Young teens who preferred a warm chatbot reported more anxiety, a study of 284 finds

Parents leaned the other way: 53.4% of 264 parents chose the relational style for their child and 28.8% chose the transparent style. Adolescents who preferred the relational chatbot reported lower family and peer relationship quality and higher stress and anxiety than peers who preferred the transparent one or both.

The authors say the findings point to conversational style as a design choice for youth AI safety, since teens with weaker offline support may be drawn to chatbots that sound caring. The data do not measure actual emotional reliance, and the design cannot show whether distress drives the preference or the reverse.

Families were recruited through CloudResearch Connect and lived in the United States. The adolescents averaged 12.98 years old, half were boys, and 90% had used ChatGPT, with 56% using it a few times a week or daily. 94% had used some AI chatbot and 58% had used an AI companion app.

The two transcripts were generated with GPT-4.5 on April 8, 2025, and were matched in length at 175 words (relational) and 179 words (transparent). The order was randomized, each participant rated both chatbots, then chose which they would prefer to use. The study was preregistered on OSF in July 2025.

Family relationship quality, stress and anxiety were measured with PROMIS scales, with parents reporting family and peer relationships and depressive symptoms. Adding AI-use frequency and mental health diagnoses as covariates one at a time did not remove the links with family relationships, peer relationships, anxiety or social isolation, while the stress link weakened to trend level.

The authors argue the transparent chatbot offered comparable guidance without the same humanlike pull, and suggest repeated reminders that the chatbot is not a person, distress detection and escalation to caregivers, and AI literacy teaching. They also note that relational chatbots might help isolated teens when other support is limited.