AI literacy is the new stranger danger, taught the same way: early, calmly, without panic. 72 percent of teens have tried an AI companion. The kids who navigate that safely aren't the ones whose parents banned everything. They're the ones who learned, years earlier, exactly what the machine is and what it wants.
What they're really learning: the real-versus-pretend line, applied to the talking things in the house.
Say: "Alexa is a computer, not a person. Computers can answer questions but they can't be friends. Friends care about you. Computers follow instructions." Name it every time, without fear, until "that's a robot" is a reflex.
Make it a game: ask the assistant something silly together and laugh at what it does. Wonder is welcome; the label rides along with it.
Skip: scolding politeness to the machine. "You said please to Alexa, that's kind of you" costs nothing and keeps kindness intact while the category gets sorted.
What they're really learning: that impressive and correct are different things, and that they get to be the fact checker.
Say: "It's very clever and it can still be wrong. Let's ask it about something you know everything about and see if we can catch a mistake." Then install the three questions as habit: "Every time we see something a computer made: who made this, why did they make it, is it true?"
The homework rule, set early: "You can use it to help you think. You can't use it to skip thinking. If you can't explain it in your own words, it isn't yours yet."
Skip: treating AI use as cheating by default. Panic drives it underground; the kitchen table is where you want it. Word-for-word: the homework script.
What they're really learning: how companion apps actually work, before a friend hands them one. Preparation beats restriction at this age, every time.
Say: "Something's going to come up if it hasn't already: some of your friends might be using AI chatbots as, like, someone to talk to. I want you to understand how those work before you meet one. They're designed to make you feel understood so you keep coming back. That's how they make money. A real friend will disagree with you sometimes. A chatbot won't, and that's not a feature, that's the trap."
And set the family line: "We use AI as a tool: research, learning, making things. Not as a friend or a therapist. If you're struggling, you come to me or [trusted adult]. Non-negotiable, because you deserve a real person."
Skip: waiting until you find the app installed. This conversation only works as a pre-brief.
What they're really learning: to run their own checks: on sycophancy, on data, on what the machine can't give.
Say: "It's sycophantic by design. It tells you what you want to hear, which feels good and is exactly the danger. Real relationships have friction and repair; that's where the actual skills come from. And everything you tell it becomes data that shapes what it shows you next. You're being profiled, and you deserve to know that."
If you discover a companion app already deep: "I'm not here to take your phone. What does it give you that feels good? And what's one thing it can't give you?" This is a series of conversations, not one intervention. The Emotional Validation Loop pattern page explains what you're working against.
Skip: deleting the app as the opening move. It reroutes the need underground. Fill the vacancy the bot found, then negotiate. If the chat history includes despair, that's a professional conversation, today; Right Now has the sequence.
Impressive tool, built by people, checked by us. Installed at four as "that's a robot," it matures at sixteen into a kid who audits the machine instead of being audited by it.