A small design lever for kids’ AI companions: 284 teens and parents compared two
A small design lever for kids’ AI companions: 284 teens and parents compared two
In a preregistered U.S. study, 11–15-year-olds and parents read matched responses to a school peer problem: one used first-person affection and promises of continued support; the other made the bot’s nonhuman limits clear and gave similarly constructive advice. Order was randomized. Among the teens included in the preference analysis, 66.8% chose the relational reply; trust and emotional-closeness ratings rose with it, whereas helpfulness did not differ significantly. Children with poorer existing family and peer relationships were more likely to choose relational language, an association that cannot show the bot harmed their relationships.
The evidence map is more mixed than 'bots replace friends': a U.S. survey found most teen users still spend more time with human friends; original tests also found scripted adolescent crises often went without a specific human referral on companion products. Adult companion experiments find immediate loneliness relief, while a two-week college comparison favored messaging a human peer for reduced loneliness. None measures long-run teen developmental harm. The missing test is live, consent-based repeated use with age and handoff audits plus actual peer contact and validated outcomes.