Consistency and WHO Guideline Alignment of ChatGPT-Generated Antenatal Care Information: A Repeated-Session Comparative Study

Background/Objectives: This study evaluated the consistency and WHO guideline alignment of antenatal care information generated using ChatGPT (OpenAI) under two access configurations: free-tier (GPTa) and Plus (GPTb), during the transition from GPT-4o to GPT-5 (July–August 2025), including changes following verification prompting. The exact underlying model versions were not recorded. Methods: Fifty-one questions based on the 2016 WHO antenatal care recommendations were evaluated across four initial sessions and a separate verification condition (Session 5). Responses were scored as non-compliant (0), partially compliant (1), or fully compliant (2). Friedman and Holm-adjusted Wilcoxon tests assessed differences across Sessions 1–4, while paired Wilcoxon and McNemar tests evaluated verification-related changes. Results: Full-compliance rates across Sessions 1–4 were 74.02% for GPTa and 72.06% for GPTb. Friedman tests showed no significant overall differences across these sessions for GPTa (p = 0.056) or GPTb (p = 0.758). No significant between-configuration differences remained after Holm correction. Following verification prompting, GPTb demonstrated significantly improved ordinal alignment scores (adjusted p = 0.042) and full-compliance rates (adjusted p = 0.032), whereas GPTa showed no significant change. Conclusions: Both configurations frequently aligned with the selected WHO recommendations, without significant overall differences across the initial sessions or between configurations. Verification prompting improved WHO guideline alignment for GPTb, but this does not establish genuine clinical error correction or improved patient safety. AI-generated antenatal care information requires independent verification against authoritative clinical guidance.