Talking to Machines About Feelings: How AI Chatbots Shape Loneliness, Help-Seeking, and Mental Health

Engineering notes

Engineering notes will be added by the aipentium editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为大语言模型、生成式AI、ChatGPT相关技术、计算机视觉、深度学习等高价值论文补充中文说明。

Original abstract

Artificial intelligence chatbots such as Replika, ChatGPT, and Woebot are now a regular part of many people's emotional lives. Millions of people talk to these systems when they feel sad, anxious, or alone, and some describe their chatbot as a friend or even a partner. This raises an important psychological question: does talking to an AI chatbot actually help people feel less lonely and more willing to seek real support, or does it quietly pull them away from human connection and professional care? This article reviews the available peer-reviewed evidence on AI chatbots as sources of emotional support, focusing on three linked outcomes: loneliness, help-seeking intentions, and broader mental health. It draws on major psychological theories, including the evolutionary theory of loneliness, attachment theory, Para social relationship theory, and the long-running debate between the social compensation and rich-get-richer hypotheses of technology use. It reviews controlled trials of therapeutic chatbots such as Woebot, large observational and experimental studies of companion chatbots such as Replika and ChatGPT, and survey research on stigma and help-seeking. The evidence is mixed rather than one-sided. Structured, therapy-style chatbots show modest but real short-term benefits for mood and anxiety. Companion chatbots can ease loneliness for some users in the short term, particularly those who are already isolated, but heavier and more emotionally intense use is repeatedly linked to greater dependence and, in some studies, more loneliness over time. Chatbots also appear to lower the stigma barrier that keeps some people from seeking help, yet there is little evidence that chatbot use reliably increases contact with human professionals when it is needed most. The article ends by identifying gaps in current research, particularly the lack of long-term studies, the near-absence of research from South Asia, and the reliance on self-report data, and proposes a research direction that could help close these gaps. Understanding this issue matters because AI chatbots are already woven into how young people cope with distress, and psychology has a responsibility to understand this change rather than simply react to it.

Links and sources

Comments