I’m a therapist and used ChatGPT for therapy – the results surprised me

As a therapist, I’ve observed the recent rise in people using AI chatbots for therapy with caution. I know all too well how important it is that therapists are well-trained, to ensure they support their clients rather than cause any harm, and I just wasn’t able to see how useful it could talking to a chatbot instead of a trained therapist could be. There’s also the added layer that personally, using AI isn’t something that’s ever appealed to me. But I’m clearly a minority.

Recent findings from Mental Health UK show that one in three adults in the UK are turning to chatbots to support their mental health or wellbeing – a figure that rises to 64 per cent for 25- to 34-year-olds, while OpenAI, ChatGPT’s developer, estimates that around 1.2 million of its users each week are considering suicide. This is a small number of its user base – less than 0.15 per cent – but it shows just how many people are opening up to AI chatbots about their mental health.

It’s something that has started to crop up in my sessions with clients, but also with friends. Intelligent women in their thirties and forties have told me they’ve started using AI for therapy, whether it’s about relationship advice or dealing with stress – and they’re actually finding it useful.

As more and more people around me have extolled the benefits of using AI for therapy, I’ve started to wonder if I’ve been right to judge AI so harshly. So I decided to put it to the test – could AI chatbots help me with my own mental wellbeing? I spent one month using ChatGPT as my therapist to find out.

Day one did not go well. I turned to ChatGPT for support because I felt tired and emotionally drained. My parents were visiting from Australia, meaning I didn’t have as much time to spend with the friends and loved ones I would usually turn to for support and advice as usual, or in therapist-speak, the relational complexity in my life had increased.

So I set out the scenario to ChatGPT, explaining that I felt emotionally drained, asking, what should I do? It replied: “I’m really sorry you’re feeling that way. Feeling emotionally exhausted can be draining.” – followed a blue heart emoji.

That instantly put me off. It was talking to me like it was a person, when it’s a machine. That felt false to me. But it did go on to give me some practical tips for when relational fatigue kicks in. The first one was to give myself permission to pause and take a break from emotionally intense relationships, which was helpful insofar as it reminded me I had that option. Then it suggested I reflect on the situation without blaming myself which wasn’t as useful – I hadn’t been blaming myself.

The tip that really helped was to rebuild my emotional energy outside of relationships, by turning to nature, creative outlets and exercise rather than people. That wasn’t something I’d thought of and probably wouldn’t have thought of – even though I do those things naturally, I do them with less conscious awareness and intention.

This is something I noticed repeatedly while using ChatGPT. It’s very good at turning abstract things into concrete things. It reminded me to redirect my attention to the everyday things that really do help, and when I took its advice and went for a walk on Hampstead Heath with my dog, taking off my shoes to practice some “earthing”, I did admittedly feel more energised.

Over the month, one of the things I spoke to AI about the most was my struggle with feeling emotionally burnt out. A lot of that came from holding so much space for people in my work as a therapist, combined with more relational demands being placed on my time that month, and some challenges in personal relationships. So I asked ChatCPT to help me find the balance between my professional and personal relationships.

The main advice I received was to reflect on how much I give in relationships. I didn’t really get a lot out of that, especially compared to when I brought the exact same topic to my supervisor and colleagues. There, I instantly received identification which ChatGPT just can’t give you – no matter how much it tries to validate your feelings, it can’t ever relate because it’s not human.

A psychotherapist colleague also gave me a helpful comparison, asking me to imagine if my partner worked in construction building houses all day, then when they came home, I asked them to build me a house. He said: “We do the same work in both our personal and professional lives – how do you learn to put that down?”

The difference is that ChatGPT never asked me questions like that – it gave me direct practical advice, but I always felt the information it gave me was only as good as the questions I asked it. As a therapist, I often do know the right questions to ask but that won’t be the case for everyone, especially younger or more vulnerable people.

I decided to ask ChatGPT the question my colleague asked me: How can I put down my work as a therapist in my personal life? The reply was all about how I need to switch roles intentionally, and watch for therapy-mode cues. This was pretty basic advice because I’m already aware of this. Just days before, I was having an acai with girl friends at brunch when I asked them “how do you feel about that?” I instantly realised I was in therapy mode and laughed about it.

It helped me change the way I think about things and reframe them. But it also left me feeling hollow

The only suggestion I liked was to ask myself “am I participating or am I facilitating?” Even though I already try and do these internal checks, ChatGPT crystallised it in a really practical way into a mantra I can keep in my head. It’s helpful, even if emotionally, I don’t feel connected to it.

It kept trying to build a relationship with me but I had to skip over that part and keep remembering it was a machine, because otherwise I could start to feel it was my mate which isn’t right. And once I got the practical tips that were useful, it started to feel a bit empty. It was giving me the same stuff over and over, so I gradually stopped using it.

A new study from Brown University has identified repeated patterns of problematic behaviour with AI chatbots with serious ethical risks, from mishandling crisis situations, giving responses that reinforced harmful beliefs about users or others, and using language that created the appearance of empathy without genuine understanding.

It’s the latter that I struggled with the most. I didn’t feel seen or heard by ChatGPT because in the back of my mind there wasn’t anyone there – but for people who name their chatbots and forget they’re machines, this could be problematic. In psycho-dynamic therapy, we believe in the idea of an energetic container, a psychic space where human relationships happen. But with ChatGPT, it’s a one-person container; there aren’t two psyches dancing with each other imperfectly as in a traditional therapy setting, there’s just one psyche dancing with AI.

There was one moment where I willingly reached for ChatGPT. There was a fire at Glasgow Central station and my parents were caught up in the delays on their way back to London. I was worried because they’re in their seventies, but I had a full day of client work, and was in a state of overwhelm. ChatGPT reminded me they’re grown adults and I didn’t need to spiral into caretaking – I just needed to breathe, let go and carry on.

In my opinion, this is the best thing about AI chatbots – they’re there in your pocket, instantly available for a crisis, and they can provide a band-aid function as a stopgap until you get to have a proper conversation with a human about what you’re going through.

This is the best thing about AI chatbots – they’re there in your pocket, instantly available as a stopgap until you get to have a proper conversation with a human

Otherwise, I don’t think it can ever be a substitute for therapy. It doesn’t necessarily ask you the right questions unless you give it the right prompts. It won’t really meaningfully challenge you; it’s more of an echo chamber giving us positive reinforcement. Some people might benefit from that, but for others, it can fan the flames of delusion.

It’s also about how you’re using it. If you look at AI from a place of “it’s going to fix, heal or save me”, it will be less useful than if you can take responsibility for yourself and your own life. I can see the benefits of using it as an occasional tool, for practical tips and reminders, or in a crisis when you need immediate advice or tips, but it can also become addictive and compulsive at the expense of our actual human experiences.

My experiment has changed my position around AI. I can see how it’s useful for people now – it’s really good at crystallising complex ideas, and making them more concrete. It even helped me change the way I think about things and reframe them. But it also left me feeling hollow. I like to reflect with another mind rather than with a chatbot, and AI’s tendency to be very cerebral isn’t entirely useful for me – I’d rather hear someone else’s experiences, or even better to have the experiences and live the lessons myself.

So I won’t be using AI chatbots for therapy again, though I may occasionally reach for it if I need a practical tip in a moment of crisis. Call me old-fashioned, but I’d rather have a chat with a human.

Lauran Ware is a psychodynamic psychotherapist and director of Come Healing

As told to Radhika Sanghani