Nearly half of American adults have turned to AI chatbots for mental health support in the past year, according to a 2025 American Psychological Association survey—yet a rigorous new study from Drexel University in Philadelphia reveals they're approaching these tools with clear-eyed caution, viewing them as companions to therapy rather than replacements for human care.
Researchers led by Shadi Rezapour, an assistant professor in Drexel's College of Engineering and Computing, analyzed more than four million Reddit posts across 47 mental health–focused subreddits to understand how people actually use AI for emotional support and what they think about the experience. The team winnowed this vast dataset down to 5,126 carefully examined posts, applying two established sociological frameworks—one traditionally used to measure the therapeutic bond between therapist and client, the other to assess how people adopt new technologies. What emerged is a portrait of pragmatic users who see genuine value in AI support, but with their eyes wide open to its limits.
The appeal is real and varied. People turn to these chatbots seeking emotional support, empathy, and reassurance for anxiety management. Many rely on them for coping strategies and companionship. Others use AI for practical, task-oriented help: organization, planning, and support managing challenges related to attention deficit/hyperactivity disorder or autism. In these moments—when a human therapist is unavailable, inaccessible, or insufficient—AI fills a genuine gap.
But users are thinking critically about what they're doing. In 51% of the posts analyzed, people explicitly acknowledged the risks and limitations of using AI in a therapeutic context. Many expressed worries about emotional dependence or addiction to the programs. Crucially, "few people are using AI as a replacement for therapy," Rezapour said. "More often, they described using it alongside therapy or during moments when human care is unavailable, inaccessible or insufficient."
The research uncovered what the team calls a "bond paradox"—a tension that matters for anyone developing AI mental health tools. Users reported the most positive experiences when AI helped with specific, concrete tasks: reflection, coping strategies, or organization. But when users developed a strong emotional bond with AI without clear goals or useful tasks attached, that relationship became more likely to carry risks, particularly when people sought repeated reassurance or treated the AI as a primary source of companionship.
This distinction matters because it suggests that AI mental health support works best when bounded and purpose-driven, not when it edges toward replacing human connection. The findings arrive at a moment of genuine concern: Brown University research has shown that as many as 1 in 8 young adults are turning to AI for mental health advice, even though these programs were never designed or clinically validated for therapeutic use.
Rezapour's lab, which specializes in analyzing online narratives and building socially aware AI for vulnerable populations, has built a reputation for studying how people's relationships with AI evolve over time. These latest findings will be presented at the 2026 Annual Meeting of the Association for Computational Linguistics in San Diego in July—a venue where technologists, linguists, and researchers can grapple with what responsible AI mental health support might look like. The message from real users is clear: AI can help, but only when its role is honest, its limits are acknowledged, and human care remains the foundation.