Key points are not available for this paper at this time.
Objectives Generative artificial intelligence tools such as ChatGPT and DeepSeek are increasingly used by parents seeking information about autism spectrum disorder. However, their effectiveness in addressing sensitive, multidimensional needs especially in early childhood neurodevelopment remains underexplored. This study aimed to compare the accuracy, content adequacy, and reproducibility of ChatGPT and DeepSeek in responding to frequently asked questions by parents of children with autism.Methods A total of 41 validated parent-centered questions were presented to ChatGPT and DeepSeek in two separate sessions, resulting in 164 responses. Each answer was evaluated by three blinded experts using a 4-point scoring system.Results ChatGPT provided comprehensive and correct responses to 82.9% of questions, while DeepSeek achieved 73.2%. ChatGPT performed better in family and psychosocial categories, whereas DeepSeek showed stronger performance in diagnosis and treatment. Reproducibility rates were 78.0% for ChatGPT and 73.2% for DeepSeek. Inter-rater agreement was statistically significant for both models (κ = 0.611 for ChatGPT; κ = 0.820 for DeepSeek), with highest agreement in clinical categories.Conclusions ChatGPT and DeepSeek show strong potential as supplementary tools for autism-related guidance. Although neither model fully addresses all informational needs alone, their complementary strengths highlight the value of a multimodal approach supported by ongoing evaluation and ethical oversight.
AVCI et al. (Sat,) studied this question.