Artificial intelligence chatbots as sources for patient education material on child abuse

The World Health Organization defines c hildhood maltreatment as any form of abuse or neglect affecting children under 18 years of age that can cause actual or potential harm. Child abuse is a form of interpersonal trauma that can critically impact neurodevelopment and increase the risk of developing psychiatric disorders. With the increasing power and accessibility of artificial intelligence (AI) large language models, patients may turn to these platforms as sources of medical information. To date, no studies have evaluated the use of AI in creating patient education materials in childhood maltreatment and the field of psychiatry. Eight questions on child abuse from the National Child Traumatic Stress Network (NCTSN) were input into ChatGPT, Google Gemini, and Microsoft Copilot. A team of child psychiatrists and a pediatric psychologist reviewed and scored the responses by NCTSN and each AI, assessing quality, understandability, and actionability. Secondary outcomes included misinformation, readability, word count, and top references. The analysis of 32 responses showed good quality (mean DISCERN score 51.7) and moderate understandability (mean PEMAT 76.5 %). However, actionability was poor (mean PEMAT 64 %). Responses averaged a tenth-grade reading level, with ChatGPT being more difficult to read than NCTSN. AI-generated responses were significantly longer (p < 0.001). Findings of this study suggest that AI chatbots may currently be able to provide accurate, quality information on child abuse comparable to authoritative sources, albeit of significantly greater length. However, all sources lack actionability and exceed recommended reading levels, which limits effectiveness. These constraints suggest that AI chatbots should supplement rather than replace primary medical information sources. Urgent efforts are needed to improve the accessibility, readability, and actionability of patient education materials generated by AI and standardized sources on topics like child abuse and neglect. • AI chatbots produce quality child abuse content comparable to trusted sources. • Chatbot responses exceed recommended reading levels, limiting accessibility. • Findings show AI materials lack actionability, reducing practical utility. • AI tools should supplement, not replace, standard medical information sources. • Improvements needed in readability and actionability of AI-generated content.