Peer-reviewed article
How US Youth Use AI Chatbots: Conversation Patterns From Naturalistic Keystroke Observations.
Maheux AJ, Boeldt D, Akre-Bhide S, Flannery J, Paige A et al. ·
Summary
Written by AI, not yet AICS's view · claude-haiku-4-5-20251001This study examined how 3,363 US youth aged 1-17 years used AI chatbots by analyzing their keystroke data from a parental monitoring app. Researchers identified 11 patterns of AI use, including tool use, role-play, emotional support, and violent content. Tool use was most common overall, but younger users engaged more with violent, sexual, and romantic role-play themes. The research reveals substantial variation in how youth interact with generative AI, with some concerning patterns suggesting potential risks to child development and safety.
Key points
- 62% of youth-app interactions involved using AI as a functional tool, but 15% involved violence, 15% involved sexual role-play, and 8% involved romantic role-play
- Violence and sexual or romantic role-play were relatively more common among younger users, while tool use increased with age
- Word counts were highest for violence, sexual role-play, and romantic role-play, suggesting greater engagement with potentially risky content
- Violence, sexual role-play, and romantic role-play frequently co-occurred within single user sessions, including instances suggestive of violent sexual role-play
- Usage patterns were highly variable, with most users accessing only 1-2 apps across a median of 4 days during the study period
Limitations
This study describes patterns of AI use but does not measure psychological outcomes or well-being impacts. The data comes from users of a commercial parental monitoring app, which may not represent all youth AI use and may skew toward families with specific safety concerns.
Abstract
Background Youth are increasingly engaging with generative AI (GenAI) chatbots; yet, little is known about the nature of their interactions. Objective This study used naturalistic behavioral observations of keystrokes typed into GenAI mobile apps to characterize the topics youth discuss with chatbots, describing themes by age and app, word count by theme (a proxy for quantity of engagement), and co-occurring themes. Methods Participants were 3363 US youth aged 1-17 years using the Aura parental monitoring app, a commercially available service that guardians install on children's mobile devices for digital safety. Participants included only those who opted into the overlay Aura keyboard (to collect keystroke data) and who engaged in some degree of GenAI use. Participants' keystrokes for distinct GenAI apps by day (ie, user-app-days) were coded using a large language model with human review. Results Usage patterns were highly variable, with most users accessing only 1 to 2 GenAI apps (range 1-15) across a median of 4 days (IQR 1-11, range 1-247; 0.4%-90% of days during the study period), sending a median of 6 (IQR 2-27, range 2-2865) messages per app-day. Results identified 11 mutually inclusive themes, including functional tool use; social practice; role-play, including general, romantic, and sexual; emotional support; friend-like interactions; and violence. Most user-app-days involved using AI as a tool (26,343/42,355, 62%), but notable proportions included violence (6460/42,355, 15%), sexual role-play (6393/42,355, 15%), romantic role-play (3538/42,355, 8%), general role-play (3707/42,355, 9%), and friend-like interaction (2624/42,355, 6%). Theme prevalence varied by age: tool use was more common among older adolescents, whereas violence and sexual or romantic role-play were relatively more common among younger users. Themes also varied by app, with general-purpose apps most often used as tools and companionship or role-play apps showing higher usage of sexual, romantic, general role-play, and violent content; however, ChatGPT appeared across all themes due to high overall use. Word counts were highest for user-app-days involving violence, followed by sexual role-play, romantic role-play, and emotional support, suggesting greater engagement when using GenAI in these ways. Co-occurrence analyses showed that tool use often appeared alone, whereas violence, sexual role-play, and romantic role-play frequently co-occurred within a user-app-day, including instances suggestive of violent sexual role-play. Conclusions Findings suggest that youth engagement with GenAI varies substantially across users. Tool use is most common, requiring more nuanced investigation into the precise ways youth use GenAI to potentially support or thwart their development. GenAI engagement also may involve developmentally salient and potentially risky themes of sexuality, romance, and violence, especially among younger users. More research is needed that pairs GenAI engagement with psychological measures of well-being to examine the impacts of AI interactions and promote safety-focused design features.