AI Chatbot Companions Linked to Lower Wellbeing in New Study

AI companions may worsen isolation for vulnerable users

People turning to AI chatbots for emotional support may be experiencing lower psychological wellbeing, according to new research from Stanford University that examined how Americans interact with AI companion platforms.

The August 2026 study analyzed 1,131 U.S. adults who use Character.AI, an AI companion platform, and reviewed 464,687 messages from 237 participants. Researchers found that individuals with smaller offline social circles who primarily used chatbots for companionship reported higher levels of loneliness, lower life satisfaction, and a diminished sense of belonging.

"AI may not be the cause of people's problems, but it might not be making them happier either," says Yutong Zhang, a study author and incoming PhD student at Stanford.

The findings arrive as AI companion usage grows rapidly. A September 2026 Elon University survey found that 27% of U.S. internet users report having "significant emotional or social interactions" with large language models like ChatGPT and Claude.

Why it matters

As tech companies increasingly position AI as a solution to loneliness—Meta CEO Mark Zuckerberg suggested in 2025 that AI could help fill friendship gaps—this research provides early evidence that chatbot relationships may actually compound isolation for the most vulnerable users rather than alleviate it.

What AI companions cannot provide

New Jersey psychotherapist Samantha Hill, a licensed clinical social worker, argues that AI chatbots have fundamental limitations that prevent them from offering genuine relationships.

First, they cannot provide physical intimacy like a hug. Second, they lack reciprocal emotions—the shared emotional experience that defines human connection. "Like, if I'm feeling sad, you're feeling sad," Hill explains.

Most critically, AI companions cannot provide the productive friction necessary for personal growth. "Humans, we are very messy," Hill says. "We have differing opinions, values, beliefs. But what actually helps you to progress in life, in these relationships, is the conflict, right? It's the drama. It's the disagreement. It's the tension."

One of Hill's clients recently admitted to seeking therapeutic advice from a chatbot, highlighting what she sees as growing discomfort with human vulnerability. She attributes some of this trend to social anxiety and diminished social skills following the COVID-19 pandemic.

Relationship drift happens gradually

Zhang's research revealed an important pattern: many users didn't initially consider their interactions with Character.AI as interpersonal relationships. However, when researchers examined chat histories, it became clear that conversations had evolved into intimate territory over time.

"Even if people don't mean to be building intimate relationships with these bots, it's possible people are doing so," Zhang says.

Hill recommends that users recognize chatbots' limits as information tools while understanding "they're incapable of offering genuine interpersonal relationships." Zhang adds that users should remember any sensitive information shared with bots is also shared with the companies operating them.

These findings were first reported by CNBC, which detailed the Stanford research and expert perspectives on human-AI relationships.