EUVIMEDEUVIMED entdecken
ÜbersichtsarbeitOpen Access · CC BY 4.0

Generative AI chatbots and mental health: A critical narrative review of benefits, risks, and clinical implications

Bruno da Silva Santos iD, Arthur Dolberth iD, Giovanni Gonçalves De Toni iD, Carlos Eduardo Freitas Dantas iD, Matheus de Lima Ruffini iD, Flavio Kapczinski iD, Jaime Eduardo Cecilio Hallak iD, Ives Cavalcante Passos iD, Thiago Henrique Roza iD

Psychiatry and Clinical Neurosciences · 09.11.2026

DOI: 10.1111/pcn.70159 ↗

Kurz gefasst

Übersichtsarbeit in der Zeitschrift Psychiatry and Clinical Neurosciences, online erschienen am 09.11.2026. Verfasst von Bruno da Silva Santos und 8 weiteren Autor:innen. Beteiligte Einrichtung: Laboratory of Molecular Psychiatry, Experimental Research Centre (CPE) and Clinical Research Centre (CPC), Hospital de Clínicas de Porto Alegre (HCPA…. Die Arbeit nennt 64 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Conselho Nacional de Desenvolvimento Científico e Tecnológico, Ministério da Saúde, Instituto Nacional de Ciência e Tecnologia em Saúde Mental Digital.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The rapid diffusion of generative artificial intelligence chatbots has repositioned these systems from productivity tools to interlocutors used for emotional support, companionship, and mental health guidance. This critical narrative review has three aims: to give mental health professionals a concise technical account of large language models; to separate the evidence and risks attached to four distinct system classes, protocolized therapeutic chatbots, clinically bounded generative systems, general‐purpose large language models used as informal therapy, and companion platforms; and to examine clinical, ethical, and regulatory implications. Literature was identified through non‐systematic, iterative database and preprint‐server searches between March and July 2026, supplemented by citation tracking and policy documents. No prespecified eligibility or screening protocol was used; the comparative claims are therefore provisional. Protocolized and clinically bounded systems carry the strongest evidence, including a meta‐analysis of generative and hybrid chatbots reporting modest short‐term symptom reduction and a randomized trial of an expert‐fine‐tuned therapeutic chatbot; these trials are brief, predominantly waitlist‐controlled, and reliant on self‐report. General‐purpose models and companion platforms may deliver immediate validation and temporary reductions in loneliness, but that evidence is largely observational, qualitative, or short‐term experimental and does not establish sustained benefit. Concerns include emotional dependence, social substitution, sycophantic reinforcement, unsafe responses to suicidal or psychotic content, commercial manipulation, and delusional beliefs temporally associated with sustained use. These systems are not a single intervention class. Their effects appear contingent on clinical structure, relational design, user vulnerability, duration and centrality of use, commercial incentives, and safeguards.

Schlagwörter

Stichwörter der Autor:innen

artificial intelligencelarge language modelsmental healthnatural language processingreview

Autor:innen und Institutionen

  • Laboratory of Molecular Psychiatry, Experimental Research Centre (CPE) and Clinical Research Centre (CPC), Hospital de Clínicas de Porto Alegre (HCPA), Porto Alegre, Rio Grande do Sul, Brazil.
  • Federal University of Rio Grande do Sul, School of Medicine, Graduate Program in Psychiatry and Behavioural Sciences, Department of Psychiatry, Porto Alegre, Rio Grande do Sul, Brazil.
  • National Institute of Science and Technology in Digital Mental Health (INCT-SMD), Porto Alegre, Rio Grande do Sul, Brazil.
  • Laboratory of Research, Prevention, and Intervention in Forensic Psychology, Paraná Tuiuti University, Curitiba, Paraná, Brazil.
  • School of Health Sciences of the Federal District (ESCS), Brasília, Distrito Federal, Brazil.
  • Digital Psychiatry Laboratory, Federal University of Paraná, Curitiba, Paraná, Brazil.

ORCID-Profile: Bruno da Silva Santos, Arthur Dolberth, Giovanni Gonçalves De Toni, Carlos Eduardo Freitas Dantas, Matheus de Lima Ruffini, Flavio Kapczinski, Jaime Eduardo Cecilio Hallak, Ives Cavalcante Passos, Thiago Henrique Roza

Förderung

  • Conselho Nacional de Desenvolvimento Científico e Tecnológico
  • Ministério da Saúde
  • Instituto Nacional de Ciência e Tecnologia em Saúde Mental Digital
  • Coordenação de Aperfeiçoamento de Pessoal de Nível Superior

Zitieren

APA

Santos, B. D. S., Dolberth, A., De Toni, G. G., Dantas, C. E. F., de Lima Ruffini, M., Kapczinski, F., Hallak, J. E. C., Passos, I. C., & Roza, T. H. (2026). Generative AI chatbots and mental health: A critical narrative review of benefits, risks, and clinical implications. Psychiatry and Clinical Neurosciences. https://doi.org/10.1111/pcn.70159

Vancouver

Santos BDS, Dolberth A, De Toni GG, Dantas CEF, de Lima Ruffini M, Kapczinski F, et al. Generative AI chatbots and mental health: A critical narrative review of benefits, risks, and clinical implications. Psychiatry and Clinical Neurosciences. 2026. doi:10.1111/pcn.70159

BibTeX
@article{Santos2026,
  author = {Santos, Bruno da Silva and Dolberth, Arthur and De Toni, Giovanni Gonçalves and Dantas, Carlos Eduardo Freitas and de Lima Ruffini, Matheus and Kapczinski, Flavio and Hallak, Jaime Eduardo Cecilio and Passos, Ives Cavalcante and Roza, Thiago Henrique},
  title = {Generative AI chatbots and mental health: A critical narrative review of benefits, risks, and clinical implications},
  journal = {Psychiatry and Clinical Neurosciences},
  year = {2026},
  doi = {10.1111/pcn.70159}
}
Kostenlos starten →