AI 'Companions' Worsen Loneliness, Stanford Study Finds

A recent study from Stanford University brings some intriguing insights into the emotional effects of using AI chatbots for companionship. Published on August 4 in Nature Human Behaviour, the research suggests that individuals seeking emotional support from these chatbots often end up feeling lonelier, especially if they already have limited social connections in real life.

The study was led by Yutong Zhang and Dora Zhao, who both worked under computer science professor Diyi Yang. They surveyed 1,131 users of Character.AI, a platform where millions engage with chatbot avatars. Interestingly, 244 participants shared full transcripts of their conversations, which the researchers analyzed with advanced tools to assess psychological well-being using the Comprehensive Inventory of Thriving which is a recognized measurement for such assessments.

While many participants may not have stated that companionship was their primary motive for engaging with chatbots, the results tell a different story. Only about 12% identified companionship as their main reason for using AI, yet over 50% described their AI chat partners with terms like “friend” or “romantic partner.” Moreover, a striking 80% of the chat transcripts focused on seeking emotional and social support.

At first glance, the more engaged users seemed to have better well-being. But this wasn’t the whole picture. Those with limited offline social circles—meaning they had few friends or family to confide in—often reported feeling worse when they turned to AI for emotional support. The research highlighted that this negative trend emerged particularly when users sought companionship and shared sensitive personal experiences like emotional pain or substance use.

What’s fascinating is that disclosing personal details to AI doesn’t seem to foster deeper connections as it typically does in human relationships. In human interactions, sharing personal struggles usually strengthens bonds and enhances emotional well-being. Zhang and Zhao point to a few reasons for this disconnect. Unlike human friends, chatbots can’t share personal experiences back, they may not fully grasp the emotional weight of conversations, and they are designed to keep users engaged, sometimes regardless of the conversation depth.

These findings challenge the earlier belief that AI chatbots might combat loneliness—a problem that Mindplex has highlighted as growing, particularly among isolated seniors, those battling depression, and young adults, with 40% of the latter group reporting feelings of loneliness. Previous discussions noted the appeal of platforms like Replika, which market themselves as personalized companions, while acknowledging that modern society’s eroding community ties have left a gap that AI is trying to fill.

Zhao describes the experience of AI companionship as a "social snack." This term captures the idea that while engaging with a chatbot might feel good in the short term, it lacks the ingredients necessary for long-term emotional health. The researchers caution that users seeking out these digital companions might end up further isolating themselves in the real world which is a truly troubling cycle.

The study emphasizes the need for public awareness regarding the potential risks of relying on AI for companionship. It also suggests that these platforms might want to consider implementing features like usage limits or directing users to human support when discussions indicate emotional distress.

It is worth noting that this study only captures a moment in time and doesn’t provide data on the long-term consequences of using these AI companions. There was a separate study coming from Harvard Business School in 2025 indicating that AI companions could alleviate immediate feelings of loneliness for a week, which contrasts sharply with the findings from Stanford. This presents an unresolved tension that suggests more research is necessary in this evolving field.