It was 9pm on a school night. The assignment was due the next morning. We’d already had the heated family discussion, the one about why something that should have taken five hours was being started now. You’ve had that one.
My son disappeared into his room. Ten minutes later he was back. “Done.”
On first glance, it was genuinely good. Well-structured, clear argument, proper references. Part of me was impressed.
Then he walked me through how he’d done it. Step one: generate the content using AI. Step two: use a different AI to verify the output. He’d built prompts that pull references off the web to anchor the claims and check the facts. Step three: run the whole thing through GPTZero, the AI detector, and flag anything that scores too high. Step four: targeted edits, rewrite flagged sections in his own voice, and deliberately “dumb down” parts of it so it wouldn’t read as too polished for a Year 10 student.
That last step is the one I keep coming back to. Because ironically, it’s the closest thing to real learning in the whole process. He has to understand a concept well enough to simplify it, to re-level it for his own voice. There’s genuine thinking in that step. But the thinking is in service of making the deception more convincing, not understanding the topic.
And here’s what got me: the generation step, the “doing the assignment” part, took five to ten minutes. Making it look like his own work took another ten to thirty. More time on authenticity theatre than on the work itself.
So I asked him a few questions about the topic. Luckily it was one I knew well, so I could probe whether he understood the material.
That’s where it fell apart. He could give me the surface-level version, but push one layer deeper and the understanding wasn’t there. He’d produced an A-grade assignment without building A-grade knowledge. The output looked like learning but it wasn’t.
That’s what bothered me as I went to bed. I wasn’t worried about him getting caught, his de-AI workflow was better than most detection tools could handle. I was worried about what was happening underneath. Every time he takes that shortcut, he’s reinforcing a loop: skip the struggle, get the hit of I did a thing, move on. Quick, easy, dopamine. And the knowledge gap just gets wider.
Here’s what I’m hearing from parents right now, and I bet you recognise some of it in yourself.
Most parents are secretly in two camps at the same time. One part of you knows this is a shortcut, that “using AI to write your essay” is just a polished version of the thing you would have called cheating ten years ago.
But there’s another part, the part you don’t say out loud at the school gate, that’s a little bit relieved. The work looks good. The marks are fine. One less battle at the dinner table in a week that already has too many.
They’re getting good marks. They seem to know what they’re doing with the technology. Maybe this is just… how it works now?
I’ve felt that pull too. But it’s worth naming the conflict, because leaving it unnamed is how kids end up two years deep into a pattern that’s genuinely hard to reverse. The concern isn’t the marks, it’s that your kid is building a dependency where AI gives them the answer, their brain registers done, and they move on without ever wrestling with the material. Skip it enough times and you end up with a kid carrying a transcript full of good grades and a head full of gaps.
That’s a problem that doesn’t show up until it really shows up. In an exam they can’t use AI for, in a uni tutorial where they’re expected to think on their feet, in a job interview where someone asks them to explain their thinking.
Last edition I talked about slop, the gap between polished output and understanding. My son’s workflow is a perfect example of what slop looks like now: the end result reads well, but look closely and you’ll find subtle holes, errors that don’t quite track with the broader argument. The structure is strong but the understanding is thin. Not obviously AI-generated. Just quietly hollow.
Which means the old question, “does this look like AI wrote it?”, is a dead end. The question that works is what I think of as the slop test: can your kid defend what they handed in? Can they go deeper than what’s on the page? Can they surprise you with something they learned?
If they can, it doesn’t matter how much AI they used. If they can’t, the mark is meaningless.
Here’s the reframe I want you to sit with: your kid is demonstrating real skills in that workflow. Research, verification, editing, quality control. But all of it is pointed at making AI output look authentic, not at understanding the material. The sophistication is there. The direction is wrong. And that means the conversation is about what you should expect them to do with AI, not whether they should use it.
Think about it this way. If you hired your child a tutor, a really good one, available 24/7, who knew every subject, you wouldn’t be satisfied with the same marks they were getting before. You’d expect improvement, depth, and for them to come home and say something about the topic that surprised you.
AI is that tutor, except it’s infinitely patient and knows more than any single human could. So why are we still benchmarking against the old bar?
The question shouldn’t be “did my kid use AI to do their homework?” It should be “did my kid use AI to understand this topic better than they would have without it?”
Here’s what I want to see when my son shows me an assignment:
If your kid has AI and is producing quality work with no real insight, that’s the red flag.
These workflows are social, not solo. My son didn’t figure this out in isolation. Kids share AI workflows the way they used to share cheat codes. Which tools to use, which prompts work best, how to beat the detectors. If your kid has a system this sophisticated, assume their friends have a version of it too. This is a generation-wide shift in how homework gets done.
The reinforcing loop is the real danger. Quick shortcut → good mark → dopamine hit → repeat. Each cycle makes the next shortcut more automatic and the knowledge gap wider. This is the same pattern we saw with social media. By the time the habit is visible, it’s already entrenched.
Prompting sophistication is not the same as understanding. Being good at getting AI to produce good output is a skill, but it’s not the same skill as understanding the material. The test is always: can they go beyond what they prompted for?
“What’s something about [the topic they’re working on] that I don’t know, that you could teach me?”
This is what I think of as the Sergey Brin question. Don’t ask “what did you write about?”, that just gets you a summary. Ask “teach me something I don’t know.” That forces a different level of engagement entirely.
If they can teach you something specific that goes deeper than the assignment required, they did the work, regardless of how much AI they used.
If they can’t, follow up with this: “Walk me through how you researched this. What did you ask AI to help with?”
That’s where the truth lives. If their answer is basically “I told it to write me an essay on [topic]”, that’s the slop loop. But if they say they asked AI to explain a concept, then challenged that explanation, then explored a different angle and synthesised across sources, that’s a kid using the superpower properly. The AI was the tool, not the substitute.
For bonus points: “How could you apply what you learned here to something completely different?” If they can transfer the knowledge, they own it. If they can’t, they rented it for a night.
This one came from watching my own kid and being honest about what I saw. If your kid has a clever AI workflow, and they probably do, the goal is to make sure the workflow includes the part where they learn something. Share your own version of the 10-minute assignment story. I know you have one.
Next time: how to tell where your kid’s school sits on the AI curve, and what questions to ask at the next parent-teacher night.
The Parent Edit lands roughly every other week. No fluff, no AI hype, just the honest version of the conversation parents keep asking me for.
Get on the listOr forward this to someone who needs it.
What’s this looking like in your house?
Share your experience. Comments are moderated.