All posts

The "how do you know?" question is the foundation of most verification habits. Ask it consistently and a child eventually asks it automatically. Ask it the wrong way and you get an eye roll, a defensive shrug, and a conversation that goes nowhere useful.

The difference between those two outcomes is almost entirely in the delivery, and the delivery is something families can get right from the start, rather than discovering through a pattern of frustrating dinner table arguments.

Why the question is powerful

"How do you know?" does several things at once when it lands well. It assumes the child has done some thinking, rather than accusing them of not thinking. It asks for evidence rather than judgment. It invites reflection rather than demanding defense.

A child who is regularly asked "how do you know?" starts asking it internally, before they've said anything out loud. That internal check (the self-directed version of the question) is exactly what metacognitive oversight looks like from the inside. It's the goal.

The question works for AI-generated claims, but it also works for information from teachers, from friends, from news, from books, and from family members. It's not an AI-specific tool. It's a thinking habit that AI just makes unusually urgent to build.

The lecture failure mode

The question fails when it becomes a test. "How do you know?" said with the tone of someone who already knows the child can't answer correctly becomes an indictment rather than an inquiry. Children pick this up instantly. They stop answering the question honestly and start managing the interaction -- giving whatever answer seems least likely to produce a lecture.

The version that works sounds genuinely curious, not performatively skeptical. "Oh interesting, where did you find that?" is the same question without the adversarial edge. "That's a big claim. What's it based on?" accomplishes the same thing at a slightly higher sophistication level. The actual words matter less than the genuine interest they imply.

If you don't know the answer yourself, say so. "I'm not sure either -- let's see if we can find out" is an honest answer that models exactly the right behavior: curiosity rather than certainty, searching rather than asserting.

Age calibrations

For children between roughly seven and ten, the question works best as a game with a shared answer. You ask, and then you find out together. The child isn't expected to already know the answer or already know how to check -- the point is simply experiencing the habit of checking.

"ChatGPT said penguins live in the Arctic -- is that right? Let's look." (It's wrong -- penguins live in the Southern Hemisphere, not the Arctic.) That fifteen-second interaction teaches the question and answers it in a way that's memorable, and the child gets to be the one who caught the error.

For children between ten and thirteen, the question can include some expectation of follow-through: "How would you check that?" as a follow-on to "how do you know?" The two together start building the actual verification process, not just the instinct that checking matters.

For teenagers, the question can be asked with the expectation that they'll do the checking independently and report back. At this stage, "how do you know?" functions more like professional accountability -- the parent checking in on whether their child is applying their own standards -- and less like direct teaching.

Building the reflex

A reflex is built through repetition at low stakes. The goal is not to ask "how do you know?" only when the child has said something wrong, or only about AI specifically. Ask it when you're genuinely curious. Ask it when the child makes an interesting claim you want to know more about. Ask it when AI said something correct and you just want to verify it together.

The habit solidifies fastest when it doesn't feel like surveillance. If the question only appears when a mistake needs to be corrected, children learn to associate the question with criticism. If it appears when things are going well too, they learn to associate it with being taken seriously as a thinker.

One practical structure: once a week, after a meal or at another natural family checkpoint, ask about one thing anyone learned that week and follow it with "how do you know?" Not as a challenge -- as genuine interest. That weekly pattern, maintained over months, builds the habit more reliably than any intensive exercise.

The deeper purpose

"How do you know?" is ultimately about intellectual humility and intellectual honesty, which are not the same thing as skepticism or doubt. A child who asks "how do you know?" of themselves and others is not a contrarian. They're someone who cares about the difference between believing something and knowing something.

In a world where confident-sounding claims arrive from AI, from social media, from political messaging, and from advertising at a volume that has no historical precedent, that distinction is not an academic nicety. It's a navigational tool. Building it as a reflex, through a simple question asked consistently over years, is one of the most durable things a family can offer a child.