Summary: More than half of U.S. teens now regularly use AI companion chatbots. A new Drexel University study warns these digital relationships can become a form of behavioral addiction, with emotional dependence that disrupts daily life.
Researchers analyzed more than 300 Reddit posts written by users who identified as 13 to 17 years old. What often begins as casual entertainment or emotional support can, the study found, evolve into strong attachments that resemble substance or behavioral addiction. The team proposes a design framework aimed at helping developers reduce harmful anthropomorphism and protect young users.
Key Facts
- Emotional support becomes a trap: About 25% of teens in the sample used chatbots for mental health advice or to cope with loneliness, so ending the interaction felt like losing a real relationship.
- The relationship illusion: Unlike passive media, AI chatbots respond emotionally and adapt to users, which encourages anthropomorphism—treating the system as if it were human—and creates bonds that are difficult to break.
- Real-life consequences: Teens reported sleep loss, falling grades and strained family or friend relationships linked to excessive chatbot use.
- Design “off-ramps” recommended: The research team urges features such as usage tracking, emotional check-ins, and personalized limits to help users disengage when necessary.
- Memory and multimodal interaction: Chatbots that remember past conversations or use voice and images are especially likely to foster dependence compared with earlier technologies.
Source: Drexel University
Context: Companion chatbots powered by large language models—services like Character.AI, Replika and Kindroid—are marketed as sources of conversation and emotional support. The Drexel study, however, shows that these interactions can grow into problematic attachments for some adolescent users, with consequences that reach beyond the screen.
Presented at the Association for Computing Machinery’s conference on Human Factors in Computing (ACM CHI), the study examined 318 Reddit posts from teens describing their experiences with Character.AI and similar systems. Many authors reported that what began as helpful or enjoyable interaction gradually became something they could not easily stop, even when it harmed their offline lives.
The researchers mapped these accounts onto established components of behavioral addiction and found evidence of all six commonly cited elements: conflict, salience, withdrawal, tolerance, relapse and mood modification.
- Conflict — feeling torn between continuing to interact with the chatbot and recognizing the harm caused by overuse.
- Salience — developing a central emotional attachment to the chatbot that replaces or displaces human relationships.
- Withdrawal — experiencing sadness, anxiety or a sense of incompleteness when unable to communicate with the bot.
- Tolerance — needing more frequent or longer interactions to achieve the same sense of comfort.
- Relapse — attempting to stop or cut back, only to resume heavy use later.
- Mood modification — relying on the chatbot to manage stress, loneliness or negative feelings.
Lead and contributing researchers noted that the interactive and responsive nature of chatbots makes them feel like relationships rather than mere tools. That emotional quality can make stepping away more psychologically difficult than abandoning other screen-based habits.
Unlike video games or social feeds, which primarily offer challenges or content consumption, chatbots create a feedback loop: they respond to personal disclosures, recall past conversations and adapt to user preferences. This persistent personalization—combined with multimodal features such as voice and images—heightens the potential for attachment.
To address these risks, the Drexel team proposes a design framework that balances supportive behavior with safeguards against overreliance. Key recommendations include building features that encourage offline connections, offering gentle disengagement options, and providing mechanisms that help users monitor and manage their time with the chatbot.
Suggested safeguards include:
- Usage tracking and summaries so users can see how much time they spend interacting with a bot.
- Regular emotional check-ins that assess well-being and suggest real-world coping strategies.
- Personalized usage limits and nudges that respect user autonomy while discouraging excessive dependence.
- Design consultations with teens and mental health professionals to ensure features are age-appropriate and clinically informed.
“Designers have a responsibility to create systems that protect young users and promote healthy social habits,” the researchers wrote. They emphasize that chatbots should help users build confidence in offline relationships rather than replacing them or encouraging anthropomorphism that leads to attachment.
For future work, the team recommends expanding research to larger, more diverse user samples and to other platforms and chatbot services. Surveys, interviews and broader community studies could help clarify how different demographics experience and manage companion AI use.
Key Questions Answered:
A: The difference lies in the social feedback loop. Games are challenges to overcome; chatbots respond emotionally and can “remember” personal details. The brain interprets those interactions more like social relationships, so quitting a chatbot can feel like ending a friendship rather than just stopping a pastime.
A: In some cases, yes—about a quarter of teens in the study reported short-term comfort from chatbot interactions. But researchers found that this assistance can become a psychological crutch that slows development of real-world social skills and may increase isolation over time.
A: Implement design “off-ramps” that encourage real-life engagement and self-regulation: personalized limits, prompts for offline social activities, usage summaries, and input from mental health experts and adolescent users during development.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- Journal paper reviewed in full.
- Additional context added by staff.
About this AI and psychology research news
Author: Britt Faulstick
Source: Drexel University
Contact: Britt Faulstick – Drexel University
Image: The image is credited to Neuroscience News
Original Research: Findings presented at the ACM CHI Conference on Human Factors in Computing Systems