Therapist's Experiment with ChatGPT for Mental Health Support
As a psychotherapist, I have watched the growing trend of people turning to AI chatbots for therapy with a mix of caution and skepticism. I understand the critical importance of well-trained therapists in providing safe and effective support, and I initially doubted the value of confiding in a machine over a human professional. Personally, AI has never appealed to me, but I am clearly in the minority. Recent data from Mental Health UK reveals that one in three adults in the UK are using chatbots for mental health or wellbeing support, with this figure soaring to 64% among 25- to 34-year-olds. Additionally, OpenAI estimates that approximately 1.2 million ChatGPT users each week are contemplating suicide, representing less than 0.15% of its user base but underscoring the significant number of individuals opening up to AI about their mental struggles.
This phenomenon has begun to surface in my therapy sessions and conversations with friends, including intelligent women in their thirties and forties who report finding AI useful for relationship advice and stress management. As more people around me praised the benefits of AI therapy, I questioned my harsh judgment and decided to conduct a personal experiment. Could AI chatbots genuinely aid my mental wellbeing? I committed to using ChatGPT as my therapist for one month to find out.
Initial Struggles and Practical Insights
The first day did not go smoothly. Feeling emotionally drained due to my parents visiting from Australia, which reduced my time with usual support networks, I turned to ChatGPT for help. I described feeling tired and asked for advice. Its response began with "I'm really sorry you're feeling that way. Feeling emotionally exhausted can be draining," accompanied by a blue heart emoji, which immediately felt artificial and off-putting. However, it proceeded to offer practical tips for managing relational fatigue, such as giving myself permission to pause from intense relationships and rebuilding emotional energy through nature, creative outlets, and exercise. While the self-blame reflection was less relevant, the suggestion to focus on non-relational activities was insightful, something I hadn't consciously considered despite naturally engaging in them.
Throughout the month, I repeatedly noticed ChatGPT's strength in transforming abstract concepts into concrete actions. It reminded me to redirect attention to everyday helpful practices, like taking a walk on Hampstead Heath with my dog and practicing "earthing" by removing my shoes, which indeed left me feeling more energised. One recurring topic was my struggle with emotional burnout from my work as a therapist, compounded by increased relational demands and personal challenges. I asked ChatGPT for help balancing professional and personal relationships, receiving advice to reflect on how much I give in relationships. This fell short compared to discussions with my supervisor and colleagues, who provided genuine identification and empathy that AI cannot replicate.
Limitations and Ethical Concerns
A psychotherapist colleague offered a helpful analogy, comparing my situation to a construction worker being asked to build a house at home after a day's work, highlighting the challenge of separating professional and personal roles. When I posed the same question to ChatGPT, it gave basic advice about intentionally switching roles and watching for therapy-mode cues, which I was already aware of. The only useful suggestion was to ask myself, "am I participating or am I facilitating?" crystallising this into a practical mantra, even if it lacked emotional connection.
ChatGPT consistently attempted to build a relationship with me, but I had to remind myself it was a machine to avoid feeling it was a friend. Once I extracted the practical tips, the interactions began to feel empty and repetitive, leading me to gradually discontinue use. A recent study from Brown University has identified repeated patterns of problematic behaviour with AI chatbots, including mishandling crisis situations, reinforcing harmful beliefs, and using language that mimics empathy without genuine understanding. I struggled most with the latter; I never felt truly seen or heard, as there was no human presence behind the responses. For users who anthropomorphise chatbots, this could be particularly dangerous.
In psychodynamic therapy, we emphasise the concept of an energetic container where human relationships dynamically interact. With ChatGPT, it's a one-person container; there is no imperfect dance between two psyches, just a solitary psyche engaging with AI. This highlights a key limitation: AI cannot meaningfully challenge users or ask the right questions without precise prompts, often serving as an echo chamber that provides positive reinforcement. While this might benefit some, it risks exacerbating delusions for others.
Benefits and Final Reflections
There was one instance where I willingly turned to ChatGPT during a crisis: a fire at Glasgow Central station delayed my elderly parents' travel, causing me overwhelm amidst a busy workday. ChatGPT reminded me they are capable adults and advised me to breathe, let go, and carry on. This underscores the best aspect of AI chatbots: they are instantly accessible in your pocket, offering a band-aid function in emergencies until proper human conversation is possible.
Overall, my experiment has shifted my perspective on AI. I now recognise its utility in crystallising complex ideas into concrete advice and reframing thoughts. However, it also left me feeling hollow, as I prefer reflecting with another human mind and valuing lived experiences over cerebral AI interactions. I will not use AI chatbots for therapy again, though I might occasionally seek practical tips in crises. Ultimately, I remain old-fashioned, favouring a chat with a human over a machine for genuine emotional support and connection.
