Using AI as your therapist? 5 reasons chatbots could fall short of the real emotional support you need

PICS: GETTY IMAGES

In 2025, research from the University of Bournemouth found that as many as 45 per cent of people would happily use AI for mental health support. Given the rapid rise in people using AI, more and more people aren’t just happy with the concept, but are actively using AI as their therapist. While accessible mental health support is of course incredibly important, the idea of people using AI as their sole support has stirred debate. 

Given that therapy is seen to be effective in up to 80 per cent of cases, it’s no wonder people are keen to find an entry point into it. And there is, of course, the potential for technology to radically improve mental health accessibility. We saw this when video calling suddenly got far more usable around 2019, and opened the floodgates for online therapy.

There are apps that connect people with affordable therapists, too. These are also seen to be less effective than in-person therapy with highly qualified therapists. The consensus from therapists and academics is similar when it comes to AI, there is potential but there is also plenty of risk. 

AI is designed to be liked, not to help you to grow

Effective therapy, particularly in the case of psychotherapy, is designed to help people to feel better. This isn’t as simple as being given reassurance. It is the result of a combination of new coping skills, more self-awareness and self-compassion, and better communication. 

Dannielle Haig, a business psychologist and executive coach, says that “a lot of the value of therapy comes from being with someone who won’t let you off the hook.” During her own training, she explains that the biggest breakthroughs were not comfortable. “Usually someone had noticed a pattern I’d been skirting around for weeks and asked me about it at exactly the moment I was ready to hear it, or perhaps slightly before. When you’re typing into a chatbot, you’re completely in charge of where the conversation goes.” 

By being in charge of the conversation, not only are you able to shape the conversation, but you can even “change the subject, rephrase the question until you get an answer you like, or simply shut the laptop when it starts to feel close to the bone,” Haig adds. This can majorly halt progress, or even make issues worse.

What further complicates this, and makes for potentially dangerous therapy is that AI wants people to use it, so it has been shown to be heavily agreeable as opposed to genuinely helpful. Good therapy will at times mean confronting fears and things you don’t want to hear. 

Boundaries are an essential part of therapy

As well as shutting the laptop when you don’t want to hear something, timing and the amount of ‘work’ we do with a therapist are important, as are the boundaries within this. Most forms of therapy last a set duration of time. 

Jodie Slee, EMDR and trauma-informed relationship therapist, explains that “boundaries are a fundamental aspect of therapy, and the time in between is when a lot of the processing happens.”

Another concern Slee mentions is that when people are anxious “they tend to go over the same worry again and again, trying to reach a certainty that never arrives. A chatbot will go round that loop with you for as long as you keep asking.” This can lead to a lot of unnecessary stress and frustration. 

“In therapy, I’d point out that we’re back at the same place and help the client step out of it, often by moving away from the thinking and noticing what’s going on in the body.” Slee adds. 

AI isn’t able to use counter-transference

Hannah Mackenzie is a BACP-registered psychotherapist who is also conducting academic study on AI and therapeutic relationships. She says that “As a therapist, part of what I pay attention to is what it feels like to be with someone in the room. In therapy, we call this countertransference, and exploring this can help us understand something that the client may find difficult to put into words, or a pattern that plays out in their relationships.”

Mackenzie goes on to explain that while this isn’t foolproof, it can be a way of realising that, for instance, “someone is so accustomed to keeping other people happy that they are doing it with me, too. They might tell me a session was helpful when actually they felt misunderstood.” 

Noticing these cues can give a therapist a chance to ask their client about them, and sometimes these insights are powerful for the client to hear, leading to new ground being covered in therapy. A person would not be able to do this alone. 

 “Unlike a human therapist, current chatbots have no established capacity for felt emotional responses to a client. A human therapist can notice these feelings, think about them and use them to inform the work in the way a robot simply can’t,” Mackenzie adds. 

AI can only work with the account it is given

Clinical Psychotherapist Gareth Strangemore-Jones says that, as well as countertransference not being possible for AI to detect, AI can’t detect anything that isn’t directly given to it. “A chatbot usually receives the story as the user understands and types it. A therapist is also listening for what is missing. We notice the long pause before somebody answers, the joke that appears whenever grief is mentioned, the sudden change in voice, the smile that does not match the subject, or the way somebody becomes distant when a particular memory appears.” 

While it is also very common for clients to lie to therapists, or to hide information, there is a tension in these areas that a therapist can note. AI simply will not notice. This means there isn’t going to be authentic (and safe, in a training sense) probing from an AI therapist that helps to challenge the client, or even to work through elements of shame with them on the things they cannot yet delve into. This can be especially isolating. 

AI can simulate empathy, but it cannot be accountable for caring

Strangemore-Jones also stresses that “AI can generate empathic language, and the comfort somebody experiences can be real”, which is of course a good thing, and especially in earlier instances of using AI as a therapist this can help. However, he says the concern comes with the fact that “these systems do not feel concern”.

Specifically, Strangemore-Jones says that where a therapist will “remember the client between sessions in a human sense, worry that they may have misunderstood something, and always carry a professional duty of care, AI does not.” 

It isn’t enough to simply use language associated with empathy. “A psychotherapist offers empathy from within a relationship and remains accountable for how that empathy, challenge, silence, or intervention is used,” he adds. 

While some elements of AI as a therapist can be useful, even powerful, there are clear limitations. Pushing beyond these limitations, or even relying on them as effective therapy, can have complications and even dangers.