AI-Powered Simulation for Nursing Education: Mixed Methods Systematic Review

Synthesis: AI-Powered Simulation for Nursing Education: Mixed Methods Systematic Review — A PRISMA-guided mixed-methods systematic review by Jiang et al. (2026) synthesizing 19 studies (N = 1,253, mostly prelicensure nursing students) on AI-powered simulations in nursing education. Evidence from the three RCTs and controlled quasi-experimental studies shows significant gains in cognitive knowledge and affective outcomes (Self-Efficacy, communication confidence), but inconsistent effects on complex psychomotor skills — one RCT even found AI-assisted simulation inferior to standardized-patient simulation. Qualitative meta-aggregation reveals learners value safe, repeatable, nonjudgmental practice that bridges the theory–practice gap, while persistent "authenticity gap" frustrations (robotic dialogue, missing nonverbal cues, technical instability) mean AI should be a complement to — not a replacement for — traditional simulation and clinical placement.

Key Findings

• Positive but conditional evidence. AI-powered simulation significantly improved cognitive knowledge and affective outcomes (self-efficacy, communication confidence) in the strongest designs (3 RCTs + controlled quasi-experiments). Effects on complex psychomotor skills were inconsistent, and one RCT found AI-assisted simulation inferior to standardized-patient simulation for these.
• Four AI modality categories. The included studies spanned (1) generative AI/LLMs (n=7, e.g., ChatGPT-driven virtual patients), (2) AI-driven virtual patients/mannequins (n=5), (3) AI-enhanced virtual/mixed reality (n=5), and (4) chatbots (n=2) — a heterogeneous field that precluded quantitative meta-analysis.
• Learners value safe, repeatable practice. Qualitative meta-aggregation found students appreciated low-risk environments where they could attempt tasks repeatedly without judgment — building confidence and bridging the theory–practice gap — and valued consistent, repeatable AI interactions for foundational skills.
• The "authenticity gap." Learners consistently perceived AI interactions as lacking emotional depth, nonverbal cues, and tactile/physical-examination dimensions. Technical failures (e.g., speech-recognition delays) disrupted interaction flow and could raise extraneous cognitive load and state anxiety. This authenticity gap is why AI is best for highly structured objectives (foundational communication, history-taking, health education) and poorly suited to emotionally or interpersonally complex scenarios.
• Learner acceptance was moderate-to-high. Usability, acceptability, and perceived usefulness were consistently rated moderate to high across included studies (SUS and TAM measures), with learners recognizing value for Accessibility and structured guidance — but acceptance of AI feedback is conditioned by trust: learners prefer human feedback sources perceived as having "benevolence" and "integrity" over AI perceived as merely "competent," especially for emotion-intensive tasks.
• A stepped simulation continuum is recommended. Rather than wholesale replacement, educators should stage AI across the curriculum — AI for pre-learning, history-taking, and foundational reasoning early; high-fidelity human simulators and standardized patients for complex psychomotor skills and emotional intelligence later. Hybrid interaction modes (menu-based for novices, open-ended/voice GenAI for advanced learners) and human-facilitated debriefing alongside AI feedback are key implementation levers.
• Evidence base is limited. The review is dominated by uncontrolled designs, relies heavily on self-report, and has no longitudinal data on skill retention or clinical transfer; the near-uniform positive result pattern raises publication-bias concerns; and no studies came from Africa, South America, or low-income countries (with a Greater China concentration).

What this means for practice

• Medical educators. Deploy AI simulation for highly structured objectives — foundational communication, history-taking, health education — and keep standardized patients and high-fidelity manikins for complex psychomotor and emotionally loaded skills, since one included study found AI-assisted simulation inferior to standardized-patient simulation for breast self-examination skills.
• Instructional designers. Stage AI across the curriculum rather than replacing a lab session: AI for pre-learning and foundational reasoning early, human simulation later, with menu-based interaction for novices and open-ended or voice-based GenAI for advanced learners.
• Instructional designers. Pair automated AI feedback with human-facilitated debriefing, because learners rated AI as merely competent and preferred feedback sources they perceived as having benevolence and integrity, especially for emotion-intensive tasks.
• Medical educators. Tell students up front that the AI simulation is for foundational skills practice, not a substitute for the full clinical experience — the review's recommended way to manage expectations and reduce disappointment caused by the authenticity gap (robotic dialogue, missing nonverbal cues, speech-recognition delays).
• Medical educators. Check the primary evidence before committing resources: only 8 of 19 included studies (42%) had a parallel control group, so large-scale adoption decisions should wait on RCTs with active comparators.

Limitations

• The evidence base is dominated by uncontrolled designs: only 3 of the 19 studies were RCTs (2 rated low risk of bias, 1 with some concerns), and just 8 of 19 (42%) included a parallel control group.
• Outcome measurement relies heavily on self-report: self-efficacy, confidence, and satisfaction were the most common outcomes, and no included study examined skill retention beyond the immediate postintervention window or transfer to clinical practice.
• Publication bias cannot be ruled out: 8 of 11 studies (73%) reporting knowledge or skill outcomes and 8 of 9 (89%) reporting self-efficacy or attitude outcomes found significant improvements, and heterogeneity was too great for formal publication-bias testing.
• The 19 studies (N = 1,253) were conducted in high-income countries or regions with robust technological infrastructure, with none from Africa, South America, or low-income countries, and several quantitative studies had small samples that limited statistical power.

Citation

Jiang, H., Wang, Z., Shen, W., Meng, M., Yang, D., Li, X., & Hao, Y. (2026). AI-Powered Simulation for Nursing Education: Mixed Methods Systematic Review. Journal of Medical Internet Research, 28, e95167.