27% of U.S. Adults Use AI Chatbots for Emotional Support

Americans are forming emotional bonds with AI

More than one in four American adults now use AI chatbots for personal and emotional conversations, marking a significant shift in how people relate to artificial intelligence tools originally marketed for productivity.

A survey conducted by Elon University researchers in collaboration with The Washington Post found that 27% of U.S. adults turn to chatbots like ChatGPT or Google's Gemini for entertainment, relationship advice, emotional discussions, or companionship when feeling alone. Among adults under 50, that figure jumps to nearly 40%—more than double the rate of older users.

The research, based on interviews with over 4,000 U.S. adults conducted by YouGov in May, reveals usage patterns that cut across demographic lines. Personal AI use was roughly similar regardless of race, gender, or education level.

Why it matters

This trend represents a fundamental change in human-technology relationships at a time when Americans report having fewer close friends than previous generations. Unlike social media, which reshaped public social interactions, AI chatbots are becoming private confidants—raising new questions about data privacy, mental health impacts, and the role of technology companies as custodians of deeply personal information. The speed of adoption is unprecedented: within three years of ChatGPT's launch, emotional engagement with AI has become normalized for millions of users.

What users are telling AI

Nearly 40% of people who use chatbots for personal queries said they sometimes talk to AI to feel less alone. Half reported that chatbot conversations make them feel better when stressed or upset. Almost a third consider their most-used chatbot to be a friend.

The intimacy of these relationships is striking: nearly 40% of regular users said they've told AI things they wouldn't share with other people. When asked to describe their relationships with AI, some respondents used words like "friend," "companion," or "a friend I can trust."

Alex Suarez, a 74-year-old retired psychologist from Washington state, exemplifies this shift. Initially skeptical of AI, she now discusses art and literature with chatbots and uses them to analyze her dreams. "It creates some feelings of being grateful to a machine," she told researchers.

Companies see the same patterns

AI providers' internal data confirms the survey findings. OpenAI reported that nonwork messages to ChatGPT increased from 50% of usage in August 2024 to 70% by June 2025. Messages categorized as "self-expression"—including casual chats, relationship advice, and personal reflection—more than doubled during that period to 9% of all interactions.

Anthropic found that about 6% of Claude users seek "personal guidance," primarily for health and career issues, with 12% of that group looking for relationship advice.

Privacy and safety concerns emerge

The emotional reliance on AI chatbots has created new risks. AI companies typically retain conversation logs to train future models, and law enforcement agencies have begun requesting chat records for investigations. Those conversations can also surface in civil litigation.

More troubling are cases where intensive chatbot use preceded tragic outcomes. Ongoing lawsuits allege that OpenAI and Google failed to adequately safeguard users who developed delusional thinking or harmed themselves after extended AI conversations. Both companies have denied responsibility and released product updates intended to prevent inappropriate responses.

The Elon University survey found mixed user experiences: while nearly 60% said chatbots were helpful for personal decisions, over one-third reported that bots agreed with them too much, and 15% said the interactions made them feel less in touch with reality.

Jessica Dai, a UC Berkeley PhD student who analyzed three years of AI discussions on Reddit, noted that emotional engagement posts increased steadily over time. "This is the first three years that anyone in the world has had access to technology of this form," she said. "You can see the trends of this becoming a normalized part of people's lives."

The findings were first reported by The Washington Post, with survey methodology showing a margin of error of 3.7 percentage points for the representative subsample of 1,000 regular AI users.