Children Who Confide in AI — Research on Emotional Attachment and How Families Can Approach It

4 min read日本語版あり
Audience
Parents raising children in a world where generative AI is widespread
Target length
~1,500 words
Status
Draft v2 (translated from Japanese v1)
Original
../187_ai_chatbot_kids.md

Lead

The question "can a robot be a friend?" was once philosophical. By 2024, Character.ai was reporting roughly 200 million monthly active users. Children routinely talking to AI and sharing their feelings with it is no longer an unusual scenario.

For parents, "how to navigate AI" is no longer a future problem. But the research that would allow definitive answers — "allow it" or "ban it" — does not exist as of 2026. This article is not a prescription. It is an attempt to organize what is known well enough to ask the right questions.

What Is Actually Happening

Emotional attachment to AI companion apps — Character.ai, Replika, and others — has been observed even in adults. Skjuve and colleagues (2021) documented long-term attachment formation to a chatbot in a case study and reported that users experienced feelings of friendship and support [1]. Brandtzaeg and colleagues (2022) analyzed, through qualitative research, cases in which users of social chatbots understood their human–AI relationship as genuine friendship [2].

Children present distinct additional concerns. The "" that Piaget observed — the tendency to attribute life and feelings to non-living things — is broadly seen from early childhood through around age 10 [3]. Children treat AI more naturally as "something with feelings" than adults do. This tendency fades with age, but as AI responses become more sophisticated, the impression that the AI "seems to have emotions" grows stronger.

In 2024, a case in the United States involving a teenage user's suicide drew attention to the potential role of an AI companion in the lead-up to the event, generating significant ethical and legal discussion. Individual cases cannot be used to draw general conclusions, but "the risk that emotional dependence on AI could affect children" is now a recognized real-world concern, not a theoretical one.

Learning Support: Possibility and the Problem of Over-Reliance

Using generative AI to support children's learning carries both promise and pitfall.

UNESCO, in 2023, was the first UN agency to publish guidance on generative AI in education, taking the position that "AI is a useful educational tool but cannot replace critical thinking or human interaction" [4].

Gerlich (2025) found a possible association between ChatGPT use and a decline in critical thinking, though the authors simultaneously noted limitations in sample and method — the finding should be treated as preliminary [5].

The important distinction is between "using AI to get answers" and "using AI to help frame better questions." In the first mode, the child may be bypassing the opportunity to think; in the second, thinking may be expanded. The same tool, used differently, produces different outcomes. What matters is how the interaction is designed.

AI-Generated Images and Family Records

A separate area deserves attention. As AI image synthesis becomes routine, the temptation to "create moments that never happened" grows. Editing or compositing a child's photos or family mementos with AI introduces the risk that a family's record becomes a mixture of things that actually occurred and things that did not. The problem of a child growing up and being unable to verify the authenticity of their own history is new enough that research has not yet caught up to it.

Framing the Questions — What Parents Can Do Now

There are few research-settled answers, but some practical orientations can be identified.

Ask rather than surveil: Monitoring a child's AI use may compromise their developing sense of digital autonomy. Instead, "What did you ask the AI today?" or "What did it say?" woven into everyday conversation gives the child a regular opportunity to reflect on their own use.

Talk about what AI is, calibrated to age: There is an argument that a child can accurately understand that AI does not understand emotions around third or fourth grade (the late stage of concrete operations). But being told "it doesn't have feelings" early does not entirely prevent attachment from forming. Explanation repeated over time, combined with ongoing conversation, is what gradually builds understanding.

Design the rules for learning use together: "Think it through yourself before asking the AI," "rewrite the AI's answer in your own words before using it" — building these norms collaboratively with the child, rather than imposing them as restrictions, creates an opportunity for the child to think actively about how they relate to the tool.

A Graduated Approach

The choice is not between "block AI entirely" and "allow unlimited use." A staged involvement is more realistic:

Summary

Children's use of AI is, right now, not a problem with a clear answer. Research hasn't caught up; the technology is changing quickly. That is precisely why designing "how to engage with it" is more durable than deciding "ban or allow."

The fact that AI can become a child's conversational partner is not going to change. A parent who remains curious about those conversations — rather than trying to eliminate them — is taking the most sustainable form of involvement available.

References

  1. Skjuve M, Følstad A, Fostervold KI, Brandtzaeg PB. My chatbot companion — a study of human-chatbot relationships. Comput Human Behav. 2021;119:106722. doi:10.1016/j.chb.2021.106722
  2. Brandtzaeg PB, Skjuve M, Følstad A. My AI friend: how users of a social chatbot understand their human–AI friendship. Hum Commun Res. 2022;48(3):404–429. doi:10.1093/hcr/hqac008
  3. Piaget J. The Child's Conception of the World. London: Routledge; 1929/2002. ISBN: 978-0415168502
  4. UNESCO. Guidance for Generative AI in Education and Research. Paris: UNESCO; 2023. Available from: https://doi.org/10.54675/PCNF4316
  5. Gerlich M. AI tools in society: impacts on cognitive offloading and the future of critical thinking. Societies. 2025;15(1):6. doi:10.3390/soc15010006

Related