Knowledge, Attitudes, and Use of Generative AI (ChatGPT) Among Nurses in Iraqi Teaching Hospitals: A Multicenter Cross-Sectional Survey

Abstract

Background: The rapid uptake of generative artificial intelligence (GenAI) tools, particularly ChatGPT, has outpaced institutional guidance in many low- and middle-income health systems. Evidence on Iraqi nurses' knowledge, attitudes, and clinical use of these tools is limited. Aim: To assess Iraqi nurses' knowledge, attitudes, and self-reported clinical use of ChatGPT, and to identify factors associated with its use. Methods: A multicenter cross-sectional survey was conducted between January and March 2026 in three teaching hospitals in central Iraq, in line with the STROBE statement. The protocol was prospectively registered (ATU-RIR-2025-19, registered 12 December 2025). A 50-item self-administered questionnaire covering demographics, knowledge of GenAI (15 items, multiple-choice), attitudes (20 Likert items), and use practices (15 items) was developed and pilot-validated (Cronbach's α: knowledge 0.81, attitudes 0.84). A stratified sample of 412 registered nurses across medical–surgical, intensive care, and emergency wards was approached; 387 returned complete responses (response rate 93.9%). Descriptive statistics, t-tests/ANOVA for group differences, and a multivariable logistic regression model (with cluster-robust standard errors at the hospital level) were used to identify predictors of ChatGPT use. Results: Mean knowledge score was 64.2 ± 14.8 of 100, with the lowest scores on data privacy and patient confidentiality (38.4% correct). Attitudes were cautiously positive (mean 3.43 ± 0.71 on a 5-point scale): 64.1% agreed GenAI could enhance work efficiency, but 51.4% worried it might erode clinical judgment. In total, 38.0% (147/387) reported any clinical use of ChatGPT in the prior six months, most commonly for literature search (28.2%), drug information (22.0%), and patient-education materials (18.1%). The strongest concerns were information accuracy (78.0%), patient privacy (72.1%), and professional liability (67.7%). In the adjusted model, age <30 years (aOR 2.12, 95% CI 1.34–3.36), high English proficiency (aOR 2.84, 1.78–4.52), and prior digital health training (aOR 2.41, 1.55–3.74) were independent predictors of clinical use. Conclusion: Iraqi nurses are using ChatGPT in clinical contexts despite moderate knowledge and unresolved concerns about accuracy and privacy. Hospitals should issue institutional GenAI policies, integrate AI literacy into nursing curricula, and establish patient-data safeguards before deployment broadens further.