Is ChatGPT safe for kids?

“Safe” is doing too much work. Three risks, three different answers — and the most feared one is where the usual advice is actively wrong.

Published 13 September 2026 · Wisesoft.ai · 6 min read

“Safe” is doing too much work in that question. There are three separate risks, they have different answers, and the one most parents worry about is not the one that actually bites. Taken in order of how likely they are to affect your child this year:

1. What your child tells it

This is the real one, and it is mundane. A chatbot keeps whatever you type into it. A child asking a perfectly innocent question will often hand over their full name, their street, or their school along the way — not because they were tricked, but because that is how people talk.

Here is a message of exactly the shape we see constantly:

The bot can explain fractions perfectly well without knowing who is asking. The skill is not refusing to use AI — it is keeping the question and dropping the identifiers.

2. Believing a confident answer

An AI does not know when it is guessing, and a guess and a fact leave its mouth sounding identical. For homework this is the risk that costs marks; for anything medical or financial it costs more.

This one is genuinely teachable and we have written about it at length in why AI makes things up. The short version for a child: ask which part of that it actually knew, and which part it filled in.

3. Not knowing who is on the other side

This is the one parents fear most and it is the one where the common advice is worst. The usual heuristic — bad spelling and lowercase means a real person, proper grammar means a bot — is now false.

We measured it. Against an older set of chat transcripts, a single rule — does the reply start with a capital letter? — identified the bot 32 times out of 32. “Casual writing and emoji means human” scored 31 out of 32. Those numbers are exactly why we stopped teaching it.

In 2026 a language model writes however it is asked to. “lol pizza obviously” is well within reach of any chatbot. A child taught “proper grammar means robot” walks away more foolable than before, which is worse than teaching them nothing.

What works instead is asking — a question whose answer requires having been somewhere, or having a body, or remembering something from earlier in the conversation. Not “are you a real person?”, which both a person and a bot answer yes to.

So — is it safe?

Used with an adult who knows these three things, yes, and it is genuinely useful. Handed over unsupervised to a child who has been told only “be careful online”, not especially — not because the tool is malicious, but because “be careful” is not a skill.

Worth knowing as well: most mainstream AI tools set their own minimum age at 13, and their terms say so. Below that, a child is generally using an adult product on an adult account. That is a separate question from whether they are mature enough, and it is covered in what age kids should learn about AI.

Read next

• What not to tell a chatbot
• Why AI makes things up