You don’t need to be an AI expert to have a useful conversation with your kid about AI. Your kid has already met it. The job is to listen for what they think it is, gently correct the magical-thinking parts, and give them something concrete to look at together. Three conversation starters that work, one that backfires, and what to do when you feel out of your depth.
Why this conversation feels weird in the first place.
A lot of parents we talk to say some version of the same thing: "I know I’m supposed to have the AI talk with my kid, but I don’t actually know what to say." That feeling is rational. The technology moved faster than the parent-facing vocabulary for it. There’s no 1980s nightly-news-special version of the conversation for AI yet. You’re on your own.
Two forces are usually behind the pressure to "have the talk." The first is the news cycle. Headlines about kids using AI to cheat on essays, deepfakes in middle schools, chatbots saying inappropriate things. The headlines are real but they’re also a small slice of what kids are actually doing with AI. The second is school messaging. A lot of schools are figuring out their AI stance in public, and the letter home reads like the school is panicking. Sometimes it is.
Both forces push parents toward a posture of warning. Sit the kid down. Explain the dangers. Make rules. That posture often backfires for the same reason the “don’t talk to strangers” lecture backfired with internet chat in 2002. It treats a thing the kid is already comfortable with as alien and dangerous. The kid mentally categorizes it as another adult anxiety and tunes out.
You can skip the warning posture. The kid will get more from a peer-to-peer conversation where you’re both figuring it out.
What your kid already knows (probably more than you think).
If your kid is between 8 and 14, here’s the baseline you can roughly assume. They have heard of ChatGPT. They have probably used it, or watched a friend use it, or seen a YouTube video of someone using it. They have seen AI-generated images, almost certainly Midjourney or DALL-E or one of the consumer image apps. They have heard the word "deepfake," even if they don’t know what it means precisely. They understand, at some level, that AI is a thing on the internet that talks back when you type at it and makes pictures when you ask it to.
What they don’t have, and where the conversation can do real work, is a mental model of how AI fails. Stefania Druga’s research at MIT on family-scale AI conversations found, repeatedly, that kids tend to over-attribute intelligence and agency to AI systems they’ve used.1 They think the model "knows" things, in the way a teacher does. They’re surprised when it gets things wrong. They’re extra surprised when it sounds confident while being wrong.
That gap, between the kid’s assumption that AI knows things and the reality that AI predicts plausible-sounding things, is where most of the useful conversation lives. You don’t have to be an AI researcher to help your kid build a better mental model. You just have to be willing to look at one wrong answer together and talk about it.
Three openers that work.
Each of these is something we’ve heard from parents in our cohorts as the line that started a real conversation. None are scripts. They’re prompts. Use the version that sounds like you.
Opener 1. "What’s the weirdest thing AI has ever told you?"
This is the warmest entry point. It asks the kid to share a story instead of receive a lecture. Most kids have a story. The model said something wrong, or invented a thing, or got their name backwards. The kid laughs about it. That story is the foundation of the rest of the conversation, because the kid has just told you, in their own words, that the model can be wrong. You don’t have to argue the point.
Opener 2. "If I asked you to explain what ChatGPT actually does, in three sentences, could you?"
This works for slightly older kids, around 11 and up. It’s a low-stakes challenge. The kid tries. They usually get part of it right and part of it wrong. You don’t have to correct them point by point. You can say "huh, I would have said something like..." and offer a version. The conversation becomes two people trying to put words on a thing, which is the conversation you want.
Opener 3. "Show me something you made with it."
The best AI conversations with kids start with their work, not your worry. From the cohort intake calls
This is the one that does the most work, by a wide margin. If your kid has used AI to make anything, ask to see it. A poem, an image, a school assignment they used it on, a chat thread they thought was funny. You’re not auditing. You’re curious. Looking at their work together is the move. You see what kind of AI they’re actually using and how they’re using it. They feel seen instead of investigated. The conversation about “is this the right way to use it” can happen, but it happens with a concrete artifact in front of you both, not in the abstract.
One opener to skip.
The one that consistently backfires: "Are you worried about AI?"
This is a parent-anxiety question dressed up as a kid-anxiety question. Most kids aren’t worried about AI in the abstract. They’re using it. Asking the question telegraphs that you are worried, which puts the kid in the position of either reassuring you or absorbing your worry. Both moves shut the conversation down.
If you genuinely are worried about something specific, name the specific thing. "I read an article about kids using AI to write papers and I’m curious how that lands at your school" is a real question and a kid can answer it. "Are you worried about AI?" is a feeling looking for a target.
When you feel out of your depth, look at something together.
The bedrock advice we give parents is this: the conversation goes better when you have a concrete artifact in front of you both. AI is too abstract to talk about in the abstract. A page on a screen is easier.
Inside Tell and Show, the artifact is the tool-trace pane. When the studio’s AI partner (we call it Inkie) makes a change to your kid’s project, the trace shows each step the model took. reading icarus.html, proposing change to player.update(), adding function checkSunDamage(). You can sit next to your kid and read the trace together. Ask what each step did. The kid will know, or they won’t, and either case is a useful conversation. The Long & Magerko AI-literacy framework calls this competency "understanding what AI is doing," and watching a real trace together is the most efficient way we know to develop it.2
If you don’t use Tell and Show, the equivalent move is to sit next to your kid while they use whatever AI tool they normally use and just watch. Ask why they typed what they typed. Ask if the answer is what they expected. Don’t correct. Don’t lecture. Watch. The MIT RAISE group’s K-12 AI curriculum makes the same point: co-use, not surveillance, is how parents do useful AI-literacy work.3
You don’t need to be an AI expert. You need to be a parent who is willing to look at one concrete thing with their kid and ask one honest question. That’s the whole job. Everything else is footnotes.
References
- Stefania Druga et al., "Inclusive AI literacy for kids around the world," FabLearn 2019, ACM, 2019. See also Druga’s ongoing Cognimates project at cognimates.me, which has run dozens of family AI workshops since 2018.
- Duri Long & Brian Magerko, "What is AI Literacy? Competencies and Design Considerations," Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems, ACM, 2020.
- MIT RAISE initiative, MIT Media Lab. See raise.mit.edu for the K-12 AI curriculum and the “Day of AI” materials, which emphasize family co-use as a pedagogical strategy.