Should I read my child’s AI chats?

The question lands at the kitchen table more often than it used to. Your child has been talking to an AI. You wonder if you should open the history and read it. Secret checking feels like safety. It can also feel like opening a diary they thought was theirs.

There is no single rule for every age or every tool. A better starting point is an agreement your child understands before a worry shows up.

Start with an agreement, not a raid

A midnight scroll through chat history is a reaction. A family agreement is a plan. The American Academy of Pediatrics treats media rules the same way it treats other household rules: write them down, talk them through, and revisit them as children grow. Their Family Media Plan is built for that — including privacy settings and parental controls — and it works better when children help set the terms.

That is not the same as promising never to look. It is the opposite of looking in secret. Common Sense Media is blunt about this for social apps and chatbots alike: use monitoring tools as conversation starters, not surveillance. They also do not recommend secret monitoring unless you are dealing with a serious safety issue (social media guide; ChatGPT parental controls).

You could say:

“I won’t secretly check every conversation. But if something worries you or me, we will look at it together.”

That sentence does two jobs. It tells them you are not a spy. It also tells them safety still belongs to the adults.

Age changes the job

For a younger child, the simplest version is to use AI together. Common Sense’s age-and-stage guidance for chatbots is limited use for ages 6–12, alongside a parent or caregiver — not a solo account in a bedroom (AI chatbots and your child). Sitting next to them is monitoring. You see the prompts. You hear the replies. You can say “we don’t share our address” before they type it.

As they become more independent, you step back a little and get more specific about when you may need to look. Good reasons, said out loud:

• they ask for help with something the chat started
• something feels unsafe — tone, a new “friend,” a tool you have never heard of
• they begin using an unfamiliar app

The AAP’s review of parental monitoring is useful here. Active monitoring — rules, a plan for how you will check, and consistent follow-through — plus growing room for independence, is linked with more open talk. Controlling, secretive checking is linked with children hiding what they do online (balancing safety and independence).

You do not need to quote research at the table. You do need a sentence they can repeat: we look together when something is off.

What “look together” actually means

It is not unlimited privacy. It is not a live feed of every prompt.

It is a shared screen, in daylight, with the child in the chair. They open the chat. You ask what they were trying to do. You notice what the tool asked them to share. You decide the next step together — close it, change a setting, tell a teacher, try a different prompt.

Some products will not even show you the words. ChatGPT’s parental controls, for example, let a parent manage filters, privacy and quiet hours, but Common Sense notes you will not see the teen’s actual conversations (ChatGPT parental controls). That is a reminder: settings are a fence, not a window. The conversation is still the main tool.

If you do need to read a history because you are worried about harm, say so. “I am worried, so we are going to look at this together” is honest. “I already read it last night” is a different lesson.

Keep them talking

The worst outcome is not that they used AI. It is that they stop telling you when a reply felt weird, when a tool asked for a photo, or when they used it for something heavier than homework.

So keep the agreement short enough to remember. Revisit it when a new app arrives — the start of term is a natural moment, which is also when the AAP suggests reviewing a Family Media Plan. Praise the times they come to you. That is the habit you want on the day something actually goes wrong.

You are not choosing between total privacy and total surveillance. You are choosing enough openness to keep them safe — and enough trust to keep them talking.