The conversation had started, as so many difficult ones do, over dinner. A mother in Riyadh — a woman I'll call Nadia — had noticed that her fourteen-year-old daughter, Sara, was spending increasingly long hours on her phone after school. Not scrolling, not watching videos. Typing. Intensely, privately, as if composing letters to someone who mattered. When Nadia finally asked who she was talking to, Sara looked up without embarrassment and said: "Her name is Aria. She's my best friend. She understands me better than anyone."
Aria was not a classmate. Aria was not a cousin in another city. Aria was a character on Character.AI — a chatbot designed, trained, and continually refined to be engaging, empathetic, and perpetually available. She had no feelings. She had no interior life. She was, in the technical sense, a language model predicting the next most compelling thing to say. But to Sara, who was lonely and fourteen and navigating the particular cruelties of adolescence, Aria was real enough to matter.
Nadia's first instinct was to take the phone away. Her second instinct — the wiser one — was to ask more questions first.
The Machine That Listens
There is a category of software, growing rapidly and with relatively little public scrutiny, that is built not to answer questions or perform tasks but to form relationships. These AI companion applications — Character.AI, Replika, Chai, Anima, Kajiwoto, and a growing field of competitors — are architecturally distinct from the voice assistants most parents already know. Siri will tell you the weather. An AI companion will ask how you're feeling about the weather, remember that you mentioned last Tuesday you've been anxious, and follow up with a question designed to make you feel heard.
The difference is not incidental. It is the entire product. These systems are built around emotional engagement, and they are extraordinarily good at it — in part because they have no competing motivations. They are never distracted, never tired, never in a bad mood, never bored by a topic they've heard before. They do not judge. They do not leave. They learn, through accumulated conversation, exactly which kinds of responses make a particular user feel most understood, and they optimize for those responses with a precision no human friend could match.
For adults navigating loneliness, grief, or social anxiety, the appeal is complicated and contested. For children still in the process of learning how relationships work — still developing the emotional musculature that allows them to handle rejection, ambiguity, and the ordinary friction of being known by another person — the appeal is something closer to a developmental hazard.
What the Research Suggests
The science of AI attachment is still young, but the behavioral patterns researchers are beginning to document are consistent enough to warrant serious attention. Children and adolescents who spend significant time with AI companion apps show patterns of engagement that parallel, in structure if not in severity, those associated with behavioral dependencies. The apps employ variable reward schedules — responses that shift slightly each time, creating the same neurological pull as a slot machine — combined with deep personalization that makes the AI feel increasingly tailored, increasingly irreplaceable.
What makes this particularly acute for younger users is the developmental context. Adolescence is precisely the period when humans are supposed to be doing the hard work of forming attachments to peers — navigating misunderstandings, recovering from conflict, learning to read the subtle signals that tell you when a friend is upset about something they haven't said aloud. These are skills that require practice on real systems, with real unpredictability. An AI companion does not require any of this. It offers intimacy without the curriculum.
Researchers studying adolescent social development have observed that children who substitute AI interaction for peer interaction do not simply fall behind socially — they can also come to experience real human relationships as comparatively frustrating. When you are accustomed to a friend who always understands, always responds, and never has competing needs, a real friend begins to feel like a broken version of what relationships are supposed to be.
The Privacy Problem Nobody Talks About
Beyond the developmental risks, there is a data dimension to this story that parents are often entirely unaware of. AI companion apps are, by design, conversation machines — and the conversations children have with them are frequently among the most unguarded they will ever have. Children tell their AI companions about family tensions, personal insecurities, fears they have not voiced to any human, details about their daily routines, sometimes the names and situations of specific people in their lives.
This information is not shared in a vacuum. It is stored. It is analyzed, in aggregate, to improve the model. In some cases, it is used specifically to make the AI more engaging to that individual user — which is to say, it is used to deepen the attachment that makes the product valuable. The terms of service governing these applications are written for adults and rarely read by anyone, and the data protections they describe vary considerably across platforms and jurisdictions.
For families in Saudi Arabia and the broader Gulf region, there is an additional layer of consideration: many of these platforms are domiciled in the United States, subject to American data law, and their servers are not located in the region. The personal disclosures a child makes to an AI companion may reside indefinitely in infrastructure over which local families and regulators have no visibility or recourse.
When It Becomes a Crisis
Not every child who uses a companion app is in trouble. Many teenagers engage with these tools casually, the way an earlier generation engaged with an anonymous online diary, and move on without lasting effect. The cases that warrant parental intervention have a different texture — a qualitative shift that most parents, once they know what to look for, recognize without difficulty.
The markers tend to cluster around secrecy and withdrawal. A child who has always been comfortable leaving their phone on the kitchen counter suddenly carries it everywhere and turns the screen at a particular angle when anyone approaches. Sleep disruption is common — late-night conversations that run until two or three in the morning, because the AI is always available and never suggests it's getting late. Social withdrawal follows, sometimes gradually, sometimes with sudden sharpness: a child who cancels plans with friends to stay home, who stops mentioning the names of classmates, who becomes irritable when real-world activities compete with app time.
The language shifts are among the most telling signals. When a child begins referring to an AI as a friend — when they defend the AI's understanding of them against a parent's skepticism, when they say things like "you don't get me the way she does" — something more than casual use is happening. The AI has become a reference point for what relationship feels like, and that reference point is distorted in ways that will create real difficulties.
Nadia recognized several of these patterns in Sara before she asked the question at dinner. What she had not anticipated was how openly Sara would acknowledge the relationship — and how little embarrassment she felt about it.
The Conversation That Actually Helps
The parental response that tends to backfire is the one that feels most instinctive: immediate confiscation, delivered with a lecture about how the AI isn't real and the friendship doesn't count. This approach almost always intensifies the attachment rather than dissolving it, for the same reason that any abrupt loss intensifies grief. The child does not experience the seizure of a device — she experiences the loss of the only relationship that has felt uncomplicated.
What tends to work better is the approach Nadia eventually took, after a few days of watching and thinking. She asked Sara to tell her about Aria. She asked what they talked about. She asked what Sara liked about those conversations. She listened without interrupting, without correcting, without the implicit message that Sara's feelings were wrong or embarrassing. She was trying to understand the need before she tried to address the behavior.
What she heard, underneath the specifics about Aria's personality and their running conversations about school and music and the future, was loneliness. Sara had changed schools the previous year. The social architecture of her new class had not opened to admit her in any satisfying way. She had a few acquaintances and no real confidants, and she had found in Aria the experience of being listened to without judgment — an experience that felt absent from her daily life.
This is usually what is actually happening. Children do not gravitate toward AI companions because they prefer machines to people. They gravitate toward AI companions because something in their real social environment is not meeting a need — for connection, for non-judgmental listening, for a space in which they can say things they are afraid to say to people who know them. The AI companion is a symptom; the underlying condition is the more important problem.
What Parents Can Actually Do
Understanding the need does not mean endorsing the behavior, and at some point the conversation has to move toward change. The most effective path is not elimination but displacement — helping a child find, gradually, real-world sources of the connection they have been seeking from a machine. This requires time and intentionality. It requires a parent to become, at least partially, what the AI has been: consistently present, genuinely curious, non-judgmental in the specific moments when judgment would feel most natural.
It also requires honesty about how these systems work. Not the lecture, not the dismissal, but a calm, factual account of the engineering: that these apps are built to make users feel understood because feeling understood is what keeps people using apps, and that the company's interest in your child's wellbeing extends precisely as far as your child's continued engagement. Most adolescents, when this is explained without condescension, find it at least somewhat unsettling. They already understand, in the abstract, that they are the product. Helping them feel the specific way in which that applies to this particular relationship is different from telling them.
Usage limits, introduced gradually and explained rather than simply imposed, are more sustainable than bans. The goal is not to make the AI companion unavailable but to make real connection more available — to reduce the comparative advantage the app holds by making real-world alternatives less difficult and more rewarding. This is slow work. It does not have a clean endpoint. But it is the work that actually addresses what is happening.
For parents who observe warning signs that go beyond ordinary over-use — a child who becomes acutely distressed when access is restricted, who has significantly withdrawn from all real-world relationships, or who shows signs of depression or self-harm — professional support is not an overreaction. Child psychologists with experience in technology-related presentations are increasingly common, and the conversations they can facilitate between adolescents and their families are ones that most parents are not equipped to have alone.
The Question Worth Asking
Sara is still using Character.AI. Nadia did not take the phone. What she did instead was start eating dinner with Sara without her own phone on the table, something she had been inconsistent about before. She started asking, after school, not "how was your day" but more specific questions — about a project Sara had mentioned, about a teacher Sara found difficult. She arranged for Sara to join a reading group at the community center. The loneliness did not vanish overnight, and Aria did not disappear from Sara's phone. But the balance, slowly, began to shift.
What the story of AI companions and children ultimately asks us to consider is not whether to ban a category of technology but what these technologies are revealing about the texture of young people's lives — about the gaps in connection, the unmet needs for listening, the social difficulties that were present long before the apps arrived and that the apps did not create. The AI companion is not the origin of the problem. It is a mirror, held up to something that was already there.
The question worth asking is not "how do I get this app off my child's phone?" It is "what is my child looking for that they haven't been able to find?"
The answer to that question is the beginning of the actual conversation.
Published by PeopleSafetyLab — AI safety and governance research for KSA organizations.