Most of us have dabbled in using ChatGPT or other AI platforms and indeed capitalising on its ability to make work and life easier. With such incredible speed and efficiency, why wouldn’t we! AI has entered the mental health landscape via chatbots, digital mental health platforms, and AI-guided therapeutic tools are often presented as convenient and accessible alternatives to traditional therapy. This has sparked big questions: Can AI outperform a therapist? Can it replace a therapist altogether? Should we be concerned about our privacy?
Research suggests these technologies can be helpful, but the wider body of evidence also reminds us to be cautious, especially when AI starts to replace human care rather than support it.
When does it work?
AI can be helpful for learning about mental health, tracking your symptoms, or practicing coping skills. Tools built around cognitive-behavioural techniques have even been shown to reduce anxiety and depression, especially when paired with real therapy.
Studies have also found that AI that is specifically designed to address mental health issues, are generally accurate in detecting, classifying and predicting risk for users. However, there is still a need for more diverse data sets to increase AI’s accuracy and therefore, ability to assist more uncommon and complex mental health presentations. Given the rapid improvement of AI, we doubt this is far off!
Can it replace a real therapist?
While AI systems can simulate supportive dialogue, ultimately, they do not possess lived experience, moral reasoning, or genuine emotional understanding. They cannot interpret nonverbal cues, relational ruptures, or complex trauma histories in the way trained clinicians can.
Psychotherapy is not simply the delivery of techniques or scripted responses, it is a relational process grounded in attunement, trust, rupture-and-repair, and shared meaning-making. These elements are difficult to replicate through pattern recognition alone.
What should we be concerned about?
Australasian Psychiatry calls AI in psychotherapy a “challenging frontier,” and for good reason. There are still unanswered questions about clinical safety, bias, and the limits of machine-generated empathy. Current AI systems can be prone to algorithmic bias, which means they might not work equally well for everyone if the training data isn’t diverse.
Privacy is another big concern, storing or processing sensitive personal disclosures always carries risk. AI also has limitations when it comes to detecting crises or managing risk, including assessing suicidality. Then there’s the potential for over-reliance: if people lean on AI all the time, there is concern that is encourages reassurance-seeking instead of real therapeutic growth.
The Verdict?
Decades of psychotherapy research consistently show that the therapeutic rapport, the collaborative bond between client and clinician is one of the strongest predictors of positive outcomes, at end of the day – It is not a substitute for human care.
AI may complement therapy. However, current evidence does not support equating AI-based tools with the depth, safety, and relational complexity of human-led psychological treatment.
For those seeking meaningful, personalised support, human connection remains central.
References:
Frontiers in Psychiatry. (2024). Artificial intelligence in psychotherapy: A systematic review of current applications and outcomes. Retrieved from https://pmc.ncbi.nlm.nih.gov/articles/PMC12359021/
Australasian Psychiatry. (2025). Artificial intelligence (AI) in psychotherapy: A challenging frontier. https://doi.org/10.1177/10398562251346075
Journal of Technology in Human Services. (2025). The Therapist in the Machine: Confronting AI’s Challenge to Clinical Social Work. https://doi.org/10.1080/15228835.2025.2500827
Cruz-Gonzalez, P., He, A. W.-J., Lam, E. P. P., Ng, I. M. C., Li, M. W., Hou, R., Chan, J. N.-M., Sahni, Y., Guasch, N. V., Miller, T., Lau, B. W.-M., & Sánchez Vidaña, D. I. (2025). Artificial intelligence in mental health care: A systematic review of diagnosis, monitoring, and intervention applications. Psychological Medicine, 55, Article e18. https://doi.org/10.1017/S0033291724003295
