Researchers at Stanford University, Carnegie Mellon University and other institutions published a study examining the relationship between long-term use of the AI character-chat app Character.AI and users’ well-being. They reported that users’ well-being tended to be lower as their engagement with AI continued and deepened.

The team collected data from 1,182 users between September 2024 and August 2025, then followed up with 439 of them about 12 months later. Researchers measured how frequently users engaged with AI and how dependent they felt, whether they used it for companionship by treating it as a friend or partner, and how much they disclosed personal concerns and feelings. They also asked how much time each day users spent in person with family, friends, colleagues and others.

Users who had stronger engagement with AI at the start of the study were more likely to remain strongly dependent on it and disclose more personal information a year later. Sustained, deep engagement with AI consistently predicted lower well-being. In analyzing how this relationship might work, the team found that stronger AI engagement tended to coincide with less time spent socializing in person.

The researchers suggested that social displacement—when time with people is replaced by another activity—may contribute to lower well-being by reducing time spent with family and friends and weakening real-world relationships. The findings address a debate over whether AI character chat can ease loneliness or deepen real-world isolation: the study points to AI immersion replacing real-world social relationships as a concern.

The researchers said this pattern is not limited to people who started using AI character chat to find a conversation partner. General-purpose chatbots such as ChatGPT may also become confidants after people begin using them for research or work. The study focused on Character.AI users; 439 of the original 1,182 responded to the follow-up. It examined engagement, companionship use, self-disclosure and time spent socializing in person in relation to well-being about one year later. The team also reported that relationships with AI tended to continue and intensify rather than end after brief use, and suggested that reduced in-person socializing may help explain the link between sustained AI engagement and lower well-being. Companionship use and disclosure of personal concerns or feelings were measured separately. The researchers said similar changes may occur with general-purpose chatbots.

Researchers at Stanford University, Carnegie Mellon University and other institutions published a study examining the relationship between long-term use of the AI character-chat app Character.AI and users’ well-being. They reported that users’ well-being tended to be lower as their engagement with AI continued and deepened.

The team collected data from 1,182 users between September 2024 and August 2025, then followed up with 439 of them about 12 months later. Researchers measured how frequently users engaged with AI and how dependent they felt, whether they used it for companionship by treating it as a friend or partner, and how much they disclosed personal concerns and feelings. They also asked how much time each day users spent in person with family, friends, colleagues and others.

Users who had stronger engagement with AI at the start of the study were more likely to remain strongly dependent on it and disclose more personal information a year later. Sustained, deep engagement with AI consistently predicted lower well-being. In analyzing how this relationship might work, the team found that stronger AI engagement tended to coincide with less time spent socializing in person.

The researchers suggested that social displacement—when time with people is replaced by another activity—may contribute to lower well-being by reducing time spent with family and friends and weakening real-world relationships. The findings address a debate over whether AI character chat can ease loneliness or deepen real-world isolation: the study points to AI immersion replacing real-world social relationships as a concern.

The researchers said this pattern is not limited to people who started using AI character chat to find a conversation partner. General-purpose chatbots such as ChatGPT may also become confidants after people begin using them for research or work. The study focused on Character.AI users; 439 of the original 1,182 responded to the follow-up. It examined engagement, companionship use, self-disclosure and time spent socializing in person in relation to well-being about one year later. The team also reported that relationships with AI tended to continue and intensify rather than end after brief use, and suggested that reduced in-person socializing may help explain the link between sustained AI engagement and lower well-being. Companionship use and disclosure of personal concerns or feelings were measured separately. The researchers said similar changes may occur with general-purpose chatbots.