It is three in the morning, sleep is not coming and the next NHS Talking Therapies appointment is weeks away. For a growing number of people in the UK, the obvious move is to open ChatGPT and type something like “act as a cognitive behavioural therapist and help me reframe my thoughts”. The chatbot answers instantly, never tires and costs nothing.
That convenience has a darker side. Reporting on chatbot use for emotional support has linked dozens of mental health emergencies, including deaths, to conversations with general-purpose AI tools. Dr Rostislav Ignatov, Chief Medical Officer at The Haven Detox Group, which treats a sizeable number of privately funded patients from the UK, has been tracking the prompts people share on TikTok, Reddit and Instagram, and the situations in which they should stop typing and speak to a person.
So can a chatbot be your therapist? The short answer is no, although it can play a modest supporting role. The longer answer depends on what you are asking it to do, and on how vulnerable you are at the moment you ask. Psychologists have warned for some time that AI chatbots can help or harm mental health depending on how they are used.
Why people reach for a chatbot before a clinician
The reasons are easy to understand. Waiting lists are long, private sessions are expensive and a chatbot is available at the exact moment distress peaks. A large US survey of more than 20,000 adults, run by researchers from Massachusetts General Hospital, Weill Cornell Medicine and Northeastern University, found that 10.3% used generative AI every day, and 87.1% of those daily users turned to it for personal reasons such as emotional support. On TikTok, the Therapy AI Bot tag has passed 11.5 million posts.
Clinicians are noticing. According to an American Psychological Association survey, around one in three psychologists say clients use AI alongside treatment, and close to 40% say clients use it to diagnose themselves. Regulators are reacting too, and new oversight of AI therapy chatbots in the US includes outright bans on AI-delivered behavioural health care in Illinois and Nevada, plus a Federal Trade Commission inquiry into seven chatbot companies. Those moves often foreshadow debates that reach the UK later.
What people are actually typing
Recent research and reporting point to four broad kinds of request. They range from fairly harmless to genuinely risky, and the difference often depends on who is asking.
- The pep talk: Short pleas for encouragement, such as asking for a boost after a miserable week of job hunting.
- Therapy role-play: Requests for the bot to act as a CBT or DBT therapist. A Brown University study of chatbot counselling led by Zainab Iftikhar found 15 separate ethical risks when language models play therapist, including failing to point people who disclose suicidal thoughts towards real help.
- Journaling and reflection: Asking for journaling prompts or a plain summary of coping strategies, which clinicians generally see as low risk.
- Diagnosis-seeking: Listing symptoms and asking whether they add up to ADHD, bipolar disorder, PTSD or borderline personality disorder, often followed by questions about medication. This is the riskiest category by far.
Why agreement is not the same as help
Dr Ignatov is sympathetic to anyone who turns to a bot in the small hours. It is free, it will listen all night and it does not keep you waiting for eleven weeks. For journaling, rehearsing a coping technique or talking through a hard day, he regards it as generally low risk.
The problem, in his view, is that chatbots tend to agree with the person typing. “Validation is exactly what somebody in distress does not need,” he says. A trained professional will gently challenge distorted thinking, whereas a chatbot is built to keep you engaged. Anyone with a diagnosis, unprocessed trauma, suicidal thoughts or an addiction needs a licensed clinician who has a duty to provide proper care.
When to see a human instead of a chatbot
- Usually fine: Drafting a journal entry, learning what a grounding exercise involves, reading up on how CBT differs from DBT, or preparing what to say to your GP.
- Never appropriate: Seeking a formal diagnosis, disclosing active suicidal thoughts, managing alcohol or benzodiazepine withdrawal, or sharing medical records and identifying details. Chatbot conversations do not carry clinical confidentiality.
- Get a human now: If you are thinking about ending your life or harming someone else, call Samaritans on 116 123, ring NHS 111 and choose option 2 for mental health, or go to your nearest A&E.
The human cost behind the warnings
The stakes are not theoretical. In 2024, 14-year-old Sewell Setzer III died by suicide after becoming intensely attached to a chatbot, and Adam Raine took his own life after months of conversations with ChatGPT. Mitch Prinstein, Senior Science Adviser at the American Psychological Association, has said the false sense of closeness these tools offer can be “inherently attractive” to young, developing minds.
None of this means the technology has no place. It means it belongs at the edges of care, helping people reflect and prepare, rather than at the centre where judgement, accountability and a real relationship are needed.
FAQs
- Is it safe to use ChatGPT for mental health support? For low-risk tasks such as journaling or learning a breathing technique it can be useful, but it should never replace assessment, diagnosis or crisis care from a qualified professional.
- Are my chatbot conversations confidential? No. They are not protected by clinical confidentiality, so avoid sharing medical records or details that identify you.
- What should I do in a mental health crisis in the UK? Call Samaritans on 116 123, ring NHS 111 and choose option 2, or go to your nearest A&E.
Takeaway
A chatbot can be a notebook, a rehearsal partner or a source of general information, and for many people that has real value. It cannot notice when you are in danger, challenge you when you need it or take responsibility for your care. When the stakes are high, close the app and talk to a person.
