Introduction: Therapy in an Age of Interfaces
Across many countries, mental health services are struggling to keep pace with demand. Rising rates of anxiety and depressive disorders have coincided with limited clinical capacity, long waiting lists, and financial barriers that make therapy difficult to access for many people (World Health Organization, 2023). In practice, this often means that individuals who are ready to seek support may still face delays of weeks or months before receiving professional care.
At the same time, a new layer of digital tools has begun to occupy part of the mental health landscape. Conversational agents that offer cognitive behavioural prompts, mood-tracking applications, and automated wellbeing check-ins are increasingly common. What was once almost exclusively an interpersonal process is gradually becoming something that can also occur through an interface.
This development invites a broader reflection. When psychological support is delivered by a digital system rather than a person, it raises questions not only about effectiveness but also about how people experience empathy, trust, and disclosure in therapeutic contexts. The purpose of this discussion is not to argue that AI therapy should replace traditional therapy, nor to dismiss it entirely. Instead, the aim is to consider what psychological research suggests about its potential benefits and its limitations.
Why AI Therapy Appeals to So Many People
Research on help-seeking behaviour suggests that the decision to seek support is influenced by more than the severity of distress. Perceived stigma, emotional safety, and the desire for autonomy all shape whether individuals feel comfortable discussing personal difficulties (Rickwood et al., 2005). In many cases, people recognise that they are struggling but still hesitate to initiate a conversation with another person.
Digital environments appear to reduce some of these barriers.
One explanation comes from studies of computer-mediated communication. Individuals often disclose personal information more readily when interacting through anonymous digital channels (Joinson, 2001). Without the presence of facial expressions or perceived judgement, the act of describing distress may feel less socially risky.
AI systems also offer a level of control that traditional therapy does not always provide. Users can decide when to engage, how much to share, and when to stop the interaction. For individuals who find emotional vulnerability uncomfortable, this degree of autonomy may feel psychologically safer than entering a therapeutic relationship that involves ongoing interpersonal exposure.
Accessibility is another factor. Digital tools are available immediately. In moments of distress, the ability to access support at any time of day can be meaningful, particularly for individuals who are currently on waiting lists for professional care. For some people, these systems may function as a temporary coping resource or even as a first step toward seeking more formal support.
Can Empathy Be Simulated?
One of the most discussed features of AI therapy platforms is their ability to simulate empathic responses. Advances in natural language processing allow conversational systems to identify emotional language and respond with reflective or validating statements.
From a psychological standpoint, emotional validation can have a measurable impact. Carl Rogers (1957) described empathy as one of the central conditions that facilitate therapeutic change. Feeling heard and understood can reduce immediate distress and help individuals organise their experiences more clearly.
However, empathy in psychotherapy is not limited to language alone. It also involves attunement, contextual understanding, and an ongoing relational commitment between therapist and client. Human therapists respond not only to what is said but also to tone, pauses, contradictions, and subtle emotional cues.
AI systems can generate empathic wording, but they do not experience concern or responsibility in the way a therapist does. This distinction raises an interesting psychological question. If a person feels understood by an AI system, does the subjective experience of empathy matter more than the actual presence of a human relationship?
In some contexts, simulated empathy may be sufficient. Structured exercises that focus on emotional regulation or cognitive reframing do not always require deep relational engagement. Yet when therapy involves complex trauma, relational wounds, or long-standing patterns of behaviour, the absence of a responsive human presence may become more noticeable.
When Support Becomes Substitution
A related issue concerns the role that AI tools ultimately play within mental health care. Are they primarily supplementary resources, or could they begin to replace professional support in some circumstances?
For individuals experiencing mild or moderate distress, structured prompts and coping exercises may provide meaningful assistance. However, complex psychological presentations often require detailed assessment and flexible therapeutic responses that adapt to the individual (Torous & Roberts, 2017).
Digital systems offer certain advantages. They are consistent, available around the clock, and unaffected by fatigue or emotional reactivity. At the same time, their responses remain limited by design. Detecting risk, responding appropriately to crisis situations, or challenging deeply entrenched beliefs requires nuanced judgement that may be difficult to encode into automated systems.
There is also a behavioural dimension to consider. If distress is repeatedly managed through immediate digital reassurance, individuals might become less inclined to engage in longer-term therapeutic work that requires sustained effort. On the other hand, some users may find that structured digital tools increase their confidence in discussing emotional issues, which could ultimately encourage them to seek professional therapy.
In other words, the outcome may depend less on the technology itself and more on how it is used.
Ethical Questions Behind the Interface
The expansion of AI-mediated mental health tools also introduces a number of ethical considerations.
Mental health conversations frequently involve deeply personal disclosures. When these interactions occur on digital platforms, they generate data that may be stored, analysed, or shared in ways that users do not fully anticipate (Floridi et al., 2018).
This raises several practical questions:
• Do users clearly understand how their information is processed and stored?
• Are these systems designed primarily to support wellbeing, or to maximise engagement with the platform?
• Who carries responsibility if advice generated by the system proves harmful or inadequate?
Another complication is regulatory classification. Many AI mental health tools are marketed as “wellness” products rather than healthcare services. As a result, they often operate outside the regulatory frameworks that govern clinical practice, despite influencing users’ psychological experiences.
At a broader level, the spread of AI therapy may also highlight existing inequalities in mental health provision. In regions with limited access to therapists, digital systems could provide valuable support. At the same time, this raises the possibility of a two-tier model in which some populations receive human-led therapy while others rely primarily on automated systems.
The Therapeutic Relationship: Can It Be Replicated?
Within psychotherapy research, the therapeutic alliance is widely regarded as one of the strongest predictors of treatment outcomes. The quality of the relationship between therapist and client often influences progress as much as the specific techniques being used.
This raises an interesting question in the context of AI therapy: can a sense of therapeutic alliance exist without a human therapist?
Some scholars argue that perceived understanding may be enough to generate a sense of alliance. If individuals feel that their experiences are recognised and validated, the psychological effect could still be meaningful.
Yet therapeutic relationships also involve moments of tension. Misunderstandings, disagreement, and emotional rupture often become important opportunities for reflection and growth. These interactions require flexibility and relational responsiveness, qualities that are difficult to replicate in systems designed to remain consistently supportive and non-confrontational.
There is also the matter of accountability. Human therapists operate within ethical frameworks and professional standards that guide their responsibility toward clients. AI systems, even when thoughtfully designed, function according to programmed rules rather than moral judgement.
Whether perceived empathy alone can sustain a therapeutic alliance remains an open question.
Where Does This Leave Us?
AI-mediated therapy is unlikely to disappear. It reflects broader changes in how services are delivered and how people interact with technology in everyday life. For some individuals, these tools may reduce barriers to seeking support and provide immediate coping strategies during periods of distress.
Generational differences may also influence how these systems are experienced. Younger individuals who grew up communicating through digital platforms may find interactions with AI relatively natural. In that context, speaking with a conversational system may not feel dramatically different from other forms of online communication.
From a psychological perspective, the more interesting issue may not be whether AI therapy is beneficial or problematic in isolation. Instead, it is worth considering how these technologies gradually reshape expectations about vulnerability, self-disclosure, and emotional support.
As AI becomes more integrated into mental health spaces, it may also encourage professionals to reflect on what aspects of therapy remain distinctly human. Presence, accountability, and the shared experience of emotional understanding are difficult to automate.
For now, the discussion remains open. The role of AI in mental health care will likely continue to evolve, and understanding both its potential and its limitations will remain important for researchers and practitioners alike.
References
Floridi, L., Cowls, J., Beltrametti, M., et al. (2018). AI4People—An Ethical Framework for a Good AI Society. Minds and Machines, 28(4), 689–707.
Joinson, A. N. (2001). Self-disclosure in computer-mediated communication. CyberPsychology & Behavior, 4(5), 587–598.
Rickwood, D., Deane, F. P., Wilson, C. J., & Ciarrochi, J. (2005). Young people’s help-seeking for mental health problems. Australian e-Journal for the Advancement of Mental Health, 4(3).
Rogers, C. R. (1957). The necessary and sufficient conditions of therapeutic personality change. Journal of Consulting Psychology, 21(2), 95–103.
Torous, J., & Roberts, L. W. (2017). Needed innovation in digital health and smartphone applications for mental health. JAMA Psychiatry, 74(5), 437–438.
World Health Organization. (2023). Mental Health at Work. WHO.
