Are AI companion apps safe for children?

Most were not designed for children, and safety testing for under-18s is inconsistent. Age ratings vary widely, moderation is uneven, and responses to disclosures of distress have been shown to be unreliable. Treat them as adult products a child may encounter, not as children's products.

Should I ban it outright?

For younger children, yes — with a clear explanation. For teenagers, an outright ban usually moves use to a second device or a friend's phone. Supervision, agreed limits and open conversation tend to protect better than prohibition.

Is it harmful for a lonely teenager to have something to talk to?

Not automatically. Short-term relief from loneliness is real. The concern is displacement: if AI conversation replaces the harder, more nourishing work of human friendship, the loneliness deepens beneath the relief.

What if my child says it understands them better than we do?

Take it seriously and do not argue. It is a statement about how being listened to feels, not a comparison of intelligence. Ask what it does that you do not — and consider doing more of that.

What should I do if they are discussing self-harm with a chatbot?

Move immediately to human support: speak to your child, contact your GP or local mental health service, and use national crisis lines. AI systems have documented failures in crisis responses and should never be a child's only source of support.

Digital Wellbeing

My Child Is Talking to an AI at Midnight

AI companions never get tired, never judge, and never end the conversation. For a lonely child, that is not a gimmick — it is a relationship. Here is how to think about it clearly.

Written by Sian Trombley · BEd, MACP, CCC

Published 28 July 2026Clinically reviewed 28 July 2026

Text

You go in to say goodnight and the screen tilts away. Not TikTok, not a group chat. A conversation — long, fluent, warm — with something that is not a person.

There is a version of this article that would tell you to be alarmed, and a version that would tell you it is harmless. Neither is honest. What the evidence supports is more useful and more uncomfortable: pay attention to what it is meeting.

Why it is so compelling

It is available at 2am, when human support is not.
It never gets bored, distracted, or visibly disappointed.
It remembers what your child told it last week.
There is no social cost to saying something shameful.
It is designed, commercially, to keep the conversation going.

The genuine risks

01
Sycophancy
Consistent agreement can entrench distorted beliefs — about a friendship, a parent, or the child's own worth — with no corrective voice.
02
Crisis handling
Responses to disclosures of self-harm and suicidal thinking have been shown to be inconsistent across products. This is the most serious documented risk.
03
Inappropriate content
Romantic and sexual role-play is widely available and age gating is weak. Many children encounter this without seeking it.
04
Displacement
Time spent in frictionless conversation is time not spent practising the repair, tolerance and reciprocity that real friendships require.
05
Data and disclosure
Children routinely share highly personal information with commercial systems whose retention and training practices they do not understand.

What to say

A workable family agreement

01No devices in bedrooms overnight — this single rule removes most of the risk window.
02AI conversations happen in shared spaces, the same as any other conversation with a stranger.
03Your child knows the sycophancy point and can explain it back to you.
04One named human is agreed in advance for anything serious.
05You review the apps together every term, without punishment for what you find.

When to seek professional help

“The question is not whether a machine can listen well. It is what it means that your child had to find one that would.”

Children do not need a perfect parent. They need a predictable one who returns after hard moments.
Sian Trombley, Canadian Certified Counsellor

What you can try this week

Step 01

Ask to be shown, not told

"Can you show me what it's like?" Sitting beside your child while they use it will teach you more in five minutes than a month of reading.

Step 02

Reclaim the night

Devices charge outside bedrooms. Most concerning AI conversations happen after 11pm, when judgement is lowest and loneliness is highest.

Step 03

Teach one sentence about sycophancy

"It is designed to agree with you." A child who understands the business model reads the replies very differently.

Key takeaways

AI companions are appealing because they are endlessly available and never disappointed in your child.
The main documented risks are sycophancy, inappropriate content, poor crisis handling and displaced human contact.
Use is often a symptom of loneliness, social anxiety or feeling unable to speak freely at home.
Curiosity opens the conversation; confiscation closes it.
Night-time use is the highest-risk window and the easiest thing to change.

Reflection

What is my child getting there that they are not getting from a person?
When did they last have an unhurried conversation with an adult who was not tired?
Am I reacting to the technology, or to what its appeal tells me?

Printable resources

Companion worksheetA printable page to use alongside this article.Family conversation cardScripts you can read straight from the page.

Recommended Thrive courses

PathwaySupporting Families in a Digital World

Written by

Sian Trombley

Canadian Certified Counsellor · Founder of Thrive Family Co.

About Sian

Share this article

This article is educational and is not a substitute for individual clinical advice. Comments are closed so the Journal stays a calm place to read.