Generative Artificial Intelligence Literacy Scale for Nurses: Development and Psychometric Evaluation

Generative artificial intelligence (GenAI) can automate time-intensive tasks and support clinical decision-making in care settings. Nurses require appropriate competencies to ensure that integration of GenAI strengthens care quality and patient safety. However, validated literacy assessment tools remain limited. In particular, instruments tailored to nurses' role-specific GenAI competencies, including hallucination detection, risk identification, and ethical accountability, are lacking. These gaps highlight the need for a nurse-specific GenAI literacy scale.

This study aimed to develop and psychometrically validate the Generative Artificial Intelligence Literacy Scale for Nurses (GenAILS).

We conducted a two-phase, cross-sectional online survey of registered nurses nationwide in Taiwan between June 2025 and October 2025. Phase 1 involved conceptualization and item generation based on a literature review, followed by content appraisal through expert discussion with 6 external reviewers. A 50-item pool was generated. Subsequently, 5 external reviewers evaluated content validity. Items with a content validity index of <0.78 or flagged for revision were revised or deleted. Phase 2 evaluated psychometric properties (item analysis, internal consistency, split-half reliability, and criterion-related validity) and construct validity via exploratory factor analysis (factor loading ≥0.60), followed by confirmatory factor analysis (CFA). The total sample was randomly split into 2 independent subsamples for exploratory factor analysis and CFA.

In phase 1, the initial 50 items underwent expert content validation and were revised to 46 items (scale content validity index based on the average method=0.92). In phase 2, 1313 questionnaires were collected, of which 191 invalid responses were excluded; 1122 valid responses were analyzed. Participants had a mean age of 34.66 (SD 7.8) years. Extreme-group comparison revealed statistically significant differences for each item (P<.001). The final scale comprised 24 items across six dimensions: responsible use, updated competencies, risk identification, fundamental knowledge, critical evaluation, and ethics and law. The cumulative variance explained was 53.1%. The first-order CFA demonstrated excellent model fit: root-mean-square error of approximation=0.035, standardized root-mean-square residual=0.032, comparative fit index=0.99, goodness-of-fit index=0.94, adjusted goodness-of-fit index=0.93, nonnormed fit index=0.99, and parsimony normed fit index=0.84. The second-order CFA demonstrated excellent model fit: root-mean-square error of approximation=0.039, standardized root-mean-square residual=0.040, comparative fit index=0.99, goodness-of-fit index=0.94, adjusted goodness-of-fit index=0.92, nonnormed fit index=0.99, and parsimony normed fit index=0.87. All heterotrait-monotrait ratio values were below 0.85, supporting discriminant validity. The scale was moderately correlated with the Short Form Meta-AI Literacy Scale (r=0.57; P<.001). Reliability was excellent (Cronbach α=0.92; McDonald ω=0.92; split-half reliability=0.81).

The GenAILS is a concise, nurse-specific self-report instrument with good psychometric properties across 6 clinically relevant domains. It supports needs assessment, targeted training, and intervention evaluation to promote the safe and ethical use of GenAI in nursing.