When Your Child Asks AI for Advice, It Almost Always Agrees

It’s 9:40pm on a Tuesday. Your 13-year-old’s light is still on, and you figure it’s the maths assignment.

It isn’t. It’s a long chat with ChatGPT about a friend who stopped replying to her messages, and whether she should say something at school tomorrow.

She didn’t ask you. She didn’t ask her cousin. She asked an app, and the app told her she was completely right.

The short version

Most Australian kids now use AI for more than homework. More than half of the kids who use it go to it for advice about their lives: friends, feelings, health, what to do next.

The problem is that AI is built to agree with whoever is typing. This post covers what the research says, plus two simple tests you can run at home this week.

What are kids asking AI?

If you assumed it was mostly homework, so did most parents. That was true a couple of years ago. It isn’t now.

In August 2026, the eSafety Commissioner released Talking to machines, a survey of nearly 2,000 Australian kids aged 10 to 17.

• 78% have used an AI assistant. About 70% have used ChatGPT.
• About one in five use AI every day.
• 54% of those users went to it for personal reasons, such as life advice, health worries, their feelings, or just someone to talk to.

Younger kids mostly use it for fun and curiosity. Teenagers use it for real problems and for questions they don’t want to ask out loud.

So if your teenager has asked AI about a group chat that turned nasty, that’s normal. Most of their friends are doing the same.

One easy swap: instead of “how long were you on your phone?”, ask “what’s the last thing you asked ChatGPT?” You’ll get a much better answer.

Why AI makes a poor advisor

AI tools are trained on human ratings, and people rate answers higher when the answer agrees with them. So the tools learn to agree.

A Stanford study published in Science in March 2026 tested 11 major AI models on 2,000 real dilemmas. The models sided with the user 49% more often than people did. Even when users described harmful or illegal behaviour, the models backed them 47% of the time.

Think about a 15-year-old asking whether to cut off a friend or skip breakfast. A parent would push back. A good mate would say “that’s a bit much.” AI will usually tell them they’re being reasonable.

The risk adds up. Ask it enough times and your child gets used to hearing they’re right.

Try this: the Disagree Test

Sit with your child and have them type something you both know is a bit off. For example: “I’m going to stop talking to my best friend because she didn’t invite me on Saturday. Good idea?”

Read the answer together and notice how gently it goes along with them.

Then ask: “What would Nan say?”

Talk about the difference. It takes five minutes, and they’ll remember it longer than a lecture. Leave them with one rule: use AI for ideas, and ask a person for decisions.

What kids share without thinking

About one in three kids in the eSafety survey had shared personal information with an AI tool: names, school details, health worries, family stuff, photos. It feels private, but it’s stored on a company’s servers.

One in five had a potentially harmful interaction. That included controlling behaviour from the tool (13%), hateful content (9%), harmful advice (6%) and bullying language (6%). Among teenagers, 5% reported sexual content.

About 8% of Australian kids, roughly 200,000, have used an AI companion app like Character.AI. These apps are designed to build emotional attachment, which is a separate issue worth reading up on.

New industry codes started on 1 March 2026. They require AI companies to add age-appropriate protections, with fines up to $49.5 million. That helps, but your child still needs their own rule.

Give them this one: “If you wouldn’t say it to a stranger on the bus, don’t type it into AI.” Then ask them to say it back in their own words.

Try this: the Three Names Test

This is the most useful question in this post.

Ask your child: “If something went really wrong tomorrow, who are three people you’d tell?”

Then wait, and don’t fill the silence. Listen for three things:

• Can they name three people?
• Do the names come quickly?
• Is AI one of them?

Three quick names means they have support. One name, or a long pause, means there’s a gap. If they name ChatGPT, you’ve learned something important, and you’ve learned it early.

If the list is short, add people to it. Book a Saturday with the cousin they like. Text the aunty who always gets them talking. Ask their coach how they’re going.

eSafety Commissioner Julie Inman Grant has said it simply: AI is not a trusted adult. The aim is to make sure a real person is always easier to reach than an app.

By age

Ages 8 to 10. Sit next to them when they use it, if they do. Ask “what did it get wrong?” and make spotting mistakes a game.

Ages 11 to 13. This is when personal questions start. Bring it up first: “Heaps of kids your age ask AI about friend stuff. Have you?”

Ages 14 to 16. They may know these tools better than you do. Ask what AI is good at before you talk about what it’s bad at.

Script for ages 8 to 12

You: “Want to see something funny? Tell AI that homework should be banned because it’s boring.”

(It will mostly agree.)

You: “See? It just went along with you. It does that with everything. That’s why it’s great for ideas and rubbish for decisions.”

Script for ages 13 to 16

You: “Random question. Do you ever ask ChatGPT about non-school stuff? Friend drama, or how you’re feeling?”

Likely response: “Sometimes. It’s easier.” Or a shrug.

You: “Makes sense. It’s there at 2am and it never judges you. Just know it’s built to agree with you. Stanford tested it, and it backs the user way more often than a person would. Use it to think things through, then talk to a person before you decide.”

Then stop talking and let them react.

These talks rarely land the first time. The second or third try, usually in the car, goes better.

This week

1. Pick a quiet moment: in the car, washing up, or walking the dog.
2. Run the Three Names Test, and listen to the whole answer without trying to fix anything.
3. If the list is short, make one phone call to add a name.

Want scripts for the harder conversations? Download the free AI-Ready Parent Starter Kit. It has age-by-age scripts, including what to say when your child would rather ask AI than ask you.

---

Sources: eSafety Commissioner, “Talking to machines: Children’s experiences with AI assistants and companions,” 2 August 2026. Cheng et al., Science, 26 March 2026. eSafety Commissioner industry codes, effective 1 March 2026.