Artificial intelligence has moved beyond its early role as a tool for productivity, automation, and research, entering the intimate realm of emotional life. Students may use the same chatbot to summarize a journal article or draft a difficult email that it does to talk through feelings of anxiety or depression. This convergence is significant. Mental health support is no longer imagined only in clinical offices, counselling centres, or peer conversations; it is now also mediated through interfaces that simulate responsiveness and care. The urgent question is: what happens when a tool that supports objective tasks, such as organizing information or generating ideas, is positioned as a source of psychological support? This paper will argue that AI chatbots should be approached as supplementary supports within digital mental health, not as substitutes for professional or relational care, because their accessibility and responsiveness are accompanied by unresolved concerns about clinical safety and privacy.
Brief Literature Review
Current research suggests AI chatbots appear capable of producing modest reductions in depression, anxiety, stress, and related forms of distress, but their effectiveness depends heavily on design, user engagement, population, and whether the system is retrieval-based or generative. Feng et al. (2025) conducted a systematic review and meta-analysis of randomized controlled trials examining AI chatbots among adolescents and young adults aged 15 to 39. Their review included 31 randomized controlled trials with 29,637 participants, making it one of the more substantial syntheses in this area. Overall, the authors found that AI chatbots produced small-to-moderate reductions in mental distress, with statistically significant improvements in depressive symptoms, anxiety, stress, psychosomatic symptoms, negative affect, and appearance-related distress. They also found modest improvements in life satisfaction and well-being, although effects on positive affect and self-efficacy were limited (Feng et al., 2025). Notably, they also found that effectiveness varied depending on the chatbot’s design, delivery format, use of reminders, control condition, and target population. User engagement was especially important where repetitive content and technical issues were identified as barriers to continued use. This matters because effective digital mental health tools must sustain trust, relevance, and user participation as their accessibility alone is not enough. The authors concluded that retrieval-based systems showed more consistent effects, while generative AI systems “showed promise,” but their overall effectiveness remained “inconclusive” (Feng et al., 2025). In other words, the strongest evidence does not yet support the unrestricted use of generative AI as a stand-alone mental health intervention.
A similar conclusion was derived in a study by Zhang et al. (2025) which focused more specifically on generative AI mental health chatbots. Their systematic review included 26 quantitative studies, while their meta-analysis examined 14 randomized controlled trials involving 6,314 participants. They found that generative AI chatbots had a statistically significant average effect in reducing negative mental health outcomes, including symptoms such as depression and anxiety. However, the effect size was modest, and the confidence interval was wide, suggesting that outcomes may differ substantially across studies and contexts (Zhang et al., 2025). The authors also noted that most interventions took place in non-WEIRD countries and that there remains a lack of evidence on certain populations, including young children and older adults.
AI chatbots, therefore, may provide meaningful low-intensity support, particularly for users who face barriers to traditional care or need immediate help organizing distress. However, the evidence does not justify treating them as substitutes for professional mental health services. As Zhang et al. (2025) state, the promise of generative AI chatbots cannot be separated from their risks. The research therefore points toward a supplementary role where AI may assist with symptom management and early support but it should remain connected to broader systems of professional human support and care.
The Risk of Substitution
The attraction of AI mental health tools is inseparable from gaps in existing systems. Many students encounter long waitlists, limited counselling availability, stigma around seeking help, or uncertainty about whether their distress is “serious enough” to justify support. AI appears to solve these access problems by being constantly available and bypassing the fear of judgement associated with human care. For example, a student might use AI to prepare for a counselling appointment, draft what they want to say to a friend, or identify patterns in their stress. In these cases, the technology supports movement toward human care or self-understanding. The risk emerges when convenience becomes substitution. A tool that begins as a bridge can become a barrier if it encourages users to remain within a closed loop of machine-generated reassurance. The World Health Organization has warned that generative AI tools are increasingly being used for emotional support despite being “neither designed nor tested for mental health” (World Health Organization, 2026).
The Problem of Simulated Empathy
One of the most complicated features of AI mental health support is that it can sound caring without being capable of care. Chatbots can mirror a user’s language, validate distress, and produce responses that resemble empathy but empathy in mental health is a relational and ethical practice grounded in responsibility. Stanford researchers have cautioned that AI therapy chatbots can “fall short of human care” and may risk “reinforcing stigma or offering dangerous responses” (Wells, 2025). Many AI systems are optimized to be helpful, affirming, and responsive to user prompts. In mental health contexts, however, affirmation is not always constructive. A person experiencing anxiety may require help challenging catastrophic interpretations; a person showing signs of severe distress may require timely escalation to qualified human support, not continued engagement with a system that cannot understand the nuanced emotional and clinical context of human suffering.
Privacy Concerns
Mental health disclosures are sensitive. When users share fears, trauma, loneliness, family conflict, or suicidal thoughts with a chatbot, they may falsely believe that the exchange is private. However, most AI tools do not operate under the same confidentiality obligations, professional standards, or regulatory frameworks as licensed providers. In this way, psychological vulnerability can easily become data vulnerability. The American Psychological Association (APA) has warned that generative AI chatbots and wellness apps lack sufficient evidence and regulation to ensure user safety (American Psychological Association, 2025). APA argues that the mental health crisis requires systemic solutions, “not just technological quick fixes” (American Psychological Association, 2025). This is to say that technology cannot be presented as an efficiency solution to problems that are inherently social, institutional, and clinical.
Discussion and Conclusion
AI may be appropriate for organizing thoughts, identifying coping strategies, preparing for appointments, or reflecting between sessions with a professional, but it cannot be treated as a therapist or replacement for qualified care. For this reason, AI systems used in mental health contexts should include clear disclaimers, strong privacy protections, crisis referral pathways, age-sensitive safeguards, and development input from clinicians and people with lived experience. WHO has emphasized that responsible AI in mental health requires bringing together “the voices of those most affected, clinical and research expertise, governance and regulatory frameworks, and data” (World Health Organization, 2026). The goal should not be to make machines sound more like therapists but that they should be to ensure that digital tools direct people toward appropriate care.
References
American Psychological Association. (2025). Artificial intelligence, wellness apps alone cannot solve mental health crisis. American Psychological Association. https://www.apa.org/news/press/releases/2025/11/ai-wellness-apps-mental-health
Feng, X., Tian, L., Ho, G. W. K., Yorke, J., & Hui, V. (2025). The effectiveness of AI chatbots in alleviating mental distress and promoting health behaviors among adolescents and young adults: Systematic review and meta-analysis. Journal of Medical Internet Research, 27, e79850. https://doi.org/10.2196/79850
Wells, S. (2025). New study warns of risks in AI mental health tools. Stanford Report. https://news.stanford.edu/stories/2025/06/ai-mental-health-care-tools-dangers-risks
World Health Organization. (2026). Towards responsible AI for mental health and well-being: Experts chart a way forward. World Health Organization.
Zhang, Q., Zhang, R., Xiong, Y., Sui, Y., Tong, C., & Lin, F.-H. (2025). Generative AI mental health chatbots as therapeutic tools: Systematic review and meta-analysis of their role in reducing mental health issues. Journal of Medical Internet Research, 27, e78238. https://doi.org/10.2196/78238
Edited by: Kealyn McDowell