This page was translated automatically from Turkish. Read the original

Bu sayfa, 26. Dünya Psikiyatri Kongresi'nde sunulan çalışmanın yapay zekâ desteğiyle hazırlanmış Türkçe özetidir. Çalışmanın sahibi sunan yazarlardır; atıf için orijinal kaynağa başvurun.

SO244

Expert evaluation of ChatGPT's psychoeducational responses about depression

EXPERT EVALUATION OF CHATGPT PSYCHOEDUCATION ANSWERS ABOUT DEPRESSION

Psychiatrists rated ChatGPT's responses to frequently asked questions about depression as good to very good, particularly in terms of clarity and accessibility, while scores for depth and relevance were lower.

The aim was to examine the potential role of large language models (LLMs), such as ChatGPT, in counseling and psychoeducation regarding the diagnosis and treatment of depression. Ten psychiatrists identified the questions most frequently asked by outpatients about depression, and an independent researcher selected five of these based on frequency (why am I depressed; are antidepressants addictive; will antidepressants change my personality; do antidepressants cause weight gain; will I return to my old self, or will I need to take antidepressants for life). Each question was submitted separately to ChatGPT in European Portuguese on January 15, 2026, without any introductory text, user preferences, or memory features; 31 psychiatrists independently rated each response on a 5-point Likert scale across six criteria—accuracy, clarity, relevance, depth, conciseness, and suitability for the target audience—yielding a total of 930 ratings. The analysis used means, standard deviations, Cronbach's alpha, and a two-way random-effects intraclass correlation coefficient (ICC) for absolute agreement. The overall mean across all criteria was 3.92 (SD 0.82); the highest scores were for clarity (4.07 ± 0.69), conciseness (4.04 ± 0.79), and suitability for the target audience (4.05 ± 0.76), while the lowest scores were for relevance (3.71 ± 0.82) and depth (3.78 ± 0.82); accuracy scored 3.86 ± 0.76. Internal consistency across raters was high (Cronbach's alpha = 0.95), whereas agreement was low at the individual-rater level (ICC = 0.05) and moderate when averaged across raters (ICC = 0.63). In conclusion, ChatGPT's responses show potential as a supplementary psychoeducational tool, particularly in terms of clarity, conciseness, and suitability; however, variability across raters and lower scores for depth and relevance warrant careful clinical oversight. The study's limitations include the evaluation of only five questions, responses generated and evaluated only in Portuguese, the absence of a comparison group, the examination of static responses rather than interactive dialogue, and the lack of measures of patient understanding and clinical outcomes.

Tüm WCP 2026 özetleri