A new international survey spanning 30 countries reveals that more than half of psychotherapists already use tools such as ChatGPT in clinical practice to support treatment planning. Researchers say adoption is outpacing regulation, training, and governance, creating both opportunities and risks.
According to a study, Generative Artificial Intelligence in Psychotherapy Practice: A Global Online Survey of Mental Health Professionals’ Adoption (1), involving 766 mental health professionals from 30 countries, 54.6% already use generative AI tools in their psychotherapy practice. ChatGPT dominates the market, with nearly 85% of AI users relying on OpenAI's chatbot, far ahead of Google Gemini or Microsoft Copilot.
The findings are consistent with previous surveys conducted by the same research team (2), which showed that 52% of UK general practitioners use generative AI tools in clinical practice, while surveys of patients with rare diseases indicate that many believe AI could significantly shorten the often decades-long journey to receiving an accurate diagnosis.
Treatment planning comes before documentation
Treatment planning emerged as the most common clinical use, reported by 41.9% of AI users. Administrative tasks followed closely (41.4%), while almost 40% used AI to generate psychoeducational materials for patients. More than 30% also reported using AI to support reflective practice or clinical supervision.
These findings challenge the prevailing assumption that clinicians mainly employ AI as a digital secretary.
Charlotte Blease, Ph.D., Associate Professor at Uppsala University and Research Affiliate at Harvard Medical School, believes the scale of adoption has taken the research community by surprise.
"What surprised me most was the scale: this is the first time we've had hard, global data – drawn from 30 countries – on generative AI use among mental health professionals, and the numbers are stark. Over half are already using these tools in practice; nobody had mapped that before now. This is comparable to adoption rates among doctors, too. In the UK, we've found that more than half of GPs are using these tools in clinical practice. Worse still, ChatGPT is the frontrunner in both studies. In the survey of mental health clinicians, of those who used AI, 85% said they adopted ChatGPT – a consumer tool, not built for mental health support."
She points out that clinicians are relying on AI for tasks that directly influence therapeutic decisions.
"What's striking within that is what mental health clinicians are using them for. The assumption has always been that documentation – letters, notes, and paperwork – is something nobody wants to do. But that's not the top box for mental health clinicians such as therapists and clinical psychologists. Treatment planning comes first. Yet that's clinical judgment – the core of the therapeutic relationship – not clerical overflow."
Adoption lacks governance. It’s dangerous
The survey also exposes a striking gap between adoption and preparedness.
Although AI has entered routine practice, institutional support remains scarce. Only 18% of respondents said their employer or professional organization had encouraged the use of AI, while just 18.3% had received any formal training. More than four out of five clinicians reported using AI without professional education or structured guidance.
According to Blease, this creates a dangerous blind spot.
"And here's what nobody's talking about: mental health professionals are using generative AI at rates very possibly higher than patients themselves. Yet the entire public conversation about AI and mental health is fixated on the risks of patients turning to chatbots in distress. We've built a whole discourse around protecting patients from AI. While ignoring that the clinicians treating them use the same unregulated tools in clinical decision-making, with no training and no oversight. That's a staggering blind spot. It looks suspiciously like an assumption that professionals must know how to use these tools responsibly simply because they're professionals. Our data says otherwise: 81% have had no training whatsoever."
She argues that healthcare systems need to respond before AI becomes fully embedded in clinical workflows: "Training and governance can't remain an afterthought while adoption, particularly in a profession aimed at helping the most vulnerable among us, races ahead."
The researchers conclude that professional competency frameworks, evidence-based guidance and regulatory clarity are urgently needed to ensure safe integration of AI into mental healthcare.
Mixed feelings about AI, but the benefits come before fears
The study also highlights why clinicians continue adopting generative AI despite uncertainty.
Among therapists who already use these tools, nearly 83% reported that AI reduces their overall workload. In a profession struggling with burnout, workforce shortages and increasing administrative burden, these efficiency gains are difficult to ignore.
Blease stresses that this should not be interpreted solely as a warning story: "But this isn't simply alarmist. Four in five clinicians say these tools are cutting their workload – a genuine benefit in a profession buckling under burnout."
Yet she believes the greatest unresolved issue is confidentiality rather than accuracy.
"What worries me most is privacy. Clinicians are very likely pasting client details – distress, disclosures, diagnoses – into general-purpose tools never built for clinical confidentiality. That's a quiet breach happening at scale, one conversation at a time, and yet here again, the focus is entirely on patients' leaking their information and not therapists."
