Large language models can help improve questionnaires used to diagnose mental illness by optimizing symptom generalizability and reducing redundancy.