Large Language Modelle (Chat-GPT, Gemini et al.) als "Psycho-Coach" für Menschen mit depressiven Erkrankungen (Umfrage "Large Language Modells und Depression")

Key Findings

• — 65% of all respondents (N=2,500) had used an LLM chatbot to talk about their own mental health problems as with a trusted friend or therapist (76% of 16 to 19-year-olds); among respondents with a medical diagnosis (N=537), 69% had done so, 56% in the past 30 days and 35% about their depression, and 10% in longer, dialogue-style conversations.
• — Among diagnosed respondents who had discussed depression with an LLM in the past 30 days (N=186), 85% described the conversations as helpful, and 65% of those who had received psychotherapy (N=143) rated LLM conversations as as good as or better than psychotherapeutic conversations.
• — In the same group (N=186), 53% agreed or rather agreed that they had experienced increased thoughts of self-harm or suicide after use and 62% agreed or rather agreed that the conversations had made seeing a doctor or psychotherapist unnecessary; the foundation recommends against LLM use for diagnosis or as sole treatment.

Methodology Notes

Online survey of German private households, 5 to 24 March 2026, fielded by bilendi (ISO 26362 certified); N=2,500 aged 16 to 39 (47.6% women, 52.0% men, 0.5% diverse). Subgroup bases are stated per chart (N=537, N=186, N=143, N=58 for open-text answers). Self-report; the self-harm item is an agreement statement without a causal control or a baseline for comparison. Publication date is the 28 April 2026 date printed on the press release. The foundation's own host returned HTTP 403 to direct fetches, so both PDFs (19-page chart book, 2-page press release) were read through a text relay of the same files; figures were checked against both documents. German-language source.

Sources