Learners' experiences of generative artificial intelligence-supported simulation and skills training in nursing education: a meta-synthesis

Synthesis: Generative AI is moving into nursing simulation and skills training faster than evidence about learner experiences can accumulate, leaving implementation dependent on learner-centered evidence about value, limits, and safety. This meta-synthesis of 15 reports representing 14 studies synthesizes nursing learners' experiences of GenAI-supported Simulation and skills training. Four analytical themes emerged: accessible rehearsal with perceived gains in confidence and readiness (moderate confidence); AI-mediated interaction supporting communication, reasoning, and case application (high confidence); perceived realism bounded by persistent relational and technical limits (moderate confidence); and verification, Feedback, and Reflection as perceived safeguards (moderate confidence). The authors conclude that learners' experiences support treating Generative AI and Conversational AI tools as supervised rehearsal partners, tied to explicit Guardrails, educator oversight, Pedagogical Safety, and human oversight, with direct implications for Nursing Education and Medical and Health Professions Education.

Key Findings

1. Scope. Fifteen reports representing 14 studies were included; database searches identified 689 records, with 44 duplicates removed before screening.
2. Four themes, four confidence ratings. The synthesis generated four analytical themes and 13 descriptive subthemes, rated moderate to high via GRADE-CERQual.
3. Accessible rehearsal (nine studies, moderate). Learners valued accessible, repeatable, low-pressure practice linked to perceived gains in confidence and readiness — one called it "a safe space to learn."
4. AI-mediated interaction (12 studies, high). Learners reported support for clinical reasoning and assessment, knowledge application, patient-centered communication, and ethical, cultural, and professional judgment.
5. Realism with limits (nine studies, moderate). Responses could feel lifelike, but robotic voices, absent body language, response lag, and audio failures constrained authenticity.
6. Perceived safeguards (ten studies, moderate). Learners verified generated content against other sources and valued immediate specific feedback and structured debriefing.
7. Recent and global corpus. The 14 studies were published in 2025 (n = 3) or 2026 (n = 11) across eight countries or regions; the United States contributed four.

Why nursing education is adopting GenAI in simulation

Simulation-based education lets nursing students combine clinical reasoning with communication and technical performance before patient care, a rehearsal that matters when placements offer few chances to manage high-risk situations. Generative AI extends this opportunity because, unlike earlier rule-based systems, it can generate context-responsive content as an interaction unfolds: large language and multimodal models can construct cases, enact virtual patients, vary dialogue, answer learner questions, and generate natural-language feedback. Virtual patients can disclose information gradually, vary their emotional responses, and follow different trajectories across attempts, and the World Health Organization identifies simulated patient encounters as a potential application of large multimodal models in medical and nursing education. Yet fluency does not guarantee quality — outputs can be inaccurate, biased, or poorly aligned with clinical context — so responsible integration requires a clear pedagogical purpose, AI literacy, and verification of generated content.

The meta-synthesis method

Searches covered MEDLINE, Embase, CINAHL, PsycINFO, and ERIC, with citation tracking, from November 2022 — when ChatGPT was publicly released — through June 21, 2026. Eligibility used the SPIDER framework, quality was appraised with the JBI Critical Appraisal Checklist for Qualitative Research, findings were synthesized using Thomas and Harden's thematic synthesis, and confidence was assessed with GRADE-CERQual. All 14 studies met JBI criteria for methodological congruity and ethical conduct. Activities spanned chatbots, immersive virtual reality, text- and avatar-based virtual patients, adaptive narratives, case analysis, role-play, and a retrieval-augmented generation (RAG) case system; communication was the most common focus. Qualitative data came from interviews, focus groups, open-ended responses, written reflections, debriefing records, and learner–GenAI transcripts.

Four themes of perceived educational value

The first theme, accessible rehearsal, describes GenAI-supported activities as accessible, repeatable, and comparatively low-pressure; learners valued revisiting difficult encounters without immediate consequences for patients, linking rehearsal to reduced anxiety and perceived readiness. The second, AI-mediated interaction, gathered the strongest evidence: across 12 studies, learners described interpreting clinical, emotional, and contextual cues, reconsidering judgments as adaptive narratives varied, applying theory to cases, and adapting their language and pacing to virtual patients. The third theme captures the trade-off between realism and limits: lifelike responses coexisted with robotic voices, absent body language, overly cooperative patients, and technical interruptions. The fourth theme, perceived safeguards, shows learners checking generated content against prior learning, instructors, or guidelines, and using feedback to guide another attempt.

Safeguards and the supervised-rehearsal-partner conclusion

Learners' perception of educational value was shaped by low-pressure practice, responsive interaction, and opportunities to evaluate generated content; relational limitations, technical disruption, and unreliable outputs constrained those experiences. Verification, feedback, and reflection operated as complementary forms of critical engagement — learners wanted sources and explanations rather than "black box" outputs, and feedback and debriefing were valued when specific enough to guide the next attempt. The authors conclude that learner experiences support using GenAI as a supervised rehearsal partner alongside established simulation modalities, contingent on educator-reviewed scenarios, prebriefing on system limitations, and structured debriefing. They favor staged implementation within defined learning activities — beginning with repeatable dialogue or formative feedback and progressing to human-facilitated practice.

What this means for practice

• Instructors. Position GenAI as a supervised rehearsal partner: prebrief learners on system limitations and source checking, and debrief afterward, since learners treated verification, feedback, and reflection as safeguards.
• Curriculum designers. Favor tasks that require learners to explain and justify responses, and retain human interaction where nonverbal communication or sustained empathy is central.
• Administrators. Stage implementation within defined learning activities before progressing to human-facilitated practice.
• Researchers. Run longitudinal and comparative studies of observed competence, retention, clinical transfer, trust calibration, and equity.

Limitations

• All 14 studies were published in 2025 or 2026, and many involved short or single-session exposures, limiting insight into sustained learning.
• Several findings relied on brief open-ended responses rather than in-depth interviews, with uneven data adequacy for feedback, reflection, and human oversight.
• Researcher positioning was not consistently reported, and novelty and social desirability may have favored positive accounts.
• GenAI roles ranged from interactive virtual patients to case-based and procedural support, so theme relevance may vary; the learner-only focus also excluded educator perspectives.
• Publication bias may underrepresent unsuccessful implementations.

Citation

Zhang, Y., Tang, M., Cao, S., Yang, L., Hu, Y., Chen, Q., & Li, Y. (2026). Learners' experiences of generative artificial intelligence-supported simulation and skills training in nursing education: a meta-synthesis. BMC Medical Education (Springer Nature), open access.