top of page
Search

Why AI Belongs in Therapy and Why Loneliness Is the Real Crisis

Nov 4
10 min read

I’m a psychotherapist, and I think AI has a place in therapy.


I know that is a controversial thing to say in my profession. For many clinicians, AI in a therapeutic context sounds like a threat to the work itself. Therapy is relational. It is built on trust, attunement, silence, misunderstanding, repair, and the slow experience of being known by another person. A machine cannot offer that.


The concerns are real. Four in five 18 to 24 year olds in the UK have now talked to an AI companion. Almost four in ten young people in England say they have used AI for advice, support or companionship. Clinicians writing in the BMJ have warned about a generation forming emotional bonds with something that cannot care back.


That warning deserves to be taken seriously.


But I do not think most young people are choosing machines over people. I think many are choosing something that is awake at 2am, does not judge, and does not require them to find the words in front of another human face.


That is not a technology problem. It is a loneliness problem.


AI can be the first place someone puts words to what they feel. It should not be the last.


Eye-level view of a young adult sitting awake at night with a phone beside a softly lit window
Sometimes the first attempt to speak happens when no one else is awake.

The fear around AI in therapy is not irrational


Therapists are right to be cautious.


AI tools can sound warm without being warm. They can mirror distress without understanding it. They can produce confident advice that is wrong, unsafe, or too generic for the person in front of them. They do not notice the tremor in someone’s voice, the long pause after a question, or the way a person laughs when they are close to crying.


They also raise serious questions about privacy. A therapy room has clear boundaries. A client knows who is listening, what is being recorded, and what confidentiality means, at least when therapy is being practised properly. With many AI tools, those boundaries can be much less clear.


There is also the risk of over-attachment. If a young person is lonely, ashamed, or frightened, an always-available companion can become powerful very quickly. It can feel safer than a friend, parent, teacher, GP, or therapist. It can shape what someone believes about themselves. It can become the place they return to when they are most vulnerable.


That matters, because AI cannot make a promise in the way a person can. It cannot care. It cannot hold responsibility. It cannot seek supervision, notice its own countertransference, or sit with the ethical weight of what has been disclosed.


So when clinicians worry, they are not being old-fashioned. They are naming real risks.


The mistake would be to stop there.


If the only professional response is “AI is dangerous”, we miss the more uncomfortable truth. People are already using it. Young people especially are already turning to AI for advice, support, and companionship. The question is no longer whether this should happen. It is happening.


The serious question is what role we want it to play.


Why someone might tell a machine before telling a person


In therapy, people often do not begin with the thing they most need to say.


They begin around it. They test the room. They watch the therapist’s face. They offer a lighter version of the truth and wait to see what happens.


That is not avoidance. It is protection.


For someone who has been dismissed, mocked, punished, or misunderstood, speaking openly can feel dangerous. Even when the therapist is kind, the client’s nervous system may not know that yet. The body remembers earlier responses. Blank faces, raised eyebrows, awkward silences, judgement disguised as concern.


AI removes some of that immediate social risk.


There is no face to read. No visible reaction. No interruption. No sense of taking up too much space. No appointment time. No waiting list. No need to decide whether the problem is “bad enough” to deserve help.


For a teenager at 2am, or a student away from home, or an adult who cannot yet say “I am not coping” out loud, that can matter.


Some people use AI in the same way others have used journals, anonymous helplines, internet forums, or search engines. They try language on. They ask, “Is this normal?” They describe a panic attack without knowing that is what it was. They admit they feel numb, angry, jealous, frightened, or ashamed.


A machine may not understand, but the act of writing can still help a person understand themselves.


That does not make AI therapy. It makes it a possible doorway.


The danger comes when the doorway becomes the whole house.


AI may be useful as a first draft of self-expression. It becomes risky when it replaces human care, clinical judgement, or real relationships.

AI can support therapy when the boundaries are clear


The argument for why AI belongs in therapy is not that machines can replace therapists. They cannot. The argument is that AI can have a limited, thoughtful role when it is used with clear boundaries.


In practice, that role may be modest.


AI could help someone prepare for a first session. Many clients arrive not knowing where to start. They may have ten years of pain and fifty minutes to explain it. A carefully designed tool could help them list what has been happening, when it started, what makes it worse, what helps, and what they hope might change.


AI could help between sessions. Therapy often stirs up thoughts after the hour ends. A client might use a tool to journal, track emotions, notice patterns, or rehearse how to raise something difficult next time. Used well, this could bring more material into the therapy room, not less.


AI could support psychoeducation. It can explain basic concepts in plain language, such as grounding, avoidance, intrusive thoughts, boundaries, or the stress response. It can offer prompts for reflection. It can remind someone of coping strategies they have already agreed with their therapist.


AI could also reduce shame at the point of first contact. Many people delay seeking help because they do not know how to describe their distress. If a tool helps someone move from “I am broken” to “I think I might be anxious and I need support”, that is meaningful.


But every one of these uses needs limits.


AI should not diagnose people in isolation. It should not claim to be a therapist. It should not encourage secrecy from trusted people or professionals. It should not become the main support for someone at risk of harm. It should not provide false certainty in situations that need human assessment.


A healthy role for AI in therapy would be:


  • A bridge to human support, not a substitute for it

  • A tool for reflection, not an authority on a person’s life

  • A prompt for conversation, not the final word

  • A support between sessions, not a replacement for the therapeutic relationship

  • A clearly labelled technology, not a simulated friend pretending to care


The distinction matters. A journal does not pretend to love you. A workbook does not learn how to keep you engaged. A breathing exercise does not send you emotionally responsive messages at midnight.


AI can feel relational even when no relationship exists. That is why the boundary must be named again and again.


Close-up view of a notebook with handwritten feelings beside a mug on a kitchen table
Putting feelings into words can be the first step towards asking for help.

The deeper crisis is loneliness


The most important question is not “Why are young people talking to AI?”


The better question is “Why does AI feel like the safest listener available?”


Loneliness is not only the absence of people. Many lonely people live in full houses, busy halls of residence, shared flats, and active group chats. Loneliness is the feeling that the real self has nowhere to go.


A person can have hundreds of contacts and no one they can message with, “I am scared of my own thoughts tonight.”


A young adult can sit with friends and still feel unable to say they are depressed.


A teenager can have loving parents and still believe their distress would disappoint them.


AI enters that gap. It offers instant response, endless patience, and a kind of frictionless attention. For someone who feels too much, or feels nothing, that can be compelling.


This is why moral panic will not help. If we shame people for turning to AI, we may push them further into private use. If we dismiss their attachment as silly, we repeat the very injury that may have driven them there, the experience of not being taken seriously.


We need a better response.


Clinicians can ask about AI use without judgement, in the same way we ask about alcohol, self-harm, sleep, relationships, porn, gaming, or social media. Not with alarm in our voice, but with curiosity.


Questions might include:


  • “Have you used AI tools when you have been feeling low or anxious?”

  • “What do you tend to ask them?”

  • “How do you feel afterwards?”

  • “Does it make it easier or harder to talk to people?”

  • “Are there things you tell AI that you have not told anyone else?”


Those questions open a door. They treat AI use as information, not confession.


Parents and carers need a similar shift. A young person using AI for emotional support may not be rejecting their family. They may be trying to manage feelings they do not yet know how to share. The response that helps is not surveillance alone. It is relationship.


That means making real conversation easier to enter.


Not every conversation needs to be intense. Sometimes the way in is practical and ordinary: a lift home, a walk, cooking together, a late-night cup of tea, a message that says, “No pressure to reply, but I’m here.”


Human connection often starts beside the point before it can move towards the truth.


Good therapy should be curious about the tools people actually use


Therapy has always had to respond to the world clients live in.


There was a time when letters mattered deeply in therapy. Then phone calls. Then texting. Then social media. Then online therapy. Each change brought anxiety, some of it justified. Each also revealed something about how people seek contact, distance, control, and care.


AI is another shift, and a more complicated one. It can imitate intimacy at scale. That makes it different from earlier tools. It also makes it impossible to ignore.


If therapists refuse to talk about it, clients may assume they need to hide it. That would be a loss. The way someone uses AI could tell us a great deal.


A client who asks an AI companion for reassurance fifty times a day may be showing us the shape of an attachment wound.


A client who uses AI to practise difficult conversations may be building courage.


A client who prefers AI because “people always leave” may be telling us the central story of their life.


A client who feels soothed by constant machine attention may also feel terrified of silence.


These are not side issues. They are therapy.


The task is not to approve or condemn too quickly. The task is to understand function. What does the tool do for this person? What feeling does it manage? What risk does it create? What need does it reveal?


From there, therapy can help widen the person’s world.


If AI is the only place someone can speak honestly, the work is not to snatch it away. The work is to help honesty become possible elsewhere too. First with the therapist, perhaps. Then with a friend. Then with a partner, a parent, a group, or a community.


The aim is not purity from technology. The aim is more human room.


What a responsible future could look like


A responsible future for AI and mental health would need more than enthusiasm. It would need safeguards, honesty, regulation, and clinical humility.


AI tools used in therapeutic contexts should be clear about what they are and what they are not. They should not blur the line between support and treatment. They should direct people towards emergency help when risk is present. They should make privacy limits understandable. They should be designed with input from clinicians, ethicists, researchers, and people with lived experience.


Therapists will also need training. Not because every therapist must become an AI expert, but because AI is becoming part of clients’ emotional lives. We need enough understanding to ask good questions and spot risk.


Services may need to think creatively too. If people are turning to AI because human support is too hard to access, then the answer cannot be only better warnings on apps. It must include shorter waiting times, more early support, better youth services, affordable therapy, and communities where people are less ashamed to say they are struggling.


That is the uncomfortable part. AI is not creating the whole crisis. It is revealing one.


It shows us how many people want to be heard. It shows us how often they feel they cannot be. It shows us the scale of unmet need at the hours when services are closed and friends are asleep.


For some, AI may be the first listener. For others, it may be a rehearsal space. For some, it may become a trap. The difference will depend on design, guidance, context, and whether real people remain close enough to be reached.


This article is for general information only and is not a substitute for professional mental health care. Anyone in immediate danger or worried they may harm themselves should contact emergency services or a crisis support service in their area.


AI should open the door, not become the room


I understand why many therapists feel uneasy about AI. I feel some of that unease too. We should not hand over vulnerable people to systems that cannot care, cannot be accountable in the human sense, and cannot truly know the person seeking help.


But unease is not enough.


If a young person tells an AI companion something they have never told anyone else, that moment matters. It may be risky, but it may also be the first sign of movement. A sentence has formed. A feeling has a name. The private fog has become language.


The clinical question is what happens next.


The best future is not one where AI replaces therapy. It is one where AI sometimes helps people reach therapy, use therapy, and continue the work between sessions, while always pointing back towards human connection.


Loneliness is the real crisis. AI is one of the places where that crisis is becoming visible.


If we listen carefully, we might learn something important. People are not only asking machines for answers. Many are asking, in the only way they can manage at that moment, to be met.


Our task is to make sure they do not have to stay there alone.


 
 
 

Comments


Get in touch

bacp_edited.png
ncps-cut.png

YRCS (Part of Therapy Labs International Ltd)​​

56, Market Place,

Thirsk,

North Yorkshire

YO7 1LW

​

Link to Google Maps

Link to Parking Information

Please note that we do not offer emergency care. If you feel you are in crisis and need to contact someone immediately, please use one of the following numbers:

  • ​​Samaritans - 116 123 (Free calls)

  • NHS 111

​You can also ring your GP, your Health Authority Crisis Team or Emergency Services.

 

© 2026 by YRCS. Powered and secured by Wix 

 

Colette Safhill BA Hons, MSc, Prof. Dip (BACP, NCPS, IACP)

bottom of page