Generative AI at the Bedside: An Integrative Review of Applications and Implications in Clinical Nursing Practice

ABSTRACT

Aim

The aim of this integrative review is to critically appraise and synthesise empirical evidence on the clinical applications, outcomes, and implications of generative artificial intelligence in nursing practice.

Design

Integrative review following Whittemore and Knafl's five‐stage framework.

Methods

Systematic searches were performed for peer‐reviewed articles and book chapters published between 1 January 2018 and 30 June 2025. Two reviewers independently screened titles/abstracts and full texts against predefined inclusion/exclusion criteria focused on generative artificial intelligence tools embedded in nursing clinical workflow (excluding nursing education‐only applications). Data were extracted into a standardised matrix and appraised for quality using design‐appropriate checklists. Guided by Whittemore and Knafl's integrative review framework, a constant comparative analysis was applied to derive the main themes and subthemes.

Data Sources

CINAHL, MEDLINE, and Embase.

Results

Included literature was a representative mix of single‐group quality improvement pilots, mixed‐method usability and feasibility studies, randomised controlled trials, qualitative descriptive and phenomenological studies, as well as preliminary and proof‐of‐concept observational research. Four overarching themes emerged: (1) Workflow Integration and Efficiency, (2) AI‐Augmented Clinical Reasoning, (3) Patient‐Facing Communication and Education, and (4) Role Boundaries, Ethics and Trust.

Conclusion

Generative artificial intelligence holds promise for enhancing nursing efficiency, supporting clinical decision making, and extending patient communication. However, consistent human validation, ethical boundary setting, and more rigorous, longitudinal outcome and equity evaluations are essential before widespread clinical adoption.

Implications for the Profession and Patient Care

Although generative artificial intelligence could reduce nurses' documentation workload and routine decision‐making burden, these gains cannot be assumed. Safe and effective integration will require rigorous nurse training, robust governance, transparent labelling of AI‐generated content, and ongoing evaluation of both clinical outcomes and equity impacts. Without these safeguards, generative artificial intelligence risks introducing new errors and undermining patient safety and trust.

Reporting Method

PRISMA 2020.

Summary.

• What does this paper contribute to the wider global community?

  • ○
    Synthesises international evidence on generative AI applications in bedside nursing, highlighting both benefits and critical risks.
  • ○
    Defines essential safeguards (e.g., human oversight, robust governance, equity evaluation) for safe, effective clinical integration.
  • ○
    Sets a global research agenda by identifying key gaps in long‐term outcomes, economics, and patient equity.
• Impact 1.1: What problem did the study address?

  • ○
    This study addressed the fragmented and under‐synthesised evidence base on how generative artificial intelligence (GenAI) is actually used in day‐to‐day clinical nursing practice. Specifically, it examined the benefits, risks, and implementation challenges GenAI tools pose for nursing care at the bedside.
• Impact: 1.2: What were the main findings?

  • ○
    All GenAI interventions in nursing required ongoing human validation and clear ethical boundaries.
  • ○
    GenAI tools could demonstrate potential to reduce documentation and routine task burden, streamline workflows, and improve patient communication, but evidence of long‐term patient outcomes, economic impacts, and equity was limited.
  • ○
    Safe and effective GenAI integration depends on robust nurse training, governance, transparent content labeling, and systematic evaluation of clinical and equity outcomes.
• Impact: 1.3: Where and on whom will the research have an impact?

  • ○
    This research may impact frontline nurses, nurse leaders, and healthcare organisations by providing practical guidance for integrating generative AI safely and ethically at the bedside.
  • ○
    Findings can inform policymakers, educators, and researchers globally on critical gaps and priorities, supporting evidence‐based decision‐making and workforce development.
  • ○
    Ultimately, patients and health systems worldwide will likely benefit from safer, more effective, and more equitable AI‐enabled nursing care.

1. Introduction

Generative artificial intelligence (GenAI) refers to advanced technologies that can produce new content, such as text, images, and simulations, by analysing and summarising vast datasets, and these tools are now increasingly accessible in healthcare environments (Shen et al. 2025). The rapid advancement of GenAI may hold significant promise for transforming various aspects of healthcare delivery (Saban and Dubovi 2024). For example, applications like ChatGPT, which utilise large language models (LLM), are able to interpret a wide range of patient information communicated in natural language, including free‐text clinical notes, symptom descriptions, past medical history, medication lists, laboratory results, and imaging indications. These applications can then generate relevant clinical responses. This capability could demonstrate potential for supporting tasks such as rapid diagnostic suggestions and imaging referral recommendations through intuitive, conversational interfaces (Rosen and Saban 2023, 2024).

Artificial intelligence (AI) is now increasingly prevalent in healthcare, and turning a blind eye is no longer viable (Douglas et al. 2023; Jackson 2025; Russell et al. 2022; von Gerich et al. 2022; Watson 2024, 2025). The critical conversations in nursing have shifted from whether we should use emerging technologies to how we should apply them responsibly and what core values must be safeguarded along the way (Jackson 2025; Johnson and Galatzan 2025; Watson 2024; Yip et al. 2025).

Artificial intelligence has been progressively integrated into nursing informatics and critical care practice in multiple forms. For example, early predictive models embedded in electronic health records (EHRs) have been used to flag patients at high risk of sepsis or falls, allowing nurses to intervene earlier in the clinical trajectory (Douglas et al. 2023; Russell et al. 2022; von Gerich et al. 2022). Expert systems such as sepsis early warning scores and drug‐interaction alerts have supported nurse decision‐making by providing real‐time prompts at the bedside (Watson 2024). Natural language processing (NLP) is also increasingly applied in documentation workflows, such as ambient listening technologies that automate progress notes, reducing nurses' time spent on clerical tasks (Peterson et al. 2023; Watson 2024). Similarly, machine learning algorithms embedded in clinical dashboards can detect subtle physiological trends and notify nurses of deteriorating perfusion before overt shock develops (Douglas et al. 2023; Stokes and Palmer 2020). These tools highlight how AI is already shaping daily surveillance and monitoring responsibilities. In acute care settings, AI‐powered chatbots have been piloted to answer routine patient questions and provide mental health support, freeing nurses for more complex care. However, risks have also been reported; some chatbots failed to provide appropriate responses in sensitive scenarios such as self‐harm or abuse disclosures (Freitas and Araújo 2024; Sweeney et al. 2021). Within critical care pharmacology, nurses are beginning to encounter AI‐assisted titration technologies. Closed‐loop vasopressor systems and AI‐driven monitoring dashboards are being trialed to adjust infusion rates in real time, shifting the nurse's role toward supervisory oversight and rapid troubleshooting rather than manual titration (Joosten 2022; Porcellato et al. 2025; Watson 2025 ). Studies show these systems can improve stability but still require nurse judgement to override automation when patient safety is at risk (Almagharbeh et al. 2025; Vincelette et al. 2025).

Nursing scholarship also highlights the ethical tensions of this technological shift. Concerns include algorithmic bias, such as cardiovascular algorithms that under‐diagnose women's myocardial infarction symptoms (Starke et al. 2021), and unsafe chatbot responses that encourage harmful patient decisions (Watson 2024). Scholars emphasise that while GenAI and other AI systems offer opportunities for efficiency, they must be implemented in ways that protect nurses' relational roles, empathy, and professional judgement (Michalowski et al. 2025; Pepito et al. 2023; Watson 2025).

Thus, as GenAI becomes more advanced and deeply embedded in clinical systems, nursing must move beyond hesitation and toward meaningful engagement. Our task is to harness these innovations in ways that enhance patient outcomes and support nurses to give the best possible care (Jackson 2025). As noted by Ronquillo et al. (2021), it is vital that nurses are not passive recipients of change, but rather active participants in guiding the adoption of AI within care environments.

GenAI is increasingly trialled as a partner in nursing workflows, challenging and reshaping the foundational practices and values of the profession (Alruwaili et al. 2025; Bienefeld et al. 2024; McDonald 2024; Watson 2024). However, as McCormack (2025) warns, the adoption of digital healthcare technologies can sometimes create an ‘illusion of progress,’ where technological innovation is mistaken for genuine improvements in care quality. Accordingly, critical reflection is key as GenAI becomes further integrated into clinical nursing practice.

1.1. Rationale and Gap

Given the rapid advancement and increasing integration of GenAI in various healthcare settings, it is essential to understand the scope of the empirical literature on its applications in clinical nursing practice. This integrative review is timely and necessary to ensure that technological innovation translates into meaningful improvements in both nursing practice and patient outcomes. The aim of this integrative review is to critically appraise and synthesise empirical evidence on the clinical applications, outcomes, and implications of generative artificial intelligence (GenAI) in nursing practice.

1.2. Review Question

What empirical evidence exists regarding the clinical applications, outcomes, and implications of generative artificial intelligence in nursing practice since 2018?

2. Methods

2.1. Design

This integrative review was guided by the five‐stage methodological framework proposed by Whittemore and Knafl (2005), which supports a systematic approach to identifying, appraising, and synthesising diverse forms of empirical evidence, including quantitative, qualitative, and mixed‐methods studies. The integrative review methodology was purposefully selected over a scoping review to enable not only a comprehensive mapping of the literature but also an in‐depth critical appraisal and thematic synthesis of findings. This approach facilitated the evaluation of study quality, comparison across methodological designs, and the extraction of nuanced themes and patterns related to GenAI implementation in clinical nursing practice. Integrating and interpreting heterogeneous evidence allowed the results of this review to move beyond cataloguing existing knowledge and into developing practice‐relevant insights, identifying gaps, and offering evidence‐based recommendations for clinical application, policy, and future research.

2.2. Search Strategy

Systematic searches were conducted across three databases: Cumulative Index to Nursing and Allied Health Literature (CINAHL), Medical Literature Analysis and Retrieval System Online (MEDLINE), and Embase. This comprehensive search strategy was designed by an academic health sciences librarian using the following search string:

(genAI OR “gen AI” OR “generative AI” OR “artificial intelligence, generative”) AND nurs* AND (patient OR “clinical decision making” OR “clinical practice” OR “clinical applications” OR documentation)

Limitations were placed to retrieve literature exclusively published between 1 January 2018 through 30 June 2025, reflecting the release of generative GenAI technologies. The final search was performed on 30 June 2025. Searches were augmented through citation‐tracking, reference‐list checks, and a structured review of relevant grey literature, including book chapters containing primary empirical data.

2.3. Inclusion and Exclusion Criteria

Included studies met the following criteria: (1) articles were published in English from January 2018 onward, (2) articles focused explicitly on GenAI tools (e.g., GPT‐based chatbots, large language models) used by nurses in direct clinical workflows, and (3) articles or book chapters were directly relevant to clinical practice, including workflow efficiency, clinical decision‐making, patient care or communication.

Date range of publications were from 1 January 2018 onward because 2018 marks the first year in which transformer‐based and other GenAI models were publicly released, ushering in a new class of AI technologies distinct from earlier predictive or rule‐based systems. Professional and scholarly dialogue about applying these generative systems in healthcare, particularly within nursing contexts, also began in 2018. Aligning the search window with this technological and conceptual inflection point ensures that included studies evaluate genuinely GenAI interventions relevant to contemporary nursing practice.

This integrative review also excluded studies whose primary aim was to use GenAI for pre‐licensure, postgraduate, or continuing education purposes (e.g., virtual tutoring, simulation debriefing, examination item generation). This was integrated into the search design to better capture interventions that could be embedded in direct clinical workflow (e.g., documentation, triage, decision support, patient communication) as these applications have the most immediate implications for bedside practice, workload, and patient outcomes. Bracketing out education‐specific research helped to maintain conceptual coherence in the synthesis and avoided conflating pedagogical innovations or concerns with tools designed to augment day‐to‐day clinical care.

Book‐chapter contributions located during screening were treated as grey literature. Chapters that presented novel primary empirical data not duplicated in a peer‐reviewed journal were retained and subjected to the same quality‐appraisal rubric as the articles in the review; purely narrative, conceptual, or duplicative chapters were excluded.

2.4. Screening and Selection

After removing duplicates, two independent reviewers (AW and CB) conducted initial title and abstract screenings based on predefined inclusion/exclusion criteria. Conflicts at the title/abstract stage were resolved by consensus or by a third reviewer (DJ). Full‐text screening followed the same independent, dual‐reviewer approach, with conflicts again resolved through consensus discussion. Reasons for exclusion at the full‐text level were documented systematically. Screening reproducibility was assessed at both stages. Dual independent reviewers achieved 88% agreement at the title/abstract stage and 92% agreement at full‐text screening, with discrepancies resolved by a third reviewer. These agreement rates demonstrate sufficient reliability. The screening and selection process adhered to the PRISMA 2020 checklist, and outcomes are summarised in a PRISMA flow diagram (See Figure 1).

FIGURE 1.

PRISMA 2020 flow diagram. [Colour figure can be viewed at wileyonlinelibrary.com]

2.5. Critical Appraisal/Quality Assessment

Given the methodological diversity of the included studies, the research team used design‐specific quality appraisal tools recommended by the Joanna Briggs Institute (JBI 2017) Critical Appraisal Checklists. Quantitative studies (RCTs, pre/post, cross‐sectional) were evaluated for internal validity, sampling, measurement reliability, and bias. Qualitative studies were assessed on credibility, transferability, confirmability, and dependability. Mixed‐method studies underwent dual appraisal, assessing both qualitative and quantitative rigour.

To enhance transparency, appraisal outcomes are reported in aggregate. JBI checklist scores across included studies ranged from 6 to 11 out of 13 possible points (mean = 8.7), indicating overall moderate‐to‐high methodological quality. As recommended by Whittemore and Knafl (2005), appraisal results informed interpretation of findings rather than exclusion of studies.

2.6. Ensuring Rigour

Methodological rigour and trustworthiness were enhanced by maintaining a comprehensive audit trail documenting each decision point in the review process. Dual‐independent reviewer processes for screening, selection, quality appraisal, and data extraction were strictly followed. Regular team discussions ensured consensus on inclusion criteria and quality appraisal in order to help minimise bias, maximise rigour, and maintain transparency in decisions. Reflexivity was ensured by acknowledging and discussing researcher assumptions regarding GenAI's potential clinical benefits and challenges throughout analysis meetings. The standardised data extraction matrix was piloted on three studies by the review team, refined collaboratively, and then applied to the full set of included studies. This step ensured inter‐reviewer consistency and helped to minimise the risk of interpretive bias during data abstraction.

2.7. Data Extraction and Synthesis

Data from included articles was systematically extracted into a standardised matrix. documenting author(s), year, study aims/objectives, settings, participants, design, data collection, analytical methods, and significant findings (see Table 1).

TABLE 1.

Data extraction matrix.

Author/Country	Research objectives	Sample/Setting	Study design	Data collection	Analysis	Key findings
Alruwaili et al. (2025) Saudi Arabia	Explore NICU nurses' experiences with AI	33 NICU nurses, 4 units	Interpretive phenomenology	Interviews, focus groups	Thematic analysis	AI supports decisions, changes workflow, requires infrastructure/training, supplements nursing expertise
Bienefeld et al. (2024) Switzerland	Compare data scientists' and clinicians' views on AI in ICUs	19 data scientists, 61 ICU clinicians, 6 ICUs	Multimethod	Observations, survey, interviews	Consensus, grounded theory, stats	AI welcomed for monitoring, documentation; nurses want control for patient care
Chen et al. (2024) Taiwan	Improve documentation with ChatGPT tool	ICU & ward nurses, Taiwan	Development/implementation	Needs assessment, feedback	Descriptive/qual	Time cut by 66%, improved accuracy/workflow, nurse oversight crucial to success
Chen et al. (2025) Hong Kong	Compare AI chatbot versus nurse hotline for anxiety/depression	124 parents, Hong Kong	Pilot RCT	Pre/post questionnaires	t‐tests, regression	Chatbot measured similar effects to nurse in mental health hotline 24/7 support, reduces anxiety/depression
Chenard et al. (2024) USA	Test ChatGPT as safe/effective for post‐op ortho triage	3 nurses, ChatGPT, 3 surgeons	Preliminary	Nurse/AI responses rated by surgeons	ANOVA, t‐tests	AI and nurse triage responses judged as similar in safety and completion of responses
Elhilali et al. (2025) Switzerland	Evaluate GenAI‐powered triage (STT + LLM) in ED	Simulated triage cases, experts	Proof of concept	Audio, AI processing	Accuracy (WER, ESI), comparison	Noted potential for high triage accuracy; may reduce nurse documentation and workload
English et al. (2024) USA	Assess large language model (LLM) usefulness for care teams, prompt optimization	166 users, 9 clinics, 2023–24	Prospective QI	LLM draft replies, user surveys	ANOVA, JMP Pro	Nurses found AI helpful for efficiency, empathy, and tone in communication
Soddu et al. (2025) Italy	Assess ChatGPT‐4 on pressure injuries in infants	Expert evaluation in simulation	Cross‐sectional	AI responses, expert review	Mixed‐methods	AI may be a safe and effective adjunct for clinical support
Shah‐Mohammadi and Finkelstein (2025) USA	Evaluate ChatGPT‐4o for extracting symptoms/signs from nursing notes	MIMIC‐III dataset	Experimental	Automated data extraction	Accuracy, reliability stats	AI can extract key info for clinical informatics
Saban and Dubovi (2024) Israel	Explore ChatGPT as clinical support for nurses	30 ER RNs, 38 students	Cross‐sectional, scenario‐based	Online Qs, AI responses	Comparative, descriptive	Highlights need for careful AI integration in nursing support
Lee et al. (2024) Korea	Test generative AI for reducing repetitive new nurse documentation tasks	Univ. hospital, 30 nurses, 50 K EHR entries	Mixed‐methods, pilot.	EHR data, user surveys	Descriptive stats, qual feedback	AI reduced documentation, improved usability, facilitated patient care
Lee and Shin (2025) Korea	Develop/evaluate chatbot for coronary artery disease recovery Q&A	5 experts, 27 post‐PCI patients	Mixed‐methods usability/feasibility	Usage logs, surveys, interviews	Descriptive, qualitative	Chatbot supported education and post‐discharge support, reducing clinician workload
Huang et al. (2024) USA	Assess LLMs (e.g., GPT‐4) for patient‐friendly ED discharge instructions, patient perspective	156 MTurk respondents; randomised to GPT‐4 versus standard instructions	Blinded, randomised survey	Survey on interpretability, satisfaction	Statistical survey analysis	AI instructions rated higher for clarity and patient satisfaction compared to non‐GPT discharge instructions
Garcia et al. (2024) USA	Evaluate LLM use for drafting inbox replies to patient messages	Academic medical center; attendings, APCs, nurses, pharmacists	Prospective, single‐group QI	EHR log analysis, clinician surveys	Pre‐post, paired stats	AI reduced clinician burden; improved communication; oversight and ongoing risk evaluation needed

Data synthesis was guided by the constant comparative analytic method, a hallmark of integrative reviews, facilitating iterative identification and refinement of emergent themes across the studies. Each extracted study was repeatedly compared with others to identify commonalities, contrasts, and patterns. Initial thematic codes were independently generated by team members, followed by collaborative discussions to achieve interpretive consensus. Through this iterative comparison and categorization, a coherent thematic structure reflecting GenAI use in clinical nursing practice emerged, ensuring analytical rigour and interpretative depth.

3. Results

3.1. Study Characteristics

Articles chosen for this review totaled 14 academic articles (n = 10) and book chapters (n = 4). The articles in this review encompassed a wide range of research designs, including single‐group quality improvement pilots, mixed‐methods usability and feasibility studies, randomised controlled trials, cross‐sectional surveys, and qualitative inquiries such as interpretive phenomenology. Those studies were conducted in varied clinical settings (e.g., emergency departments, intensive care units (ICUs), neonatal intensive care units (NICUs), and cardiology), and primarily assessed workflow efficiency, clinical reasoning enhancement, patient communication, and ethical or professional implications of GenAI. The included studies originated from a diverse range of international contexts, reflecting a global interest in GenAI applications within clinical nursing practice. Specifically, articles originated primarily in the United States (n = 6), followed by Korea (n = 2), Switzerland (n = 2), and one article each from Israel, Hong Kong, Taiwan, and Saudi Arabia. See Figure 2.

FIGURE 2.

International origins of articles. [Colour figure can be viewed at wileyonlinelibrary.com]

This geographical diversity underscores the widespread recognition of GenAI's potential in nursing practice internationally, while also highlighting the need for future research to explicitly address cross‐cultural differences, resource variability, and context‐specific implementation challenges to enhance the generalizability and applicability of findings across diverse healthcare settings. See Table 2 for an overview of themes.

TABLE 2.

Overview of main themes and subthemes.

Main theme	Subthemes	Empirical evidence	Main points
Workflow integration & efficiency	1.1 Documentation time‐savings	Implementation of the “A+ Nurse” documentation tool reduced average charting time from roughly 15 min to 5 min per patient (Chen et al. 2024), and a parallel generative‐AI pilot demonstrated comparable efficiency gains among newly qualified nurses (Lee et al. 2024)	Automation demonstrably releases nurse time for direct care; time‐motion metrics should be core outcomes
	1.2 Messaging & inbox relief	LLM‐draft replies trimmed composition time while preserving empathy (Garcia et al. 2024; English et al. 2024) AI wrote patient‐message drafts so quickly that nurses and doctors spent less time typing while still sounding caring and empathetic	Embedding AI in comms reduces invisible cognitive load. Need to track impact on burnout
	1.3 Infrastructure, training & policy enablers	Seamless Epic/Nursing Information System integrations; 84.8% NICU nurses requested formal AI training (Alruwaili et al. 2025) While the AI system plugged easily into the hospital's Epic/NIS software, nearly 85% of NICU nurses said they still needed formal training to feel prepared to use it	Realising efficiency improvements is contingent on dependable information technology infrastructure, clear governance, and comprehensive workforce training
AI‐augmented clinical reasoning	2.1 Symptom/sign extraction	GPT‐4o mined nurse notes with high accuracy (Shah‐Mohammadi and Finkelstein 2025). (i.e., GPT‐4o could read nurses' ICU notes and correctly pick out the important symptoms and signs most of the time)	AI can structure free‐text but must embed validation checkpoints

3.2. Main Theme 1: Workflow Integration & Efficiency

Nurses play a central role in guiding how GenAI tools are implemented, ensuring these systems are meaningfully aligned with clinical needs and patient care priorities (Alruwaili et al. 2025; Bienefeld et al. 2024). Their expertise is essential for interpreting, validating, and contextualising AI‐generated content, as well as identifying potential workflow disruptions or unintended consequences (Chen et al. 2024; Lee et al. 2024; English et al. 2024). Without nurse leadership in both the adoption and ongoing evaluation of GenAI technologies, the risk remains that efficiency gains may be superficial, or offset by new burdens and safety concerns, rather than resulting in truly improved patient outcomes (Alruwaili et al. 2025; Garcia et al. 2024).

3.2.1. Subtheme 1.1: Documentation Time‐Savings

While GenAI‐powered tools such as the “A+ Nurse” documentation assistant show substantial promise in reducing documentation workload, these efficiency gains must be interpreted with caution. Chen et al. (2024) reported that A+ Nurse reduced average charting time from approximately 15 to 5 min per patient, with similar gains among newly qualified nurses in a separate GenAI pilot (Lee et al. 2024).

Automated systems cannot replicate the adaptability of experienced clinical judgement, contextual adaptation, and patient‐specific insights that underpin high‐quality nursing records (Chen et al. 2024; Lee et al. 2024; Alruwaili et al. 2025; Bienefeld et al. 2024). Nurses' expertise is crucial for validating, editing, and contextualising GenAI‐generated documentation, ensuring that essential clinical details are neither omitted nor misrepresented (Chen et al. 2024; Lee et al. 2024). As such, time savings should never come at the expense of thoroughness, individualised care, or safety; ongoing human oversight by nurses is essential to prevent errors, omissions, or the loss of critical information, safeguarding both patient outcomes and professional standards (Alruwaili et al. 2025; Bienefeld et al. 2024; Chen et al. 2024; Lee et al. 2024).

3.2.2. Subtheme 1.2: Messaging & Inbox Relief

GenAI systems designed to draft patient messages and manage clinical inboxes may show potential to reduce the time required for communication tasks, with the goal of still maintaining empathy and a professional tone (English et al. 2024; Garcia et al. 2024). In both related studies, clinicians reported that GenAI‐assisted messaging streamlined workflows. Clinicians also experienced a decrease in the “invisible” cognitive load of large message volumes, known contributors to communication‐related burnout. However, these apparent gains must be interpreted with caution.

Automated responses may introduce errors, misinterpretations, or send inappropriate information to patients (Garcia et al. 2024). Nurses play a critical role in reviewing, editing, and contextualising GenAI‐generated content. English et al. (2024) and Garcia et al. (2024) observed that oversight is necessary to preserve clinical quality. Another priority in these scenarios is to ensure that sensitive or complex issues are addressed appropriately. There is also a need for clear protocols to escalate patient queries that exceed the capabilities of automated systems. Ultimately, these considerations may serve to reinforce the ongoing importance of nurse involvement in patient communication (Alruwaili et al. 2025; Garcia et al. 2024).

3.2.3. Subtheme 1.3: Infrastructure, Training & Policy Enablers

Despite evidence that GenAI tools can be integrated with existing hospital infrastructure such as Epic and Nursing Information Systems, successful and safe implementation is far from guaranteed. Alruwaili et al. (2025) found that nearly 85% of NICU nurses expressed the need for formal GenAI training to feel competent and confident in using these technologies. This highlights that infrastructure alone is insufficient; sustained investment in comprehensive nurse education, explicit governance policies, and clear operational frameworks is critical to ensure both technical competency and ethical accountability (Alruwaili et al. 2025).

3.3. Main Theme 2: AI‐Augmented Clinical Reasoning

The capacity of GenAI to augment clinical reasoning is emerging as one of its most promising, and most closely scrutinised, contributions to nursing practice. Across the reviewed literature, multiple studies illustrate how GenAI tools are being deployed to extract and synthesise clinical data, support complex triage decisions, and provide real‐time guidance for clinical scenarios (Elhilali et al. 2025; Saban and Dubovi 2024; Shah‐Mohammadi and Finkelstein 2025). These applications highlight GenAI's potential to enhance the speed, consistency, and accuracy of clinical assessments, particularly in structured or protocol‐driven environments. At the same time, evidence consistently emphasises the necessity for robust human oversight, careful system validation, and explicit safeguards to prevent automation bias and ensure that technology complements, rather than supplants, professional nursing judgement (Bienefeld et al. 2024; Chenard et al. 2024; Soddu et al. 2025).

3.3.1. Subtheme 2.1: Symptom/Sign Extraction

Recent advances in GenAI demonstrate strong potential to enhance the speed and accuracy of extracting clinically relevant information from complex datasets. For example, Shah‐Mohammadi and Finkelstein (2025) found that ChatGPT‐4o was able to extract structured clinical data, such as symptoms and signs, from free‐text nursing notes in a given dataset. This capability could facilitate real‐time clinical analytics, supporting rapid identification of patient deterioration or evolving conditions. Similarly, studies by Chen et al. (2024) and Lee et al. (2024) show that GenAI‐powered tools can streamline the extraction and organisation of documentation in everyday nursing workflows. However, the literature consistently emphasises the need for regular validation of these systems by clinical experts, to safeguard against misclassification, ensure clinical relevance, and maintain the accuracy and integrity of extracted data (Lee et al. 2024; Shah‐Mohammadi and Finkelstein 2025).

3.3.2. Subtheme 2.2: Triage Decision Support

GenAI is also making significant inroads in supporting clinical triage and decision‐making. Elhilali et al. (2025) reported that an integrated speech‐to‐text and LLM pipeline achieved 90%–100% concordance with expert benchmarks for Emergency Severity Index (ESI) acuity classification and chief complaint coding in simulated emergency scenarios. In real‐world applications, Chenard et al. (2024) found that ChatGPT‐generated replies for postoperative orthopaedic triage were rated comparable to those drafted by experienced triage nurses. Yet, there is a persistent need for human oversight to avoid automation bias and to ensure that AI‐supported decisions are contextually appropriate and tailored to each patient's unique presentation (Bienefeld et al. 2024; Chenard et al. 2024; Elhilali et al. 2025).

3.3.3. Subtheme 2.3: Clinical Q&A/Pressure‐Injury Coaching

GenAI‐driven clinical Q&A tools are emerging as effective adjuncts to nursing expertise, providing rapid, evidence‐informed guidance in complex clinical scenarios. Soddu et al. (2025) found that expert review affirmed the accuracy of ChatGPT‐4‐generated clinical advice related to pressure injuries in infants, while Saban and Dubovi (2024) demonstrated that ChatGPT scored comparably to expert ER nurses in standardised decision‐making vignettes, though its performance varied with case complexity. These findings indicate that GenAI tools can supplement nursing judgement, particularly for well‐defined clinical questions and education. Still, the need for robust clinical oversight remains paramount, as the quality and appropriateness of GenAI outputs can fluctuate in more ambiguous or atypical cases (Bienefeld et al. 2024; Saban and Dubovi 2024; Soddu et al. 2025).

3.4. Main Theme 3: Patient‐Facing Communication & Education

As GenAI becomes more embedded in clinical practice, its impact on patient‐facing communication and education is increasingly evident (Bhuyan et al. 2025; Park et al. 2024). The reviewed literature highlights how GenAI technologies can enhance the clarity, accessibility, and continuity of information provided to patients across the care continuum. For example, studies have shown that large language models can generate discharge and after‐visit instructions that are clearer and more understandable than standard summaries, thereby improving patient comprehension and engagement (Huang et al. 2024). Additionally, GenAI‐powered chatbots are emerging as valuable tools to support patients through recovery phases, offering sustained, personalised education and support after hospital discharge (Lee and Shin 2025). Early evidence further suggests that GenAI interventions, such as AI‐driven chatbots, have the capacity to deliver accessible and continuous mental health and crisis support, augmenting traditional services and expanding the reach of nursing care (Chen et al. 2025; Garcia et al. 2024).

3.4.1. Subtheme 3.1: Discharge & After‐Visit Instructions

Recent studies demonstrate that large language models, such as GPT‐4, can generate patient discharge instructions that are not only clearer and more readable than standard documents, but also rated higher for satisfaction by lay readers (Huang et al. 2024). This improvement in health communication suggests a meaningful opportunity to enhance patient understanding, engagement, and self‐management after clinical encounters. Ethical and practical considerations emphasise the importance of transparent labeling of AI‐generated materials (Huang et al. 2024), and several studies recommend that patients be informed when content is produced or supported by GenAI to maintain trust and accountability (Garcia et al. 2024; Park et al. 2024). Continued nurse oversight is required to ensure instructions are contextually appropriate and culturally sensitive.

3.4.2. Subtheme 3.2: Recovery‐Phase Chatbots

GenAI‐powered chatbots are emerging as highly usable, patient‐centered tools for delivering ongoing education and self‐care support during the recovery phase. For instance, Lee and Shin (2025) found that patients recovering from coronary artery interventions reported high satisfaction and frequent use of a chatbot designed to answer personalised recovery questions, particularly about medication, activity, and follow‐up care. The scalability of such interventions allows education and support to be extended beyond traditional clinical boundaries without placing additional burdens on nursing staff (Bhuyan et al. 2025). However, effective deployment requires that chatbots are designed with input from clinicians and patients to ensure accuracy and relevance, and ongoing evaluation of engagement and long‐term adherence is warranted (Garcia et al. 2024; Lee and Shin 2025).

3.4.3. Subtheme 3.3: Mental Health & Crisis Support

There is growing evidence that GenAI chatbots can offer effective and accessible mental health and crisis support, supplementing existing nurse‐led services. In a randomised controlled pilot study, Chen et al. (2025) reported that AI chatbot use significantly reduced anxiety and depression symptoms among users, with effects comparable to those seen with nurse hotlines. These tools provide 24/7 availability and scalability during surges in demand, which is particularly valuable in settings with limited mental health resources (Bhuyan et al. 2025). Nonetheless, the literature stresses the need for ongoing human oversight, clear protocols for escalation and safety, and attention to ethical standards, such as transparency and equity, especially for vulnerable populations (Garcia et al. 2024; Park et al. 2024).

3.5. Main Theme 4: Role Boundaries, Ethics & Trust

The integration of GenAI into nursing practice is prompting a re‐examination of professional boundaries, ethical responsibilities, and the core values that underpin safe, effective care. Two critical dimensions emerge: a shift in nurses' roles and autonomy, and the ongoing imperative to uphold safety, transparency, and explainability in AI‐enabled workflows.

3.5.1. Subtheme 4.1: Role Evolution & Autonomy

Across multiple studies, GenAI is shown to be reshaping nursing roles—from routine, task‐focused work to more analytic, consultative, and supervisory functions. For example, Alruwaili et al. (2025) found that NICU nurses increasingly described their role as validating and contextualising AI‐generated recommendations, positioning themselves as critical decision‐makers. Similarly, Chen et al. (2024) and Lee et al. (2024) highlight that nurses act as essential validators of AI outputs, exercising clinical judgement to ensure recommendations are appropriate and safe. This trend is further supported by Shah‐Mohammadi and Finkelstein (2025), who observed the indispensable role of nurses in confirming the accuracy of AI‐driven symptom extraction, and Soddu et al. (2025), who reported that nurse oversight remained critical even with high‐performing AI tools for clinical assessment.

3.5.2. Subtheme 4.2: Safety, Transparency & Explainability

A consistent theme across the literature is the need for robust clinical validation and clear communication of GenAI outputs. Alruwaili et al. (2025) reported that nearly all NICU nurses insisted on clinician confirmation of AI‐generated recommendations, reflecting a recognition that AI can support, but not replace, human judgement in safeguarding patient safety. Transparent labeling of AI‐generated content is also crucial: Garcia et al. (2024) found that clinicians valued explicit identification of AI‐authored messages, while Huang et al. (2024) noted that clear disclosure increased patient trust and satisfaction. Additional studies (Chen et al. 2024; Shah‐Mohammadi and Finkelstein 2025) highlight the importance of explainability and traceability, with nurses needing to understand and interrogate the rationale behind AI recommendations to ensure safe and effective integration.

4. Discussion

4.1. GenAI's Transformative Potential and Workflow Efficiency

This integrative review, consistent with Bhuyan et al. (2025), highlights GenAI's transformative potential in nursing, particularly in the potential for streamlining workflows and possibly even reducing the documentation burden. Studies by Chen et al. (2024) and Lee et al. (2024) suggest that GenAI‐assisted tools can reduce charting time by as much as 66%. Such a reduction could enable nurses to shift their primary focus from administrative tasks to much‐needed clinical care and direct patient interactions. Efficiency gains can extend to patient messaging, where English et al. (2024) and Garcia et al. (2024) found that large language models may generate empathetic, timely draft replies, which could help to decrease communication‐related workload and potentially mitigate burnout.

4.2. Decision Support, Clinical Informatics, and Patient Communication

Beyond workflow efficiency, GenAI has shown promise in supporting clinical reasoning, triage, and patient communication. In structured, protocol‐driven scenarios, studies by Elhilali et al. (2025), Shah‐Mohammadi and Finkelstein (2025), and Saban and Dubovi (2024) found GenAI systems can match clinician accuracy for symptom extraction and triage. In patient education, Lee and Shin (2025) and Huang et al. (2024) document high ratings for clarity and satisfaction with AI‐generated discharge instructions and chatbots, pointing to scalable solutions for ongoing support. Chen et al. (2025) further demonstrated that an AI chatbot was as effective as a nurse hotline for reducing anxiety and depression, with the added benefit of 24/7 accessibility.

4.3. Quality of Evidence and Research Gaps

Although early outcomes from GenAI integration in nursing are promising, critical analysis reveals significant methodological and reporting limitations. Most studies remain confined to single‐center settings, short‐term pilots, or simulated environments, limiting their generalizability to real‐world clinical practice (Chen et al. 2024; Elhilali et al. 2025; Lee et al. 2024). Given that generative AI in healthcare and nursing is an emerging field, much of the current literature consists of early‐stage pilot studies and exploratory research, which contribute to the heterogeneity and limited rigour observed in the evidence base.

The evidence base is dominated by descriptive, observational, and quality improvement studies, with rigorous randomised controlled trials and robust longitudinal research still rare (Chen et al. 2025; Huang et al. 2024). As a result, it is difficult to draw firm conclusions about causality, long‐term effectiveness, or potential harms. Patient‐level outcome data, formal economic evaluations, and analyses of equity impacts are largely absent, raising concerns given the widespread claims of GenAI's efficiency and scalability (Bhuyan et al. 2025; Garcia et al. 2024; Lee and Shin 2025; Rogers and Baker 2025). The lack of attention to differential effects across populations and algorithmic bias is a particularly significant gap, as highlighted in recent reviews (Park et al. 2024; Ruksakulpiwat et al. 2024). Without systematic evaluation of these domains (Aveyard 2023), there is a risk of exacerbating disparities or introducing new threats to patient safety and workforce well‐being (Alruwaili et al. 2025; Bienefeld et al. 2024; McDonald 2024; Watson 2024). In sum, the current evidence base remains provisional, underscoring the urgent need for greater methodological rigour, longer‐term follow‐up, and focused attention to economic, legal, and equity‐related outcomes (Aveyard and Bradbury‐Jones 2019, 2021).

4.4. Critical Safeguards for Ethical and Reliable GenAI Implementation

Across the literature, robust safeguards are emphasised as essential for ethical, safe, and reliable GenAI integration in nursing. Human oversight and clinical validation are consistently identified as necessary, especially in patient‐facing or high‐risk scenarios (Alruwaili et al. 2025; Bienefeld et al. 2024; Soddu et al. 2025). Oversight by nurses and clinicians is required to maintain communication quality, empathy, and accountability, even when efficiency gains are achieved (English et al. 2024; Garcia et al. 2024). Prerequisites for safe integration include comprehensive nurse training in AI literacy, explicit governance frameworks, transparent labeling of AI‐generated content, and ongoing monitoring for equity and patient safety (Bhuyan et al. 2025; Hoelscher 2024; Huang et al. 2024; Lee et al. 2024; Park et al. 2024; Rogers and Baker 2025; Ruksakulpiwat et al. 2024). Validation checkpoints, error correction, and feedback systems must be built into AI workflows to support accountability and rapid response to emerging risks (Chenard et al. 2024; Shah‐Mohammadi and Finkelstein 2025). These safeguards are the minimum requirements for GenAI to enhance, rather than undermine, person‐centered nursing care. See Table 3.

TABLE 3.

Recommended safeguards for ethical GenAI integration in nursing practice.

Ethical safeguards	Description
Human oversight	All GenAI outputs should be validated and contextualised by nurses/clinicians before use in care decisions
Transparent labeling	AI‐generated documentation, discharge instructions, and patient messages should be explicitly identified as such
Nurse training & AI literacy	Nurses should receive regular, structured education on GenAI tools, prompt crafting, ethics, and critical appraisal
Governance & policy frameworks	Institutions must implement clear policies on accountability, escalation protocols, and safe use boundaries
Equity monitoring	Ongoing evaluation of algorithmic bias and impact across populations should be considered essential to prevent disparities
Feedback & error correction	Systems should include mechanisms for reporting, correcting, and learning from AI‐related errors

4.5. The Risk of Depersonalization: Preserving Relational Care

A recurring concern is the potential for GenAI to depersonalise patient care, especially if AI systems replace or mediate key nurse–patient interactions. Strong professional resistance remains to using AI as an autonomous interface for patient engagement, with consensus that direct care should remain fundamentally human (Bienefeld et al. 2024). Nurses value GenAI's workflow benefits but insist on maintaining oversight to ensure empathy and relational care are not lost (English et al. 2024). Overreliance on AI risks eroding the human connection central to nursing, while nurse practitioners are uniquely positioned to balance technology with patient‐centered, compassionate care (Rogers and Baker 2025).

4.6. What This Integrative Review Adds: Functionalization and Conceptual Advances

Unlike prior scoping reviews that mapped the range of AI applications, this integrative review critically appraises study quality and synthesises empirical findings to clarify how GenAI is being implemented, where it adds value, and what boundaries are required for safe, person‐centered adoption (Bhuyan et al. 2025; Hoelscher 2024; McDonald 2024; Park et al. 2024; Ruksakulpiwat et al. 2024; Watson 2024). The review moves beyond theoretical discussion to highlight GenAI's practical impact on workflow, decision support, communication, and mental health interventions (Chen et al. 2024; Chen et al. 2025; English et al. 2024; Garcia et al. 2024; Huang et al. 2024; Lee et al. 2024; Lee and Shin 2025). By contextualising these findings within nursing's core values and professional identity, this review frames relational care and clinical judgement as non‐negotiable red lines for automation (Bienefeld et al. 2024; English et al. 2024; Watson 2024). In light of ongoing calls for governance and transparency, the review identifies the need for economic evaluation, equity analysis, and long‐term outcome studies to guide responsible implementation (Bhuyan et al. 2025; Garcia et al. 2024; Huang et al. 2024; Lee and Shin 2025; McDonald 2024; Park et al. 2024; Rogers and Baker 2025; Ruksakulpiwat et al. 2024).

5. Strengths and Limitations

This integrative review is grounded in Whittemore and Knafl's (2005) recognised methodology, ensuring systematic processes for search, selection, data extraction, and synthesis (Aveyard and Bradbury‐Jones 2021). All steps, including review question development, inclusion/exclusion criteria, and search strategy, are transparently documented, supporting methodological rigour and reproducibility. Data extraction and critical appraisal were performed with standardised tools and dual independent reviewers, minimising bias and enhancing validity. Thematic synthesis was iterative and collaborative, maintaining focus on comparability and quality appraisal. Adherence to contemporary reporting guidelines further strengthens the interpretive clarity and transparency of the review.

However, some limitations remain. The search was limited to English‐language studies from three databases (CINAHL, MEDLINE, Embase), which may have led to the omission of relevant evidence from other languages or sources. While grey literature (book chapters) was included, rapidly evolving GenAI innovations and unpublished studies may have been missed, introducing potential publication bias. Most included studies were short‐term pilots or quality improvement projects with limited sample sizes and follow‐up, restricting conclusions about long‐term effectiveness and generalizability. Few studies reported patient‐level outcomes, economic impacts, or equity analyses. Study quality and reporting were variable. To address these gaps, future reviews should expand database and language coverage, proactively seek unpublished or negative findings, and prioritise robust economic, equity, and longitudinal evaluations.

6. Nursing Implications

6.1. Nurse Clinicians

GenAI can ease documentation and routine task burdens, but nurses must remain actively involved in adoption, training, and daily validation of AI outputs (Alruwaili et al. 2025; Chen et al. 2024). Their judgement and feedback are essential for identifying workflow disruptions or threats to individualised care, and for ensuring patients are informed about GenAI's role in their care.

6.2. Nursing Leadership and Management

Clinical leaders must champion evidence‐informed, transparent GenAI integration, but also be willing to review, adjust, or discontinue use if technology does not deliver safe, equitable improvements (Alruwaili et al. 2025; Garcia et al. 2024). Leadership should empower nurses to participate in procurement, workflow mapping, and establish rapid feedback mechanisms, while safeguarding relational care and patient safety as non‐negotiable priorities.

6.3. Healthcare Organisations and Systems

Institutions should invest in information technology infrastructure, governance, and continuous professional development, as well as create reporting channels for issues and conduct equity audits to detect disparities (Park et al. 2024). Frontline nurse and patient input must guide all stages of selection and evaluation to ensure relevance and prevent unintended risks.

6.4. Nurse Educators and Professional Development Leaders

Education should combine GenAI training with deep emphasis on ethics, critical thinking, and relational care (Hoelscher 2024; Watson 2024). Simulation and hands‐on practice should address not just technical skills but also transparency, consent, and effective communication.

6.5. Policymakers and Nurse Researchers

Research should target real‐world outcomes, workflow, cost‐effectiveness, and equity (Bhuyan et al. 2025). Policymakers should mandate open reporting, safety monitoring, and inclusion of frontline nursing perspectives. Funding should support multisite, longitudinal studies with transparent sharing of positive and negative findings.

7. Conclusion

This integrative review highlights GenAI as a transformative force at the bedside, offering meaningful opportunities to enhance nursing workflow efficiency, support clinical decision‐making, and strengthen patient communication. However, fully realising these promising identified benefits depends critically on thoughtful integration, robust clinician training, explicit governance frameworks, and sustained ethical vigilance. As GenAI continues to evolve, the nursing profession must remain actively engaged in shaping its trajectory to reflect core nursing values and priorities. Strategic investment, inclusive collaboration, and commitment to equity will be essential to ensuring that GenAI serves as a tool for nurse empowerment and improved clinical care for patients. With collective leadership and a clear vision, GenAI can be harnessed to advance a future of nursing that is both technologically enabled and grounded in compassionate, person‐centered care.

Author Contributions

All authors have agreed on the final version and meet at least one of the following criteria: substantial contributions to conception and design, acquisition of data, or analysis and interpretation of data; drafting the article or revising it critically for important intellectual content.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1: jocn70151‐sup‐0001‐Supinfo.docx.

Watson, A. L. , Bond C., Aveyard H., Smith G. D., and Jackson D.. 2026. “Generative AI at the Bedside: An Integrative Review of Applications and Implications in Clinical Nursing Practice.” Journal of Clinical Nursing 35, no. 10: 4022–4037. 10.1111/jocn.70151.

Data Availability Statement

Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.