Study: AI chatbots can reinforce five mental-health risk patterns
Key points
- Journal of Psychopathology and Clinical Science paper maps five chatbot risk pathways.
- About 70% of AI chatbot conversations involve personal/emotional topics, not only work.
- Nearly half of adults with diagnosed conditions have used bots for emotional support.
- Authors say ChatGPT/Gemini/Copilot-class tools lack clinical crisis protocols.
Seven in ten chatbot sessions, researchers estimate, already drift into feelings, relationships and mental distress — not spreadsheets. Eastleigh Voice reported findings from the Journal of Psychopathology and Clinical Science that also put nearly 50% of adults with a formal mental-health diagnosis among people who have leaned on bots for emotional support, often without telling their clinician.
The paper’s warning list runs to five loops: putting off human care because a bot is always “there”; feeding compulsive reassurance (a known trap in anxiety and OCD); replacing friends with a tireless virtual listener; nodding along with distorted beliefs instead of challenging them; and training users to outsource ordinary decisions until self-trust shrinks.
Therapy-shaped products vs chat toys
Purpose-built clinical tools tested under supervision can teach skills and signpost help. Mass-market systems branded like ChatGPT, Gemini or Copilot were trained for general conversation. They typically do not carry the crisis playbooks, mandatory reporting norms or liability chains expected of a psychologist. Comforting tone is not a safety case.
Authors stress the evidence base is still thin — observational work and early reports more than huge long trials. That argues for labelling and guardrails, not for pretending zero risk.
Policy window in Kenya
As Nairobi debates AI rules, mental-health use should be an explicit chapter: local-language crisis links, bans on “therapy” marketing without clinical oversight, and guidance for schools and clinics. Understaffed counselling services will inherit worse caseloads if bots become the only nightly confidant. If self-harm thoughts appear, contact human emergency and counselling services immediately — not another prompt.
Universities training counsellors should add chatbot literacy to practicum modules: how to ask patients what apps they use nightly, and how to document bot-reinforced delusions without shaming help-seeking. App stores operating in Kenya can pre-empt regulation by shipping crisis buttons that dial local hotlines, not only global web forms.
Based on Eastleigh Voice summary of research in the Journal of Psychopathology and Clinical Science. Not medical advice; clinical decisions require qualified professionals.
