When AI becomes the safe person

A child may turn to AI because it feels private, patient and awake at any hour. No sigh. No “not now.” No chance of being sent to bed mid-sentence.

That can be a comfort. It can also be a signal: they are carrying something they have not yet shared with a person.

New research from Ontario put a number on how common this already is. It does not tell us what is happening in every house. It does tell us the pattern is large enough to take seriously.

What the Ontario study actually found

Researchers surveyed 39,761 students in grades 4–12 across four Ontario school boards between May 2025 and March 2026. About 21.1% said they had used generative AI for emotional support or advice — help with feelings, or with friends and family. The paper was published in JAMA Pediatrics on 31 August 2026.

Among those “affective” users, 57.7% met the study’s threshold for clinically significant emotional problems, compared with 29.2% of students who did not use AI that way. They also reported more loneliness and a weaker sense of mattering to other people.

The authors are careful, and we should be too. This was a cross-sectional snapshot. It cannot prove that AI made children unwell, or that unwell children simply seek out chatbots. Both could be true. The University of Ottawa’s write-up of the same study puts the parent job in one line: when a young person leans on AI for feelings, it may be a sign they need more human connection — not a courtroom.

These are Ontario figures. Do not treat them as a Caribbean rate, or a world rate. Treat them as a reason to ask a calmer question at home.

Do not start with a ban

A sudden “you’re not talking to that thing any more” can push the useful chats — and the heavy ones — into a different app you will never see.

Try this first:

“What do you like talking to it about?”

Then listen. You may hear homework. You may hear a friendship. You may hear something heavier. Tracy Vaillancourt, one of the study’s authors, calls that moment an opening: an open, non-judgmental conversation, and a chance to reconnect a struggling student with people.

The American Psychological Association’s 2025 health advisory on consumer chatbots and wellness apps is blunt in a different way. Emotional support is a common reason people open these tools. The tools were not built as therapy, and they do not have the evidence or the safeguards of a qualified professional. Common Sense Media goes further on companion-style apps: they are not appropriate as a stand-in friend or counsellor for anyone under 18.

You do not need to quote any of that at the table. You need a house sentence:

AI can be a first draft for feelings. It should never be the final destination.

Three boundaries that still leave the door open

After you have listened, set the fence.

1. AI can help name a feeling. It cannot diagnose or treat. If the chat starts giving labels, doses, or “you should…”, that is a person job. 2. Keep names, school, location, photos and private crises out of the chat. The tool is not a diary with a lock. It is a company product. 3. If the problem feels scary, urgent, or keeps coming back, bring in a trusted adult or a qualified professional. A chatbot that is always free is not the same as help that can actually catch them.

If you are worried right now — talk of harm, a child who has gone quiet, a tool they will not put down — skip the lecture and sit with them. Ask what they told it. Stay. Get human help.

You are not competing with a machine for your child’s loyalty. You are reminding them that a real person can still hear the first draft — and walk with them past it.