How Youth AI Chatbot Conversations Reveal Hidden Risks

A major new study on youth AI chatbot conversations shows that many children and teenagers type far more than homework questions into generative AI apps. Researchers analysed keystroke data from thousands of US young people using parental monitoring software. The results highlight both practical tool use and frequent engagement with violence, sexual role-play, and romantic scenarios, especially among younger users.

What the Keystroke Study Actually Measured

Researchers examined data from 3,363 US youth aged 1–17 who used at least one generative AI app on a personal device. The sample came from users of the Aura parental monitoring app who had opted into an overlay keyboard. This captured the exact text children typed into AI chatbots over hundreds of days in 2025. The study, published in the Journal of Medical Internet Research, coded more than 42,000 user app days using a large language model with human review.

Most users accessed only one or two AI apps. They typically engaged on a median of four days and sent a median of six messages per app day. Usage was highly skewed, some young people typed thousands of messages. Tool use dominated overall, appearing in 62% of user app days. However, violence appeared in 15%, sexual role-play in 15%, general role-play in 9%, romantic role play in 8%, and friend-like interaction in 6%. Emotional support themes also appeared, though less often.

Word counts used as a proxy for depth of engagement were highest for violence, followed by sexual and romantic role play. Tool use often stood alone. Risky themes frequently co-occurred on the same day. The researchers noted high rates of explicit violent sexual role play during coding.

Age Patterns in Youth AI Chatbot Conversations

Tool use was more common among older adolescents. Among 16 and 17 year olds, roughly 80% of user app days involved using AI as a functional tool. Violence and sexual or romantic role play were relatively rare in this group, typically under 8%.

The pattern reversed for younger users. Among those under 12, only 47 to 53% of user app days were pure tool use. Violence appeared in 15 to 24% of their sessions, sexual role-play in 11 to 19%, and romantic role-play in 6 to 14%. These differences matter because younger children have less developed judgment and fewer real world experiences to compare against AI generated responses.

General purpose apps such as ChatGPT were most often used as tools. Companion focused apps marketed for role play showed higher rates of sexual, romantic, and violent content. ChatGPT itself appeared across every theme simply because of its overall popularity.

Why These Patterns Matter for Development

The sample came from families already paying for parental monitoring. This means the findings likely underestimate how common risky interactions are among young people without such oversight. Even in this group, substantial numbers of children explored violence and sexual themes with AI. However, we know that parents who install monitoring tools are more likely to discuss online activity with their kids.

AI chatbots differ from traditional media. They respond, adapt, and sustain conversations. So a child can escalate a violent or sexual scenario in real time and receive detailed, personalised replies. Social cognitive theory suggests that interactive modelling can shape expectations more strongly than passive viewing. Research on violent video games has linked interactivity to stronger aggressive cognitions. AI companions may intensify this effect because the interaction feels personal.

Sexual and romantic role-play with AI also carries risks. Age appropriate exploration is normal for teenagers. However, AI systems can produce explicit content without the social feedback or boundaries that real relationships provide. Younger users exploring these themes may form distorted ideas about consent, relationships, or body image before they have the maturity to evaluate them.

Practical Steps for Parents

Start with open conversations rather than only technical controls. Ask what AI apps your child uses and what they talk about. Many young people treat chatbots as tools for homework or brainstorming. Others use them for companionship. Both patterns appeared in the data.

Review device settings and app permissions. Most modern devices allow parents to limit or block specific apps. Companion style AI apps can be removed or restricted more tightly than general tools. Combine these with existing parental controls and safety settings that already cover screen time and content filters.

Agree clear rules about AI use. For younger children, keep AI interactions visible and supervised. For older teenagers, discuss how AI responses can be inaccurate, sycophantic, or inappropriate. Encourage them to bring surprising or uncomfortable AI replies to a trusted adult. Our Digital wellbeing guidance that covers device free times and open dialogue applies equally to AI chats.

Schools can also help. Teachers who assign AI assisted work should model responsible use and discuss limitations. Families who want structured support can explore CODA school presentations that address emerging digital risks, including AI.

The study authors call for more research that links actual AI conversation content with measures of well being. Until that evidence exists, the safest approach is active parental involvement combined with technical controls. Youth AI chatbot conversations are already widespread. Understanding what children type, and how AI replies, gives families a clearer starting point for guidance.

References

• Maheux AJ, Boeldt D, Akre-Bhide S, et al. How US Youth Use AI Chatbots: Conversation Patterns From Naturalistic Keystroke Observations. J Med Internet Res. 2026;28:e95819. https://www.jmir.org/2026/1/e95819

Further Guidance from CODA

• Google Parental Controls & Safety Settings – practical steps to manage apps and content on family devices.
• Digital Wellbeing for Families – guidance on screen routines, open conversations, and balanced technology use.
• Schools resources and blog – materials that support school-based discussions of online safety, including AI.

About Children of the Digital Age

Children of the Digital Age provides online safety education and training for schools, parents and organisations. Our guides support practical conversations about online behaviour and the choices people make. However, education and app settings do not replace safeguarding or professional support.

Send a message

Contact Children of the Digital Age to enquire about online safety education, parent sessions or training for your school or organisation.