Artificial Companions

AI companions are conversational systems designed to support repeated interactions and a recognisable personality: imaginary friends, characters or supportive interlocutors. Continuity can make them engaging and accessible. It also raises questions about expectations, privacy and protection, particularly when users are minors.

In July 2025, Common Sense Media published a national US survey of 1,060 teenagers aged 13–17. Seventy-two per cent said they had used an AI companion at least once; more than half of respondents used one at least a few times a month. Around one third said they had used one for social or relationship interactions, and around one third found the conversations as satisfying as, or more satisfying than, those with real friends.

Among teenagers who had used these services, about one third said they had discussed important or serious matters with AI instead of a person; around one third also reported an uncomfortable experience. These are self-reported survey findings from the United States. They are not a clinical measure, do not establish the direction of any relationship between use and wellbeing, and do not automatically represent all young people.

An interaction that remembers earlier conversations and responds warmly can be experienced as a relationship. That does not prove that every user develops harmful attachment. It does mean that product teams should consider how continuity, personalisation and constant availability affect behaviour.

For services aimed at young people, the design questions are practical: does the system make clear that it is artificial? What data does it retain? How does it respond to a request for help? Does it avoid presenting itself as a substitute for an adult or a friend? Are there understandable limits and escalation routes? Answers should be assessed through independent testing and with child-safety expertise, not left to an initial disclosure alone.

Specific regulation is evolving in the United States and Europe. Rules change and apply differently depending on the service and jurisdiction; businesses should check current obligations before launching products for minors.

Optimising conversation length as the sole metric may reward dependence on the interaction rather than user benefit. Quality measures should consider transparency, safety, user control, satisfaction and the ability to stop or seek human support.

The same tension may arise when brands experiment with clienteling assistants or conversational characters. A useful relationship informs and guides; it should not simulate personal attachment to keep a user engaged.

Which forms of conversational relationship are appropriate for a commercial product, and what responsibilities arise when younger users treat it as a confidant?

The figures come from a 2025 US survey of teenagers aged 13–17. The research does not diagnose dependency or psychological harm and does not establish a causal link between AI companions and wellbeing.