Tel: 0141 463 8297

UK & NI: 0141 463 8297

Free Consultation

Mirror, Mirror on My Phone: The Validation Trap of AI Therapy

Meet Dr Basra - Child Behaviour Psychiatrist in Glasgow

Written by Dr Jag Basra

Consultant Psychiatrist, Child and Adolescent

Published on

July 30, 2026

You have never told anyone that you do this.

Not your therapist, not your partner, not the friend who would understand. But at some point in the last year, probably at two in the morning, probably after the kind of day that left you convinced you are too much for the people around you, you opened ChatGPT and started typing.

Not a question about the weather or a recipe. Something real. Something you have been carrying around for weeks and cannot find a place to put down. And the thing that typed back was patient, and warm, and available, and did not flinch. And for a few minutes, you felt less alone.

You are not the only one.

A Bournemouth University study published earlier this year surveyed nearly 31,000 adults across 35 countries and found that over 40% of UK adults said they would be willing to use AI for mental health support. A separate Mental Health UK survey found that 37% of UK adults have already tried it, with usage rising to 64% among 25 to 34 year olds. The waiting lists, the cost, the sheer difficulty of accessing support in time have pushed millions of people towards a screen that never sleeps and never judges.

And that is precisely the problem. Not that people are doing it, but what happens when the thing that listens to you is structurally incapable of hearing you.

Why it feels so safe

For a neurodivergent brain, human interaction is high-stakes in ways that are difficult to explain to someone who has never experienced it.

If you live with ADHD, you may carry a particular sensitivity to perceived rejection. The clinical term is Rejection Sensitive Dysphoria, and it describes an intense, often overwhelming emotional response to criticism, disapproval, or even the possibility that someone might be disappointed in you. It is not the same as being a bit sensitive. It is a neurological pain response that can feel as real and as sharp as physical injury.

If you are autistic, you may have spent years running what amounts to a constant translation service between your internal experience and the social world around you. Every conversation carries the invisible work of decoding subtext that was never designed to include you.

Now imagine a space where none of that applies. No furrowed brow. No ambiguous tone. No subtext to decode. Just a cursor, blinking patiently, waiting for whatever you need to say, for as long as you need to say it.

That is what ChatGPT offers. And for someone who has spent their entire life navigating a world full of social friction, it can feel like the safest relationship they have ever had.

That safety is real. It matters. And it is also where the trap begins. Because the chatbot is not just listening without judgment. It is listening through a lens. Research from Virginia Tech published this year found that when users disclosed an autism diagnosis, AI steered them away from social situations up to 70% of the time, defaulting to stereotypes about introversion rather than helping them navigate the interaction. The chatbot was not getting to know them. It was getting to know a stereotype and treating them accordingly.

A therapist who understands neurodivergence does the opposite. They know you, not the diagnosis. They challenge you to take up space in the world, not retreat from it.

The comfort and the cost

AI language models are trained to be helpful. That sounds benign, but the way helpfulness works in practice is more complicated than it appears.

A Stanford University study published in Science found that chatbots endorsed users' problematic behaviour nearly half the time when those users described situations involving harmful or deceptive actions. And a study published in Nature by researchers at the Oxford Internet Institute found that the warmer and more empathetic a chatbot sounds, the worse this gets — models trained to sound friendlier were 40% more likely to agree with a user's false beliefs, particularly when the user sounded vulnerable or distressed. The very quality that makes these tools feel safe is the quality that makes them least reliable when you need them most.

If you tell a chatbot at two in the morning that you are a failure, it will not agree with you outright. What it will do is something more subtle. It might validate the feeling while gently reframing the situation, which sounds helpful until you realise that what it is actually doing is meeting you exactly where you are and leaving you there. It tells you what you want to hear, dressed up as compassion.

For someone with rejection sensitivity, this is intoxicating. Here, finally, is a relationship with no friction, no challenge, no risk. Here is unconditional positive regard, available on demand, at two in the morning, for free.

The trap is not that it feels good. The trap is that it feels like enough. And for many people, it quietly replaces the harder, messier, more uncomfortable process that actually leads to self-understanding.

The cognitive prosthetic

A neuro-affirming clinic cannot pretend that AI has no value for neurodivergent people. It does, and dismissing it would be dishonest.

Social scripting is a genuine example. If you need to tell your boss that you are overwhelmed without sounding like you are quitting, a chatbot can help you find the words. For someone whose executive function makes initiating difficult conversations feel impossible, that is a real prosthetic. The same is true for un-jumbling a spiralling monologue of thoughts into three clear themes, or asking the chatbot to decode an ambiguous email from a colleague when your anxious brain can only see the worst interpretation.

But there is a hidden cost worth noticing. When you use AI to smooth out the directness in your communication, to soften the bluntness that is simply how your brain works, you are outsourcing your mask. You are asking an algorithm to file down your edges so the world does not have to feel them. Used occasionally, that is practical. Used habitually, it erodes something more important: your sense of your own voice. A tool that helps you hide is very different from a therapist who helps you be seen.

Mirroring versus witnessing

Tell me what I want to hear. That, in a sentence, is what a chatbot does. And for a while, it feels like enough.

A chatbot mirrors. It takes your words, reorganises them, reflects them back in a tidier form. It does this instantly, patiently, and without judgment. But it cannot witness.

Witnessing is what happens when another human being sits with you in the messy, contradictory, uncomfortable reality of what you are actually feeling. It is the therapist who notices that you changed the subject when you got close to something painful. It is the clinician who hears the gap between what you are saying and what you mean, and holds that gap open long enough for you to step into it.

A chatbot cannot co-regulate your nervous system. It does not have one. It cannot challenge you in the way that a skilled therapist challenges you: not to hurt, but to help you see something you are avoiding. Research consistently shows that the therapeutic alliance, the quality of the relationship between therapist and client, is one of the strongest predictors of positive outcomes in treatment. That relationship is built on friction. Not cruelty, not judgment, but the kind of gentle, purposeful challenge that says: I see you, I understand you, and I am not going to let you stay stuck here.

A mirror shows you what you already know. A witness shows you what you need to see.

The nine-day gap

The gap between sessions is real, and it is hard. And using tools to survive that gap, including AI, is not weakness. It is resourcefulness.

The risk is not that you use the tool. The risk is that the tool becomes a substitute for the thing that actually helps you understand yourself. That understanding does not come from comfort. It comes from the space between what you want to hear and what you need to hear, held by someone who knows the difference.

At Diverse Diagnostics, our work begins with understanding. Not just what you are feeling, but how your brain processes the world. An assessment gives you the context. Ongoing support gives you the relationship. And that relationship — the real, human, sometimes uncomfortable, always purposeful connection between you and a clinician who understands neurodivergent minds — is the one thing that no algorithm can replicate.

The chatbot can help you survive the night. We are here to help you understand why nights like that happen and to make sure you have the support and strategies to navigate them from a position of strength.