AI, ChatGPT and Schema Therapy: What Happens When Clients Turn to AI?
At the beginning of this year, I opened my books to new therapy clients. What happened next surprised me.
80% of the new therapy clients I have seen this year are talking to AI about their mental health and interpersonal difficulties on most days, if not every day.
As I worked with these clients, I began to notice significant differences in how they engaged with therapy at almost every stage. Over time, a clear pattern began to emerge.
How were AI-using clients engaging differently with therapy?
Clients who frequently used AI seemed to engage with therapy in ways that differed from clients I had worked with in the past.
Before Therapy:
These clients were referred to schema therapy by AI. Generally, my clients are referred by a doctor, colleague, past client, or friend.
Most of these clients only sought therapy when they realised their AI discussions did not result in feeling better or desired change. Despite this, they continued to converse with AI between sessions.
These clients often had an existing AI-generated formulation of their problems, sometimes even in the schema therapy model. Clients varied in the extent to which they recognised that their formulations were influenced by their own biases, and in how strongly they regarded these formulations as accurate.
During Therapy:
Clients used AI to complete homework activities, which at times reduced opportunities to develop the skills the activities were designed to practise. In some cases, AI use also shifted the purpose of the activity or resulted in unnecessarily lengthy responses characteristic of AI-generated content
Clients used AI to generate targets for imagery rescripting and EMDR based on past discussions with AI.
Clients used AI to monitor their well-being patterns and dream themes.
AI entered the therapeutic relationship through:
The use of AI-generated information to question or challenge the formulation, recommendations, session planning and homework.
The use of AI to generate session plans, with the expectation that they would be followed.
Requests for sessions be recorded using an app that generated an AI transcript.
Clients presenting AI-generated documents for signing, with the expectation that the content would be accepted rather than independently formulated based on my professional judgement and expertise.
What happens to schema therapy when clients frequently use AI?
I began wondering whether these changes were particular to my practice or whether other schema therapists were noticing them too. I raised the issue with colleagues on the ISST listserv. Their responses suggested that I was not alone. I have shared a summary of these perspectives* below.
AI the substitute-therapist: Clients may turn to AI to meet emotional needs that might otherwise be brought to therapy, including seeking validation and reassurance, making sense of themselves and their experiences, and pursuing self-improvement. Reliance on AI for these functions may delay seeking professional support, particularly when clients initially experience the interaction as helpful but later recognise that their underlying difficulties remain unresolved. This pattern of AI use may continue after therapy has commenced.
AI is shaping clients before they enter therapy: Clients often begin therapy with an AI-generated formulation about themselves or others that they have accepted, some with excessive certainty. AI is often used to explore “what is wrong with me?” or “what is wrong with another?” Clients vary in their capacity to recognise their own biases and to critically evaluate AI-generated responses that may affirm or reinforce their existing perspective. Therapists were increasingly encountering clients seeking professional confirmation of diagnoses, schemas, or mode formulations generated by AI.
Existing coping modes linked AI use: Whilst some respondents wondered whether we would see new coping modes arise as a result of AI use, most seemed to postulate that AI use was driven by existing coping modes in the model, most often from the overcompensation group. Overcompensation modes function to avoid vulnerability through seeking certainty, control, grandiosity, domination and avoidance of responsibility.
Outsourcing Healthy Adult Mode functions: Client reliance on AI between sessions may have the net result of limiting the growth of the Healthy Adult Mode. When clients seek validation, reassurance and guidance through AI they are outsourcing the job of the HA Mode to AI.
Attempting to meet emotional needs without relational vulnerability: Reliance on AI may limit opportunities for clients to learn to meet emotional needs within relationships, as it provides access to support without requiring the emotional vulnerability, interpersonal risk, and gradual development of trust involved in relationships with therapists and other people.
The paradox: AI may make the therapy relationship more important
Although AI can be useful for psychoeducation, reflection, and organising thoughts, its responses are based on clients' input and therefore prone to client blind spots, biases, and partial perspectives. In schema therapy, some of the most clinically meaningful information emerges within the therapeutic relationship itself, through processes such as limited reparenting, empathic confrontation, experiential and here-and-now work, and rupture and repair.
Across the world, people are using AI increasingly for mental health and social support. In Australia, 1 in 4 adults are using AI for mental health advice and support and over 50% of children and adolescents are using it for personal and social reasons. This makes it necessary for schema therapists to discuss AI use with all new and existing clients. The increase in AI use makes the therapy relationship more important than ever. Schema therapy is well positioned to address these challenges of AI use due to its focus on the relationship as a core mechanism of change.
For schema therapists, perhaps the most useful starting point is simply to ask about AI use: how often clients use it, what they use it for, how much authority they give its responses, and what emotional needs or modes may be involved.
Conclusion
AI is unlikely to remain outside the therapy room. Clients are already bringing it into their formulations, coping strategies, homework, decision-making, and therapeutic relationships. Rather than treating AI use as inherently helpful or harmful, schema therapists may need to become curious about the function it serves for each client: which needs it meets, which modes it strengthens, which capacities it supports, and which experiences it may allow the client to avoid. In this context, schema therapy's emphasis on emotional needs, modes, experiential change and the therapeutic relationship may be more relevant than ever.
*Special thanks to Nefise Ladikli, Hagara Feldman, Hilary Kloss, Sonia Wiltshire, Beverley Hyde & Peregrine Kavros for sharing insights that enriched this blog article.
**This blog was written over a 2 week period. Generative AI was used solely to assist with clarity, and language refinement. The substantive content and views expressed remain those of the author.
Nadene van der LInden is a Clinical Psychologist, EMDRAA Accredited EMDR Consultant & Therapist, ISST Certified Advanced Therapist and Supervisor in Schema Therapy and Psychology Board Approved Supervisor.
