You Cannot Check Everything

Verify Outputs is the principle people agree with fastest and abandon soonest. The reason is not laziness.

When we present our Five Principles, Verify Outputs is the one people agree with fastest and abandon soonest. The agreement is easy and genuine. Everyone says they are going to verify the outputs, check the answers, review the sources themselves. Rigorously. Every time. I have seen many such sandcastles, in professional adults.

When a student does not verify, the abandonment is not laziness either. Ask them to check everything a tool tells them and you have handed them a job they were never prepared for, on a deadline. A host of plausible answers, each one needing a second source to confirm, all before class tomorrow. The volume wins. It was always going to win.

We are the ones who did not build the habit

Before we say anything about the responsibility of a fifteen-year-old, let me make this point clear.

We spent a dozen years telling them we did not have time and to go look it up. We let the screen be the thing that kept them occupied, when they were unruly, in the car, in the waiting room, out to dinner, through the long weekends. All we needed was autoplay, the next episode already loading, and a stable connection, and there were no more tantrums. Left to their own devices, in the most literal sense the phrase allows.

Then the tools got good enough to answer in a paragraph instead of ten blue links, and we turned around and asked for scrutiny. A reflex we spent a decade not building, requested on short notice, about the one source that has never once looked unsure of itself.

That is not written to assign blame to the overworked, overcommitted, and overextended parents of the world. It is certainly not written to excuse any adult from acting now either. It is written because it explains why "just check it" fails as an instruction. You cannot ask for a habit as though it is already there.

What is actually at risk

Everyone has some version of the reflex. It is the half-second of "wait, what?" before you act on something. It sparks when a date sits wrong, when a quote is a little too perfect for the argument it supports, when a confident answer lands in a subject where you happen to know enough to be suspicious. You do not actually decide to have it. It arrives, and then you either act on it or you do not.

That reflex is not general diligence. It is a muscle, trained over years. It is built out of all the times a person was confidently wrong and found out about it. It is also the part that quietly stops working, because unlike the capacity to check, nobody notices when you stop checking with yourself. That is cognitive drift operating on a single habit.

So the version of the principle worth teaching is not the one that asks for more checking. A student needs something smaller and more durable than that. Not the capacity to interrogate every answer. The trigger that tells them which answer deserves it.

What the tool removes

Here is the part that usually goes unsaid. AI does not only make checking feel unnecessary. It removes the cues the trigger was trained on.

Think about where a student's information used to come from. A teacher who paused before answering. A classmate who said "I think the answer is X" and meant it. A book that hedged its bets, a website that looked wrong, one source that clearly disagreed with another. Every one of those carried a signal about how reliable it was. The trigger was built by that friction, and for this generation the friction was removed.

AI output has none. A sentence the model gets right and a sentence it invented both arrive in the same voice, the same authority, the same clean paragraph. Nothing is marked. So a student is not merely tempted to skip the check. They are working in an environment stripped of the cues that used to tell them a check was needed at all.

That is worth examining, because it means the erosion is not a discipline problem. A student can want to verify and simply not know where to aim.

What a parent or teacher can do

Not "check more." That instruction fails on contact with a real week, and we have already covered why we are in no position to issue it.

Ask what they did not believe. "What in here did you not fully trust?" is a better question than "did you check this," because it asks for the trigger rather than the labor. A student who can name one soft spot in their own work still has the capability. A student who says everything looked fine, every time, is telling you something.

We have to make the catch worth something, with a lesson and not a reprimand. A trigger only fires if acting on it is safe. If finding a problem late means an argument, a lost grade, or admitting the last hour was wasted, the teen reflex is to stay quiet. If a student who says "This didn't seem right" is treated as thoughtful and asked to explain, they learn to think through the answer themselves. That is most of the training, and it happens at a kitchen table or in a classroom rather than in any policy or platform.

And point it at their own work too. The reflex that catches a weak claim from a tool is the same one that catches a weak claim from themselves. It does not have a separate setting for AI. A student who learns to say when they believe something is wrong is not being argumentative. They are learning to decide for themselves.

The narrower thing

As an AI governance practitioner, I can tell you that anyone who says they never find an error with their AI is not lucky. They have stopped checking. For a student it is harder. They may not have stopped caring, and they may never have started. Either way they have stopped noticing, inside an environment built so that every answer looks equally solid.

You cannot check everything, and no honest version of this asks you to. What has to survive is the smaller thing: the instinct that still says wait. We did not give them much of a head start on it. That is a reason to begin, not a reason to expect it for free.

PS. I helped raise three kids to college. I was working so much that they all had smartphones and tablets and laptops long before they needed them, and long before they were done needing me.