Understanding Parents’ Desires in Moderating Children’s Interactions with GenAI Chatbots through LLM-Generated Probes

Paper Title

Understanding Parents’ Desires in Moderating Children’s Interactions with GenAI Chatbots through LLM-Generated Probes

Publication Info

• Topic area: Parental control design for Generative AI Chatbots.
• Keywords: GenAI Chatbots, parental controls, child safety, moderation, transparency, AI ethics, child-computer interaction, personalization, algorithmic parenting, developmental appropriateness.

Background and Problem

• Problem / challenge: Current parental control tools for GenAI Chatbots are coarse-grained and insufficient to address the unique challenges posed by open-ended, stochastic conversational systems. They fail to provide fine-grained moderation and transparency at the conversation level.
• Significance: The rapid adoption of GenAI Chatbots by children raises risks such as exposure to harmful content, undermining parental authority, and fostering overdependence. Effective parental controls are necessary to ensure safety while preserving children’s access to AI tools.
• Motivation and related work: Existing research has focused on parental controls for conventional technologies like smartphones and social media, but these approaches do not address the dynamic, individualized nature of GenAI Chatbot interactions. Recent lawsuits and incidents highlight the urgency of addressing these gaps.

Solution

• Proposed approach: A systematic investigation of parents’ concerns and desires for moderating children’s interactions with GenAI Chatbots, using LLM-generated interaction scenarios as probes.
• Novelty:
  1. Identification of specific factors that trigger parental concerns in child–GenAI Chatbot interactions.
  2. Characterization of parents’ desired moderation-related interactions at the conversation level.
  3. Exploration of parents’ transparency preferences, including involvement and content access dimensions.
  4. Development of actionable insights for personalized and age-appropriate parental control tools.
• Procedure and key techniques:
  1. Generated 160 synthetic child–GenAI Chatbot interaction scenarios using GPT-4.1-nano, filtered to 12 realistic and diverse examples.
  2. Conducted semi-structured interviews with 24 parents to evaluate concerns, moderation desires, and transparency preferences.
  3. Thematic analysis of interview transcripts to identify recurring themes and actionable insights.

Results

• Concrete findings:
  • Parents’ concerns stem from both the child’s prompts (e.g., harmful intentions, overdependence) and the GenAI Chatbot’s responses (e.g., missing intent, emotional impact, developmental mismatch).
  • Desired moderation strategies include refusing harmful requests, clarifying intent, tailoring responses to age, and deferring to human support.
  • Transparency preferences vary along two axes: involvement (e.g., alerts, post-interaction review) and content access (e.g., full transcripts, summaries).
• Advantage over baselines:
  • Provides fine-grained, conversation-level insights compared to existing coarse-grained parental controls.
  • Highlights the need for personalized and age-appropriate tools, addressing gaps in current systems.
• Experiments / evaluation:
  • Interviews with 24 parents of children aged 6–18.
  • Analysis of 12 diverse scenarios covering safety, emotional support, and rule-breaking.
  • Metrics include parental concern ratings and qualitative feedback.
• Limitations and future work:
  • Absence of children’s perspectives in the study.
  • Small sample size and potential priming effects from scenario presentation.
  • Future work should include child participation, develop interpretable models for moderation, and explore the robustness of proposed controls against circumvention.

Summary

This study investigates parents’ concerns and preferences for moderating children’s interactions with GenAI Chatbots. Using LLM-generated scenarios, it identifies key factors that trigger parental concerns, such as harmful child prompts and inappropriate chatbot responses. Parents desire fine-grained moderation strategies and transparency mechanisms tailored to their child’s age and family values. The findings highlight the need for personalized, conversation-level parental controls that balance safety, developmental appropriateness, and children’s privacy. Future work should involve children’s perspectives and develop scalable tools to implement these insights.