Quality of Autism Information Generated by Artificial Intelligence Tools: Implications for Paediatric Care

Families increasingly turn to artificial intelligence (AI) tools for information about autism spectrum disorder (ASD), often during early childhood when concerns about development, diagnosis and intervention first emerge. From a paediatric perspective, these tools increasingly function as informal sources of health information alongside primary care and developmental services. This study aimed to evaluate the quality of autism-related information generated by widely used AI platforms.

Using a descriptive research design, responses generated by six freely accessible AI platforms (ChatGPT, Gemini, Microsoft Copilot, Perplexity, Brave and Grok) were examined. Each platform was asked 15 autism questions with well-established scientific answers. Responses were evaluated across five dimensions relevant to paediatric health communication: accuracy, readability, language framing, actionability and reference quality.

Substantial variability was observed across AI platforms. Several tools produced generally accurate explanations of ASD; however, readability levels consistently exceeded recommended guidelines for paediatric health materials (i.e., the 6th-8th grade reading level). Most responses relied primarily on medicalized language rather than neurodiversity-affirming framing. Actionable guidance was limited, with only a minority of responses offering concrete next steps for families navigating early paediatric decision-making. Reference practices varied widely, with some platforms providing numerous credible sources and others offering few or none.

Although AI tools can support parental understanding of ASD, differences in clarity, tone, usability and transparency may shape families' expectations prior to or during paediatric consultations. These findings highlight the need for thoughtful use of AI-generated autism information and suggest that families may benefit from guidance from paediatric professionals when interpreting AI-based responses.