Psychologists warn of harmful mental-health advice from AI chatbots

Nearly nine in ten psychologists, counsellors and psychotherapists surveyed by the British Psychological Society say patients are relying too heavily on AI chatbots, and the BPS warns that "unqualified and harmful" mental-health advice from the tools is a real risk. The poll of 197 practitioners found a 49 percent rise over the past year in the number of patients seeking support from chatbots.

Seven in ten respondents said they were somewhat or very concerned about AI use for mental-health advice, around 76 percent said the technology is negatively affecting their clients' wellbeing, and some 80 percent reported concern about over-reliance. "The risk of unregulated open access AI chatbots providing unqualified and harmful advice remains very real," said BPS president Dr Laura Dean, adding that AI "cannot replace the vital human element in mental health care" and can only be brought into healthcare with safeguards that protect both practitioners and the people they support.

The respondents described the mechanism from their consulting rooms. One said people turn to AI because it feels "safer" — it removes the relational risk of building trust with another human being — which increases isolation and weakens social contact, "directly linked to quality of life and wellbeing." Another said they did not understand "exactly what AI does, how it works, and how it comes up with its answers." A third called for accountability, noting that the companies behind the tools, like many social media platforms, "have free rein to do as they please with little to no legal responsibilities."

The demand is already established elsewhere. A survey of 2,000 children aged 11 to 17 by the charity Barnardo's and the National Counselling and Psychotherapy Society found 43 percent had used AI for advice or emotional support. Rosie Weatherley of the charity Mind said the findings echo what Mind hears and sees in emerging evidence: even casual use of generative AI can cause harm, and much is unknown about long-term effects — though people still say they value human support over AI tools.

Two caveats sit inside the BPS poll: it records what practitioners observe, not measured clinical outcomes, and its respondents are the in-person providers whose clients are also consulting the chatbots. What it does establish is scale — schoolchildren and adult patients are taking their problems, in rising numbers, to products whose makers hold no clinical licence.

Neither survey asked the children whether the advice helped.