A scoping review on the mental health harms of LLM-based chatbots

Key Findings

• — 3,137 articles identified across ACM, IEEE, PubMed, Science.gov and Google Scholar plus a supplementary search; 119 met inclusion criteria
• — Conceptual works attribute harm to chatbot limitations named as hallucinations, sycophancy and bias, to data-security issues, and to risks in severe or high-risk psychiatric cases such as suicide and psychosis
• — Vignette studies show LLM-based chatbots respond inappropriately to mental health queries when measured against clinical standards
• — Cognitive overreliance on chatbots is associated with decreased cognitive and academic performance
• — Problematic use, marked by symptoms of emotional or social dependency and withdrawal, correlates with mental health symptoms
• — Articles on AI psychosis propose links between delusional beliefs and chatbot use, including chatbots reinforcing and validating delusional beliefs

Methodology Notes

Scoping review rather than a meta-analysis: no effect sizes are pooled and the included literature is heterogeneous in design, with conceptual pieces counted alongside empirical studies, so the category sizes describe where research attention has gone and not incidence of harm. PRISMA-based search with a validated search string across five databases plus a supplementary search. Open access under CC BY 4.0. Date established from the article page metadata (citation_online_date 2026/08/20, prism.publicationDate 2026-08-20), which agrees with the PubMed article date.

Sources