AI Therapy and Psychotherapy are not alike
Recent publicity surrounding the AI boom has raised a familiar question: Can AI therapy replace psychotherapy?
To my mind, this is a false question. It assumes that AI therapy and psychotherapy are fundamentally similar activities—differing only in efficiency or scale—when in fact they are categorically different. They do not fail or succeed at the same task, because they are not attempting the same thing.
The most obvious difference is that therapists are human and AI systems are not. But that observation barely scratches the surface. The deeper differences concern what therapy actually is, how change occurs, and what kind of presence psychological suffering requires.
What AI Therapy Actually Is
Artificial “intelligence” derives its apparent intelligence from speed and scale. AI systems can analyse and recombine information vastly faster than human beings. AI therapy programs trawl through enormous databases containing descriptions of therapeutic models, skills, techniques, and advice. They identify patterns in how therapists respond to particular client statements and reproduce those patterns fluently.
This gives the appearance of expertise. But it is important to be precise: AI systems possess information about therapy, not therapeutic understanding.
In practice, this trawling can slide into outright plagiarism, with large sections of therapists’ and researchers’ writing reproduced without attribution or consent. (This has happened to me directly.) But even when ethically sourced, the underlying process remains the same: pattern-matching and recombination, not comprehension.
That distinction matters.
Where Chatbot Therapy Can Be Useful
AI therapy can be a genuinely helpful, low-cost option for people dealing with specific, circumscribed difficulties. For example, frustration after a setback, work stress, relationship confusion, low mood, or anxiety linked to identifiable circumstances. The fact that Chatbots are available 24/7 is another advantage.
Much like a self-help book, AI can:
-
Offer psychoeducation
-
Summarise links between thoughts, emotions, and behaviour
-
Suggest coping strategies
-
Provide journaling prompts
-
Deliver CBT or Solution-focused exercises
-
Help people name and reflect on their experiences
For some users, this is enough. For others, it is a useful starting point.
But as with self-help books, there is a catch. People may feel worse after receiving advice they cannot implement, that does not fit their situation, or which they cannot sustain when they meet with obstacles on the way.
What AI Is Good At (and Why That’s Not Therapy)
AI systems excel at:
-
Sorting and categorising problems
-
Symptom checklists and screening
-
Matching problems to therapeutic models
-
Delivering standardised advice
-
Generating summaries and clinical notes
-
Administering appointments and records
Indeed, many human therapists already use AI tools to assist with assessments, record-keeping, and decision support. In these roles, AI can be genuinely valuable.
But none of this constitutes therapy in the deeper sense, unless a human therapist is there to oversee the work.
Why Therapy Is Not an Information Exchange
Psychotherapists do acquire information through training. But information alone is never sufficient. This becomes obvious the moment one begins working with real clients.
The difficult part of therapy is not knowing what to say in general, but knowing how to respond to this person, with that problem, in this context.
-
How much explanation is helpful—and when it becomes intrusive
-
How to build trust without colluding in the client’s unhelpful ideas and behaviours
-
How to pace a conversation so it does not overwhelm
-
How to recover when an intervention lands badly
-
How to remain present with distress without empty reassurances, or rushing to fix it
These are not technical skills in the usual sense. They are relational, embodied, and situational. They are learned through years of practice, supervision, and self-examination.
An AI system has no supervisor in this sense—only software engineers refining outputs. AI therapy never struggles, never doubts itself, and never notices it has missed something important.
An analogy may be helpful here. When we use GPS navigation maps on a car journey we are given expert information on how to get from the starting point to our destination in quick time. If our journey is uncomplicated this works fine for most. But what if we are inexperienced, nervous drivers navigating a complex route? In that case having an experienced driver alongside us in the passenger seat makes a real difference, and leads to a different kind of journey.
The Therapeutic Alliance: Why the human connection matters
Only a human being knows what it is like to be human.
We are equipped with nervous systems, emotions, and the capacity for empathy precisely so we can recognise one another’s suffering. A therapist does not merely register alienation, anxiety, or despair as concepts – she recognises them from lived experience.
This is why psychotherapy is far more likely to succeed when clients feel the therapist genuinely “gets” them. The working alliance—trust, attunement, and mutual understanding—has repeatedly been shown to be one of the strongest predictors of therapeutic success.
AI cannot form such an alliance. It can simulate the language of empathy, but not the reality.
The Problem of Simulated Conversation
AI therapy systems rely heavily on stock conversational moves derived from transcripts:
-
“What’s on your mind today?”
-
“It sounds like a lot has been happening for you.”
-
“What was that like for you?”
-
“I’m hearing that took courage.”
These phrases are useful openers. But over time they reveal their limits. If repeated too often conversations begin to drift. The AI becomes overly compliant, subtly reinforcing the client’s existing frame rather than questioning or clarifying it.
Human therapists do something more demanding: they risk misunderstanding, challenge assumptions, and take responsibility for the impact of their words. AI systems cannot do this because they cannot be wrong in any meaningful sense—and cannot repair ruptures when they are.
For all these reasons a study by Stanford University this year concluded that AI therapy is best employed under live supervision for support in between psychotherapy sessions.
What AI Cannot Do
All the core limitations of AI therapy stem from one fact: it processes information, but does not participate in experience.
AI cannot:
-
Detect when it’s statements are in fact hallucinations
-
Recognise irony, nuance, or emotional contradiction accurately
-
Sense when interventions are premature, invasive or a poor match for the user
-
Work with silence as a living presence rather than a gap
-
Engage in nervous-system-to-nervous-system co-regulation
-
Participate in relational repair for attachment wounds
-
Use face-to-face interactions as a source of diagnostic insight
-
Hold ambiguity without forcing coherence
-
Take ethical responsibility for risk, harm, or failure
Many psychological wounds are relational in origin—neglect, abandonment, betrayal. Healing requires not just information and explanation, but a relational encounter with another conscious, responsive human being.
AI can mimic the form of this encounter. It cannot enter it.
Accountability and Care
Human therapists are accountable:
-
To ethical codes
-
To supervisors and peers
-
To professional regulation
-
To their own conscience and intuition for how things are going
They can acknowledge mistakes, repair harm, and hold responsibility in situations involving risk, suicidality, or safeguarding. AI systems cannot do this. They cannot be trusted with vulnerability in the same way because they cannot stand behind their interventions.
Where AI Belongs
AI does not replace psychotherapy. It complements it.
Used well, AI can support:
-
Accurate assessment and diagnosis
-
Psychoeducation
-
Exercise steps
-
Reflection and journaling
-
Session summaries
-
Administrative efficiency
But therapy itself—the encounter in which suffering is met, clarified, and sometimes released—remains irreducibly human.
The question, then, is not whether AI can replace therapy. It is whether we still understand what psychotherapy actually is.
And that question matters far more.