Assessment of Large Language Models for Tracheostomy Care in Pediatrics: ChatGPT vs Claude.

INTRODUCTION: This study aimed to evaluate the accuracy, completeness, clarity, and relevance of responses provided by two freely available large language models (LLMs)-ChatGPT-4.o and Claude 4.0 Sonnet-to caregiver-focused questions about pediatric tracheostomy. METHODS: Seventeen clinically relevant questions were submitted to each LLM, and responses were evaluated by nine pediatric otolaryngologists across four domains using a 5-point Likert scale. RESULTS: ChatGPT-4.o showed significantly higher accuracy (Mean = 4.49, p = .029), while Claude 4.0 Sonnet achieved better completeness scores (Mean = 4.25, p = .022). No significant difference was found in clarity or clinical relevance. ChatGPT-4.o responses were more readable (FRES = 63.78, FKGL = 6.41) compared to Claude (FRES = 36.93, FKGL = 10.44). CONCLUSION: Preliminary findings suggest that both LLMs may offer supplementary educational value for tracheostomy caregivers, but their role in urgent situations was not evaluated. Use should remain under clinical oversight, and further research is needed.