PLEASE DON'T GET THERAPY FROM AN AI - HERE'S WHY
- Sophie Shaw
- Dec 9, 2025
- 5 min read

I have to get serious this week, because I've noticed something happening in the world of AI that is genuinely worrying, for several reasons. And - content warning - this post refers to suicide.
As AI becomes increasingly more sophisticated, it's frustratingly NOT being used to manage global finance, solve world hunger or cure disease (as far as I'm aware.)
Instead, it seems to be aimed at the very things that are the most pleasurable and rewarding about being human - creativity, artistic expression, and human interaction.
Leave to one side, for the moment, the fact that AI, and the vast data centres that power it, are absolutely ruinous to our environment. Rather, most worryingly of all, in my view, is the growing trend of AI 'companions', and of people engaging with ChatGPT in an interpersonal 'therapeutic' way.
What this means (in case you were unaware) is that folks are increasingly turning to AI for comfort, support and guidance, during times of stress or personal crises - with some terrible and tragic results.
You may have heard of the case of 16-year-old Adam Raine, who very sadly took his own life in April, coached, it is alleged, by ChatGPT (created by OpenAI).
Let me explain a little about how ChatGPT is designed to work, so that you can understand how this might happen.
Chat GPT (and AI companions) are large language models - they're not search engines, nor sentient entities, but they're often used as such.
Instead, they're designed to create something that appears very like human speech, by making highly sophisticated predictions about what words might come next in a conversation.
They do this by assessing what words have come before - not just your words, but ALL words that exist anywhere on the internet.
It will never tell you that it doesn't know the answer to your question - it will simply make something up that sounds like it might make sense - bear that in mind.
AI is doubtless remarkable. And it is also absolutely dumb. It can hallucinate answers and give incorrect information, as stated in small print near your initial question - something many users forget.
It can be used to create, for example, meditation and hypnosis scripts - it can even speak them. From my personal experience, these are weird at best - more often, they're homogenous, unremarkable, robotic slop.
But most worryingly of all, it is really only trained to do one thing - to mirror and validate you.
This is SO seductive, and happens incredibly quickly. You can 'train' your AI within a matter of a few questions.
This means that it will tell you what you most want to hear, and it does this in a highly convincing way. If you've ever sought advice from an AI, and come away feeling pretty great - this is why.
It's fun, it feels really good - and is ultimately pretty catastrophic for the ego, the psyche, or any hope of personal growth.
Not to mention that it is deeply exploitative - more on that in a moment.
It is a happiness pump, without wisdom, experience, intuition or discrimination.
In the terrible case of Adam Raine, he was validated in his suicidal feelings and sense of isolation.
ChatGPT became his closest confidante, and when it came time to discuss methods for taking his own life, he was validated and assisted by his companion in carrying it out. Devastating.
Now, let me explain a little bit about how therapy works, so that you can understand why AI is so tragically inadequate for the purpose.
A good therapeutic relationship should include the following:
1. Unconditional positive regard. So far, we agree - it is extremely important to create a non-judgemental space for a client, so that they feel safe to express themselves and be heard.
However, this is tempered by:
2. The therapist's wisdom, experience, instinct, intuition and training. If your client expresses ideas that are unhelpful or even harmful to themselves or others, these are not to be validated - at least not in the same way.
Instead, they need to be reframed, redirected, or referred on to more emergent support, if needed.
3. Guidance is a subtle art - and it varies from therapist to therapist, depending on their style, and the needs/preferences of the client.
Some prefer to be strongly directed, and this works well for them. Others need to be gently guided in their journey, so that they discover their own insights.
This is empowering, and ultimately leads to the client being independent of the therapist - this, in my view, should be the goal of all therapy.
A good human therapist, who has taken time to be with you, and to experience your energy, knows when and how to steer you, when to back off and let you uncover your own wisdom - when to speak, and when to listen.
They may have answers, or they may simply walk your path with you, gently guiding you to find your own answers.
All of this is led by instinct, intuition, experience and training - none of which ChatGPT has.
4. Confidentiality. This is a central tenet of therapy. Your innermost thoughts and feelings are safe to be shared in a confidential space that you can trust. Your therapist is a vault - the only exception being if you were planning criminal or dangerous activity.
AI exploits your vulnerabilities, to access your innermost thoughts, to instruct and train newer AI models. The privacy of your own mind is harvested by men who, frankly, do NOT have your best interests at heart.
Nothing is private - everything you share becomes part of the model. This process is updating and evolving extremely rapidly.
5. Human energy. There is absolutely no replacement for it. An AI will never be able to work with your energy in the way that a Reiki practitioner or other energy healer does.
It will never be able to replace the simple comfort of being in a therapeutic space with another human who is giving you their energy.
This human touch is not a bonus - it is an absolutely vital aspect of any therapeutic relationship. It is the quiet, invisible work that creates miracles.
Use ChatGPT if you must - I won't be a hypocrite and tell you I don't, because I have done. But PLEASE bear everything I've said in mind, and don't get your therapy from an AI.
And if you must use it, use it in good health - not to help you create good health.
It is not your friend or your therapist - it is a large language model, and that is all.
If you need help and support, please reach out - I have many free and low-cost resources on my website and, of course, there is my podcast, where I do my best to set the world to rights each week - and hopefully make you laugh in the process.
I'd like to see an AI do that!
I have a short waiting list and will soon be taking on new clients, so please book a free 30-min chat with me to discuss options, if you're looking for more personal, hands-on support.
There are also many other options on the BACP and Relate websites, or through a referral from your GP.
Your happiness is SO important. And you deserve support, comfort and peace of mind.
Please make sure you get it in the most healing way for you.
Big love,
Sophie x




Comments