How to see what your child does with AI

You can see their YouTube history. You can set screen-time limits on their phone. You might even see their texts, depending on the deal you've struck. But AI chat slips past all of it. It looks like homework from across the room. It happens in an ordinary browser tab. And the conversation itself — the actual words — lives inside an account you have no window into.

That last part is the problem. With most of the internet, knowing where your kid went tells you a lot. With AI, the address is meaningless. The same chat window holds “explain photosynthesis” and questions your child would never ask you out loud. The content is the whole ballgame, and content is exactly what the standard parental tools can't see.

Why your current tools come up short

Screen-time dashboards tell you minutes-per-app. Useful for rationing games; nearly useless here, because “45 minutes in a chatbot” could be a study session or something you'd really want to know about, and the dashboard can't tell you which. Web filters block categories of sites, but AI chat tools aren't in a bad category — they're homework tools, search engines, and everything else at once. Router-level blocking is blunt, breaks the legitimate uses, and does nothing once your kid is on cellular data at a friend's house.

None of these tools are bad. They were just built for an internet of destinations, and AI isn't a destination. It's a conversation.

What you can actually do today

With mainstream AI tools, your realistic options look like this:

• A shared account. For younger kids, have them use AI through your account or a family one. You can open the chat history and read it. Know the limits: individual chats can be deleted in one tap, and a logged-out or incognito session leaves no history at all.
• Common-room computing. The oldest trick still works. AI use happens in the kitchen, not behind a closed door. Proximity isn't surveillance — it's presence, and it changes behavior gently.
• Platform parental controls. Some AI products now offer account linking for teens. Turn them on where they exist — just read what they actually cover, which is usually settings and safeguards, not a transcript of what was said.

Notice what's missing from that list: anything airtight. It doesn't exist. Any teenager motivated to hide their AI use from you will succeed, and the arms race itself does damage. Which brings us to the part of this that matters more than the tooling.

Surveillance teaches secrecy. Visibility teaches trust.

There's a version of monitoring that reads your kid's chats secretly, catches something, and confronts them with it. It feels decisive. It works exactly once. After that, you've taught them precisely what to hide and how — and they will out-engineer you, because they have more time and better incentives.

The version that works long-term is boring and open: they know you can see it. Not as a threat — as a plain fact of how AI works in your house, the same way they know you're in the room when they swim. The point was never catching them. The point is knowing what they're curious about, so the questions they're asking a machine can become conversations they have with you.

“I can see your chats — not because I don't trust you, but because this stuff is new for both of us, and it's my job to be around while you figure it out.”

And let the visibility age with them. Full transparency at eight is appropriate. At fifteen, it should be a negotiated, shrinking thing — because the goal was never permanent oversight. It's a kid who eventually needs none.

What to actually look for when you read

If you do have visibility, resist keyword-hunting. The signal is rarely one alarming message; it's patterns. Is the AI doing their thinking for them? (See our guide on AI and homework.) Are they asking it things they used to ask you? Is it drifting from tool to confidant? Those patterns are conversation starters, not evidence for a prosecution. Lead with curiosity — “I saw you were asking about black holes, that's so cool” opens more doors than anything that begins with “we need to talk.”