Artificial Intelligence Anxiety Among Nurses: A Systematic Review and Meta-Analysis

Background/Objectives: Nurses are central to artificial intelligence (AI) implementation in healthcare, yet estimates of AI-related anxiety are difficult to compare because Artificial Intelligence Anxiety Scale (AIAS) versions have different total-score ranges. This review estimated AI anxiety using separate syntheses of range-compatible raw scores, with percent-of-maximum-possible (POMP) normalization as a secondary exploratory analysis. Methods: Following PRISMA 2020, PubMed, Scopus, CINAHL, and Google Scholar were searched from inception to 16 July 2026. Methodological quality was appraised using Joanna Briggs Institute checklists. Random-effects models used restricted maximum likelihood estimation with Hartung–Knapp inference; DerSimonian–Laird models were retained for sensitivity analysis. Prediction intervals, leave-one-out analyses, and influence diagnostics were examined. The protocol was retrospectively registered on OSF (DOI: 10.17605/OSF.IO/K3B56). Results: Twenty-five studies included 7271 nurses. The pooled mean was 45.85 (95% CI 44.45–47.26; 95% PI 41.75–49.96; I2 = 80.0%) for the 16–80 AIAS (10 studies; N = 2545) and 69.28 (95% CI 62.03–76.53; 95% PI 50.58–87.97; I2 = 94.4%) for the 21–147 AIAS (six studies; N = 2066). Pooled means were essentially unchanged under the alternative estimator. Extreme heterogeneity limited interpretation of the exploratory POMP synthesis. Conclusions: Mean AIAS scores were below theoretical midpoints. Raw AIAS totals from incompatible versions should not be pooled. Wide prediction intervals indicate substantial between-study variation, and the absence of validated severity cut-offs precludes clinical severity classification. Longitudinal, controlled, and psychometrically rigorous research is needed before specific anxiety-reduction interventions can be recommended.