisational support; (4) ethics and accountability, that is, responsibility when AI-derived advice leads to harm, duties of disclosure and documentation, privacy, equity in digital health literacy, and regulatory gaps. Context. Clinical settings at any level of care, including inpatient, outpatient, emergency, community, and long-term care, in any country, in which patients or family members use AI tools such as conversational agents, large language models, symptom checkers, or health question-and-answer applications, and bring the resulting information into their interaction with nurses. Purely online consultations without nurse involvement are excluded. Language. Records in Chinese or English will be included; the search itself will not be language restricted. Rule for determining the user of the AI tool To separate the phenomenon of interest from the far larger literature on nurses' own use of AI, studies will be classified into three tiers. Tier A: the patient or family member is the AI user and the study reports nurse-side data on at least one of the four domains; included. Tier B: nurses are the participants and the study addresses preparedness, training needs, or role perceptions regarding patients' AI use, without reporting observed encounters; included and labelled separately. Tier C: nurses use AI as a clinical tool for decision support, documentation, or triage; generic surveys of AI attitudes or AI literacy without any patient-side use; excluded, with reasons recorded. Search strategy Eleven databases will be searched: PubMed (MEDLINE), Embase, CINAHL Complete, Scopus, Web of Science Core Collection, PsycINFO, Cochrane CENTRAL, SinoMed, CNKI, Wanfang Data, and VIP. The strategy is built on the PCC elements as four blocks: nurses; the patient as AI user; AI tools; and the four concept domains. The main search combines the first three blocks (population, patient-side use, AI tools). The concept block is not used as a filter, because concept terms such as communication, role, and ethics are pervasive in nursing literature and would remove approximately half of the records; instead it is applied as a sensitivity check, in which titles in the difference set are reviewed to confirm that no key study has been missed. Free-text terms include generic and proprietary tool names, among them ChatGPT, Gemini, Copilot, DeepSeek, Doubao, Ernie Bot, Qwen, and Kimi. Controlled vocabulary was verified against the MeSH thesaurus; because ChatGPT has no MeSH descriptor, and because the descriptors for generative AI and large language models were introduced only in 2025, free-text searching is essential for records published before 2025. The strategy will be reviewed by a health sciences librarian using the PRESS checklist, and the full line-by-line strategy with per-database yields will be reported in an appendix. Reference lists of included studies and of relevant reviews will be hand-searched, and forward citation tracking will be performed. Study selection Records will be exported to Rayyan, and duplicates removed. Two reviewers will independently screen titles and abstracts after a calibration exercise on 50 records, then screen full texts independently. Inter-rater agreement will be reported; disagreements will be resolved by discussion, with a third reviewer consulted when needed. Exclusion reasons will be recorded at full-text stage and summarised in a PRISMA flow diagram. Data extraction A charting form will be piloted on five studies and refined. Data will be charted at three levels: study characteristics (author, year, country, setting, design, sample, data collection); encounter characteristics (AI tool type, patient motive, form of the information, whether it conflicted with clinical evidence); and the four concept domains, with verbatim exemplars retained for the synthesis. Where a study reports both nurse-side and patient-side data, only nurse-relevant findings will be charted. Synthesis Findings will be summarised descriptively: counts by year, country, setting, design, and tool type; an evidence map crossing the four domains with study design and care setting; and a narrative synthesis organised by domain, following JBI guidance for scoping reviews. The number of studies providing direct evidence on AI-related disagreement in nurse-patient encounters will be reported explicitly as a finding, and gaps will be stated in terms of domain, population, setting, and method. Timeline Search completion: December 2026. Screening: January 2027. Charting: February 2027. Synthesis and manuscript: March 2027. Submission: April 2027. Ethics and reporting The review uses published literature only and does not require ethics approval. Reporting will follow PRISMA-ScR, with the search reported according to PRISMA-S. Dissemination Submission to a peer-reviewed nursing journal, together with presentation at a national or international nursing conference. The charting form, the full search strategies, and the evidence map will be deposited with the registration. Funding and conflicts of interest [待填：基金名称与编号，或写 This review received no specific funding.] The authors declare no conflicts of interest. Deviations Any change to the eligibility criteria, search strategy, or synthesis plan after registration will be reported in the resulting manuscript, with the date and the reason for the change.
Source: OSF Preprints (OSF Preprints)
Author: YuXiao Zhu
Language: English
Subfield: Health Informatics
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7224845005