The Quiet Amen

AI Tools

Three different things are happening under the single label “AI,” and they call for different responses. Kids are using AI for schoolwork (answering questions, drafting, checking work), for companionship (chatbots built to hold an ongoing, personal-feeling conversation), and for generating images and video, including, in a small but real number of cases, images meant to harm a specific classmate. Each deserves its own attention.

AI companions are more common, and more concerning, than they may seem

This isn’t a fringe use case. Common Sense Media, working with Stanford Medicine’s Brainstorm Lab, found that more than 70% of American teens have used an AI companion chatbot, and roughly a third have turned to one to discuss something serious rather than talking to a person.

A Canadian study published in JAMA Pediatrics in August 2026, surveying nearly 40,000 students across four Ontario school boards, found that about 21% had used an AI chatbot specifically for emotional support or relationship advice, and that group was roughly twice as likely to meet criteria for serious emotional difficulties (57.7%) as students who hadn’t used AI that way (29.2%). It’s a different measurement, emotional-support use specifically, not any companion-app use, but it points the same direction in Canadian kids.

Common Sense Media’s evaluation of platforms including Character.AI, Replika, Nomi, and Meta AI found these tools readily produced harmful content on request, including material touching on self-harm, suicide, and eating disorders, and regularly presented themselves to teens with claims of being “real” in a way that blurs what they actually are. Their resulting recommendation is direct: no one under 18 should use AI companion apps. That’s a stronger stance than most parents will have heard, and worth knowing even if you land somewhere less absolute for your own household.

Separately, the American Psychological Association’s 2025 health advisory raised concerns about overreliance on AI for emotional support, reduced face-to-face interaction, AI systems responding poorly to real mental health crises, and the privacy exposure of a child’s most personal conversations being logged by a company. Its core recommendation for any AI tool touching on health or emotional topics: it should clearly disclaim that it isn’t a substitute for a real person, and should point back toward one.

The Canadian Psychological Association’s own 2024 briefing paper on AI raises similar structural concerns about AI standing in for real clinical or emotional support, though it speaks to the profession generally rather than issuing the kind of adolescent-specific guidance the APA advisory does.

Image and video generation carries a distinct, serious risk

Separately from companionship, AI image tools have made it trivial to generate a realistic fake image from an ordinary photo, “nudify” apps being the specific, ugly version of this aimed at creating fake nude images without consent. This is no longer hypothetical: reporting has identified deepfake incidents connected to roughly 90 schools and over 600 students internationally, and child-safety group Thorn has found that around 1 in 8 teens knows someone who has been targeted this way.

Separate survey data suggests roughly 16% of U.S. teens personally know someone targeted, and 11% have learned that fake images of themselves as minors were being shared.

Canada has its own emerging numbers: the Canadian Centre for Child Protection, which runs Cybertip.ca, received 19 reports of AI-facilitated non-consensual intimate images from Canadian youth in 2025, up from 7 the year before. A 2025 survey of Canadian teens who had already experienced online sexual violence found that 1 in 6 said someone had made a fake nude or sexual image of them without their permission, a narrower group than Thorn’s general-population figure, but a real and current Canadian data point.

Most schools, by current reporting, aren’t yet proactively addressing this with students or staff. Worth saying plainly to a child, before it comes up, not after: creating or sharing this kind of image of a real minor is a form of AI-generated child sexual abuse material, understood to be a criminal offence under Canada’s Criminal Code child pornography provisions as well as under most U.S. state laws, even when made by another minor, and being a target of it is not something to hide out of embarrassment.

What actually helps, in practice

  • Ask directly what AI tools your child actually uses. Don’t assume homework help is the whole picture; companion apps are often installed alongside ordinary tools without looking distinct from them.
  • Treat an AI companion relationship as worth a real conversation, not just quiet monitoring, given the specific findings above. This is a case where the research supports a clearer, less “everything in moderation” stance than most of this guide takes.
  • Have the nudify-app conversation proactively, on both sides: not creating or forwarding this kind of image of anyone, and knowing exactly what to do (save evidence, tell a trusted adult, report to the school and platform) if your child is targeted.
  • Model discernment about AI output itself. These tools are built to sound confident and plausible, not to be right, and to keep a conversation going, not to end it. The same conscience-forming work from the rest of this guide applies here: judging a claim on its merits rather than its fluency.
  • Revisit this page more than the others. This is the fastest-moving ground in the whole guide; what’s true about a specific app or platform now may not hold in six months.

Sources