What the research found
Vanderbilt University Medical Center researchers searched 578,058 progress notes for 215,712 patients in behavioural-health care between December 2022 and April 2026 and identified 28 patients (0.013%) with psychosis linked to AI chatbot use. 61% were in a first psychotic episode, against 29% of patients whose delusions involved AI without any chatbot use, and in 64% of cases the chatbot amplified symptoms already present; 24 of the 28 presented after GPT-4o's release in May 2024. The study is single-centre and observational, so it cannot show that chatbots caused psychosis, and it only captures chatbot use that clinicians wrote down.
Asking about chatbot use belongs in a mental-health assessment, above all for young people presenting with a first episode, because it rarely reaches the record unless someone asks.