Assessing AI-Generated Autism Information for Healthcare Use: A Cross-Linguistic and Cross-Geographic Evaluation of ChatGPT, Gemini, and Copilot

Editorial reading guide · Trust

Is an AI answer equally reliable in the language our family actually uses?

Accuracy, references, actionability, and readability did not point to one uniformly best tool. Turkish and English outputs differed, and medicalized framing remained common, showing that trust depends on language and the kind of help sought.

Study context

Researchers evaluated answers to 44 caregiver-focused autism questions in English and Turkish across the USA, England, and Türkiye.

Age / stage: Autism information for families of children roughly 2–18 years; no family users enrolled

Read with three questions

1. How did accuracy trade off against references or actionable support?
2. Which language differences matter for your family’s questions?
3. How might medicalized framing shape a parent’s view of an autistic child?

Keep the limits in view

Only three systems, two languages, and single-turn prompts were tested in May 2025. Content scores cannot show whether families understood or acted on the advice, and should not be generalized to all cultures or later model versions.