Generative AI is entering nursing classrooms, skills laboratories, and clinical simulation activities—but does it help students become safer, more capable nurses, or mainly help them complete learning tasks? A systematic review and meta-analysis published in Nurse Educator on October 23, 2026, offers a timely answer. It found positive average effects for knowledge, practice-related scores, and teaching satisfaction across 17 randomized studies involving 1,456 participants. Yet the review did not find statistically significant improvements in every outcome, including competency and self-efficacy. The authors also warn that substantial variation between studies makes the size and reproducibility of the benefits uncertain.
For nursing faculty and training leaders, the practical conclusion is not “replace teaching with a chatbot.” It is to use generative AI as a supervised learning aid, then test whether students can explain, perform, and retain what they have practised without the tool.
Featured training photograph: Navy Medicine on Unsplash.
What the 2026 nursing education meta-analysis examined
Wan and colleagues searched PubMed, Embase, Web of Science, and the Cochrane Library for studies published up to July 2026. Their review, published ahead of print on October 23, 2026, pooled 17 randomized controlled studies with 1,456 participants. The researchers used random-effects models and examined heterogeneity, sensitivity to individual studies, meta-regression, and possible small-study effects.
The outcomes were not interchangeable. Knowledge scores measured what students knew; practice-related scores captured performance on study-specific practical assessments; satisfaction measured how learners rated the teaching experience. Competency, attitudes, and self-efficacy were separate outcomes. Keeping those distinctions matters because a student can enjoy an AI-supported activity or score well on a short exercise without necessarily demonstrating reliable clinical performance in a real care environment.
Read the original review in Nurse Educator (Wan et al., October 23, 2026).
What improved—and what did not
The pooled results reported by the authors were:
| Outcome | Standardized mean difference | 95% confidence interval | Interpretation |
|---|---|---|---|
| Knowledge scores | 0.53 | 0.03 to 1.04 | Positive pooled difference |
| Practice-related scores | 1.25 | 0.23 to 2.28 | Positive pooled difference, with a wide interval |
| Teaching satisfaction | 0.58 | 0.02 to 1.13 | Positive pooled difference |
| Attitudes, competency, self-efficacy | Not significant | — | No statistically significant difference reported |
These are standardized mean differences, not percentage increases. An SMD of 0.53 does not mean that students learned 53% more, and an SMD of 1.25 does not mean that clinical skill rose by 125%. Standardized effects express differences relative to variation in the studies’ measurement scales. They help combine different instruments, but they do not translate directly into a pass rate, a patient-safety improvement, or a clinical competency level.
The wide confidence intervals are another reason to avoid headline claims. The pooled estimates suggest potential, but the plausible range around each estimate is broad. Most importantly, the review did not show a statistically significant pooled improvement in competency or self-efficacy. Educators should therefore resist turning a positive knowledge or practice score into a claim that AI has already made students clinically competent.
Why the results need cautious interpretation
The authors explicitly report extreme heterogeneity and sensitivity to individual studies. In plain language, the studies did not all test the same educational intervention under the same conditions. They may have differed in the AI tools used, the course content, the duration and structure of the intervention, the comparison teaching method, and the instruments used to assess outcomes. A pooled average can summarise this varied evidence, but it cannot guarantee that the same effect will appear in every nursing programme.
The review’s meta-regression findings were exploratory. They should be treated as hypotheses for further research, not as proof that one tool, teaching format, or duration is universally superior. The search also ended in July 2026, so the review does not represent every study published later in the year.
There is a further distinction between a controlled educational score and transfer to clinical practice. A simulated case or classroom quiz can assess selected knowledge and reasoning. It cannot, by itself, establish that a learner will recognise deterioration, communicate under pressure, follow infection-control procedures, or make safe decisions with a real patient. Those outcomes require appropriate supervised assessment and, where relevant, longitudinal follow-up.
What a second review says about AI-powered nursing simulation
A separate mixed-methods systematic review in the Journal of Medical Internet Research, published July 21, 2026, examined AI-powered simulation in nursing education. It included 19 studies and 1,253 participants, but the evidence base was methodologically mixed: only three studies were randomized controlled trials, alongside controlled quasi-experimental, uncontrolled pre–post, qualitative, and other designs.
That review found promising signals for cognitive knowledge and some affective outcomes, such as confidence and communication comfort. However, results for complex psychomotor skills were inconsistent. One randomized study found AI-assisted simulation inferior to a standardized-patient simulation for a particular outcome. Learners also described an “authenticity gap”: AI interactions could feel robotic, lacked some non-verbal cues, or were affected by technical instability.
The two reviews should not be compared as if they were identical experiments. They used different inclusion criteria and covered different forms of AI-supported teaching. Taken together, however, they point to a consistent design principle: AI may help learners rehearse and reason, but it should not displace human feedback, hands-on practice, or supervised clinical experience.
Read the JMIR review on AI-powered simulation for nursing education.
Five responsible uses of generative AI in nursing education
1. Generate varied patient scenarios for guided discussion
Faculty can use an approved AI tool to draft alternative case scenarios—for example, changing a patient’s age, presenting symptoms, comorbidities, or social circumstances while preserving the learning objective. An instructor must verify the clinical facts, remove unsafe or implausible details, and ensure that the scenario does not reproduce identifiable patient information. The purpose is to give students more opportunities to compare cases, not to let an unverified model write the clinical answer key.
2. Ask students to explain the reasoning behind a decision
Instead of asking a chatbot for a diagnosis or care plan, students can be prompted to identify relevant cues, explain what information is missing, prioritise risks, and justify their next action. The instructor can then compare the student’s reasoning with an evidence-based rubric. This makes the AI interaction a prompt for explanation rather than a shortcut to a finished response.
3. Use branching conversations for low-stakes communication practice
A conversational patient or family-member simulation can let learners rehearse history-taking, plain-language explanations, and empathetic communication. Faculty should define the scenario boundaries, make clear that the virtual character is not a real patient, and debrief the interaction. AI-generated dialogue should not be treated as a reliable measure of empathy or professional communication without human review.
4. Pair digital rehearsal with physical skills practice
AI can help students prepare for a procedure by asking them to sequence steps, identify safety checks, or explain why a step matters. The practical skill itself still needs to be demonstrated and assessed using appropriate equipment, simulation, or supervised clinical practice. A chatbot cannot verify hand placement, aseptic technique, physical dexterity, or the quality of a real patient interaction merely because a learner describes those actions convincingly.
5. Turn feedback into a structured debrief
After an exercise, students can use AI to organise their reflections into “what I noticed,” “what I decided,” “what I would change,” and “what evidence supports the change.” The instructor should correct inaccuracies and guide the debrief. Reflection is useful when it helps learners examine their decisions; it is not useful when a model supplies polished reflections that students do not understand or own.
A practical evaluation framework for nursing faculty
Before adopting a generative AI tool across a programme, define the learning outcome and decide how it will be measured. A short pilot can use the following checks:
- Knowledge: use a valid pre-test and post-test, and include a delayed check to see whether learning is retained.
- Practical performance: use an instructor-observed checklist or a validated assessment instrument. Do not rely only on a learner’s self-report or a chatbot’s rating.
- Clinical reasoning: ask students to identify cues, explain priorities, justify decisions, and recognise when they need to escalate to a supervisor.
- Transfer: examine whether the skill appears in a new scenario or supervised placement, not only in the exact case used during AI practice.
- Equity and access: check language quality, disability access, device availability, and whether the tool performs differently across learner groups.
- Safety and governance: review privacy, data retention, vendor terms, output logging, human oversight, and the process for reporting errors.
- Workload: measure the time required for faculty verification, scenario preparation, technical support, and debriefing. A tool that saves student time but creates unsustainable review work may not be an improvement.
For simulation programmes, the International Nursing Association for Clinical Simulation and Learning (INACSL) provides Healthcare Simulation Standards of Best Practice covering simulation design, prebriefing, facilitation, evaluation, and debriefing. These standards are useful when deciding where an AI component fits into a broader simulation design; they are not an endorsement of any particular AI product. See INACSL’s simulation standards overview.
Governance: protect patients, students, and professional judgement
Generative AI tools should not receive identifiable patient records, confidential placement notes, or sensitive student information unless the institution has explicitly approved the system and its data-protection arrangements. Faculty should verify outputs against current clinical guidance and course materials, disclose when AI has been used in learning activities, and provide a non-AI route where access or accommodation requires one.
The World Health Organization’s 2025 guidance on the ethics and governance of large multi-modal models in health emphasises the need to assess risks, oversight, accountability, and the suitability of systems for their intended health use. Although it is not a nursing-education curriculum, it offers a relevant governance lens for institutions using generative AI in health-related settings. Read the WHO guidance.
The American Nurses Association’s OJIN discussion, “An Ethics of Artificial Intelligence for Nursing,” also argues against treating AI as either an automatic solution or an inevitable threat. It calls attention to care relationships, equity, upstream harms, and the way systems are designed and implemented. Read the nursing ethics discussion.
For a broader approach to evaluating educational AI, see Prowad’s guides on evidence checks for AI learning tools and measuring AI literacy beyond prompting.
What nursing schools should conclude in 2026
The October 2026 meta-analysis provides encouraging evidence that generative AI-supported teaching can improve some measured learning outcomes in nursing education. It does not show that every outcome improves, that the effects are consistent across settings, or that AI can replace clinical instructors and hands-on simulation. The absence of a significant pooled effect for competency and self-efficacy is especially important for leaders responsible for professional preparation.
The strongest next step is a carefully governed pilot with a clearly defined learning objective, instructor-verified content, observed practical assessment, and delayed measurement of retention and transfer. Use AI to expand practice opportunities and stimulate explanation. Keep human educators responsible for clinical judgement, feedback, and decisions about readiness to practise.
Evidence note: This article interprets published educational research. It is not clinical guidance and does not replace local nursing education standards, clinical supervision, or patient-safety procedures.
References
- Wan, Q., Long, X., Lin, T., Zhang, Z., & Meng, Y. (2026). “Applications of Generative Artificial Intelligence Teaching in Nursing Education: A Systematic Review and Meta-Analysis of Randomized Controlled Trials.” Nurse Educator, published ahead of print October 23, 2026. https://doi.org/10.1097/NNE.0000000000002355
- Jiang, H., et al. (2026). “AI-Powered Simulation for Nursing Education: Mixed Methods Systematic Review.” Journal of Medical Internet Research, 28, e95167. https://www.jmir.org/2026/1/e95167/
- International Nursing Association for Clinical Simulation and Learning. Healthcare Simulation Standards of Best Practice. INACSL standards overview.
- World Health Organization. (2025). Ethics and governance of artificial intelligence for health: Guidance on large multi-modal models. WHO publication.
- Dillard-Wright, J., & Smith, J. (2025). “An Ethics of Artificial Intelligence for Nursing.” OJIN: The Online Journal of Issues in Nursing. Read article.