As artificial intelligence (AI) continues to permeate nearly every facet of human life, its encroachment into the domain of mental health support has sparked intense interest and debate. A pioneering study conducted by researchers at Drexel University offers a granular examination of how individuals engage with AI chatbots for emotional and psychological aid, revealing complex user attitudes and nuanced patterns of reliance. Unlike a wholesale substitution of human therapy, the investigation reveals that most users treat AI as an adjunct resource—an auxiliary to conventional therapeutic interventions rather than a replacement.
This comprehensive research draws upon an unprecedented analysis of over four million posts extracted from 47 mental health-related subreddit communities on the popular social platform Reddit. Through advanced AI-driven natural language processing techniques, the researchers isolated a focused sample set of just over 5,000 posts where individuals openly discussed their interactions with AI programs in a mental health context. This data-rich approach enabled the team to employ sophisticated sociological and psychological frameworks to interrogate the dynamics of AI use, therapeutic bonding, and technology acceptance.
The study emerges at a pivotal moment. Surveys, such as a 2025 American Psychological Association poll, report that approximately 50% of U.S. adults aged 18 to 80 have employed large language model-based AI tools for mental health purposes within the preceding year. Parallel findings from notable academic institutions like Brown University indicate that as many as one in eight young adults are actively leveraging AI programs for psychological advice. These statistics underscore a rapidly evolving digital landscape within which mental health support is increasingly mediated through algorithmic agents.
Leading the investigation, Assistant Professor Shadi Rezapour from Drexel’s College of Engineering and Computing emphasizes a critical caveat: these AI systems, while highly sophisticated, have not been specifically engineered or clinically validated to function as mental health treatments. “Our goal was to parse out not just how people are using these chatbots, but to critically understand the perceived benefits alongside emerging risks,” Rezapour explains. Her lab specializes in analyzing online narratives and is at the forefront of developing socially aware AI systems designed to serve vulnerable populations.
Through their rigorous textual analysis, the researchers unearthed a consistent theme: users predominantly sought emotional reassurance, empathetic engagement, anxiety management, and coping strategies when interacting with AI chatbots. The utility of such tools also extended into pragmatic arenas, as respondents frequently reported using AI assistance for matters such as organizational support and managing symptoms related to neurodivergent conditions like ADHD and autism spectrum disorders. This practical dimension highlights the multifaceted role AI currently occupies in mental health ecosystems.
Notably, the study surfaced a pronounced ambivalence among users toward these AI tools. More than half of the analyzed communications explicitly referenced the inherent risks and limitations of relying on AI for therapeutic support. Many posts conveyed anxiety over prospects of emotional dependency or addiction to the technology, revealing a nuanced balance between appreciating the support AI offers and fearing its possible psychological pitfalls.
This ambivalence crystallizes in what the authors term a “bond paradox.” Positive experiences largely corresponded to interactions where AI assistance was goal-oriented—such as facilitating cognitive reflection, offering coping techniques, or providing organizational help. Conversely, when users sought companionship or engaged in repeated reassurance-seeking behaviors, an intensified emotional connection with AI often correlated with adverse outcomes. These included worsening symptoms, heightened feelings of shame or guilt, and problematic dependencies on the chatbot interactions.
The researchers assert that these findings carry profound design implications. AI systems developed for mental health should prioritize establishing clear operational boundaries to mitigate risks associated with unstructured emotional attachment. “It is not sufficient for AI to simply simulate warmth or human likeness,” elaborates Elham Aghakhani, the study’s lead author and doctoral candidate. “Our research suggests the necessity of embedding safeguards particularly in contexts where companionship or excessive reassurance-seeking occurs, to prevent exacerbating symptoms or fostering unhealthy reliance.”
Importantly, the majority of users framed AI as complementary rather than competitive to human therapists. The study highlights that AI functions primarily as a stopgap in moments where human care is absent, inaccessible, or insufficient. This nuanced positioning challenges alarmist narratives that position AI as threatening traditional mental health professions, instead underscoring a hybridized model of care that integrates technological and human support systems.
Underpinning these insights is the researchers’ endorsement of evidence-based frameworks for AI mental health tools. AI’s increasing ubiquity demands that these tools be grounded in scientifically validated psychological principles to maximize benefit and minimize harm. Users must be educated about both the potentials and limitations inherent to AI-mediated support to foster informed engagement and mitigate misplaced expectations.
As the landscape of mental health services evolves, this study from Drexel University offers a timely and critical perspective. It emphasizes a pragmatic optimism tempered by caution, advocating for AI innovations that enhance rather than replace human therapeutic bonds. With millions turning to AI for emotional solace and coping strategies, ensuring these technologies operate within well-defined ethical and clinical boundaries is imperative to safeguard a vulnerable user base.
This body of work signifies a milestone in natural language processing research intersecting with mental health and social understanding. By deploying data-driven sociological assessments on vast digital diaries of user interactions, the study provides one of the most comprehensive portraits to date of how AI chatbots function within the psychological support landscape. It charts a path forward for developers, clinicians, and policymakers tasked with sculpting a digitally integrated mental health future that is both effective and responsible.
In sum, AI’s role in mental health support is complex and evolving. It exists not as a panacea but as a tool layered within human therapeutic ecosystems. By illuminating the nuanced realities of user interaction, this research equips stakeholders with critical knowledge to guide the ethical advancement and deployment of AI in sensitive and impactful mental health domains.
Subject of Research: Usage patterns and perceptions of AI chatbots in mental health contexts, focusing on Reddit user narratives.
Article Title: Like a Therapist, But Not: Reddit Narratives of AI in Mental Health Contexts
News Publication Date: 28-Jan-2026
Web References:
- https://arxiv.org/abs/2601.20747
- https://psycnet.apa.org/doiLanding?doi=10.1037%2Fpri0000292
- https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2841067
- https://arxiv.org/abs/2504.18412
Keywords: Artificial intelligence, mental health, clinical psychology, sociological analysis, natural language processing, therapy, large language models, emotional support, digital health, AI ethics

