A few years ago, a neat, correct piece of homework told you something. It meant a student had probably sat down and worked it out.
Today it tells you almost nothing. Any chatbot can produce a correct maths worksheet, a well-structured essay or a working piece of code in seconds. The homework looks the same whether your child spent an hour understanding it or thirty seconds copying it.
So parents and teachers are asking the same question: how do you tell whether a student actually understands something, when AI can fake the evidence?
The answer is surprisingly old-fashioned. Ask them to explain it, out loud.
Quick Answer: You can't reliably catch AI cheating by looking at the homework, and AI detectors are too inaccurate to trust. What AI can't do is sit beside your child and explain their thinking for them. A short oral check, asking them to walk through how they got the answer, why each step works, and what would change if the question changed, shows within minutes whether they understand it. Knowing that they'll have to explain their work also changes how students use AI in the first place: they have to think it through themselves.
Why You Can't Spot AI Homework by Looking at It
The obvious answer is an AI detector, a tool that claims to tell whether text was written by AI. Unfortunately, they don't work well enough to trust.
- They get it wrong in both directions. They flag genuine student writing as AI, and miss AI writing that's been lightly edited. Even OpenAI, the company behind ChatGPT, withdrew its own AI-writing detector because it wasn't accurate enough.
- They're easy to dodge. Ask the AI to "write like a 14-year-old", or change a few words, and the score drops.
- They don't work on maths or code at all. A correct answer to 3x + 7 = 22 looks the same whoever wrote it.
Worse, a false accusation is a terrible outcome. A child who did the work honestly and gets told a machine says they cheated learns that honesty doesn't pay.
The real problem is that we're looking at the wrong thing. Homework was never the goal. It was evidence of understanding. Now that the evidence can be faked, we need to check the understanding directly.
What AI Can't Fake: The Explanation
As we explained in how ChatGPT actually works, a chatbot predicts likely text. It's very good at producing answers. But when your child is sitting at the kitchen table and you ask "why did you do that step?", the chatbot isn't there.
A student who understands can explain their thinking in their own words, handle a follow-up question, and adapt when something changes. A student who copied can usually repeat the answer, and then gets stuck on the first "why".
That's why oral exams, called vivas, have been used for centuries at universities: to check that a student really did the thinking behind their written work. It's exactly the right tool for the age of AI.
How to Run a Five-Minute Oral Check
You don't need to be an expert in the subject. You need curiosity and a few good questions. Pick one question from the homework and ask your child to talk you through it.
- "Walk me through how you did this." Let them explain from the start, in their own words. Don't help.
- "Why does this step work?" Pick one step and ask why. Not what they did, but why it's allowed or why it makes sense.
- "What if we changed this?" Change a number, a condition or a word in the question. Can they adapt, or does the whole thing fall apart?
- "Where could someone go wrong here?" A student who understands knows where the traps are.
- "Teach it to me." Pretend you don't know the topic, and ask them to teach you. Teaching is the hardest test of understanding there is.
Two or three of these is usually enough. You'll know within a few minutes.
An Example: Real Understanding vs Copied Answers
Here's the same homework question, checked with two different students. The question was: Solve 3x + 7 = 22. Both handed in the correct answer, x = 5.
Student A understood it:
Parent: Walk me through how you did this.
A: I wanted x on its own, so first I took away 7
from both sides. That gives 3x = 15.
Parent: Why both sides?
A: Because it's like a balance. If you only take it
from one side, it's not equal anymore.
Parent: What if it was 3x + 7 = 25?
A: Then 3x = 18, so x = 6.Student B copied it:
Parent: Walk me through how you did this.
B: You move the 7 over and then divide, and it's 5.
Parent: Why do you move the 7?
B: That's just how you do it.
Parent: What if it was 3x + 7 = 25?
B: Um... still 5?Same homework, same mark. Completely different understanding. No detector could tell you that. Two minutes of conversation did.
The same works for an essay ("Why did you choose this example? What would someone who disagrees say?") and for code ("What does this line do? What happens if I delete it?").
Why It Prevents Cheating, Not Just Catches It
The best part of an oral check isn't catching anyone. It's what happens once students know it's coming.
If your child knows that every so often they'll be asked to explain their homework out loud, copying an AI answer stops being useful. They'd still have to understand it to survive the conversation. So they start using AI the right way: asking it to explain a step they're stuck on, rather than to hand over the answer. Our guide to using AI for homework covers what that looks like.
The oral check forces the thinking, and the thinking is the whole point of homework.
Tips for Parents
- Make it normal, not an interrogation. "Show me what you did today" at dinner, not "prove you didn't cheat". Curiosity works better than suspicion.
- Do it regularly, not just when you're suspicious. Random and routine is fair, and it's what makes it work as prevention.
- Praise good explanations, not just right answers. "I love how you explained that" teaches them what you value.
- If they can't explain it, that's information, not a crime. Maybe they used AI. Maybe they just didn't understand it. Either way, the fix is the same: go back over it together.
- Know which AI tools they use. Our post on AI safety and supervision explains why being able to see your child's AI chats matters.
Tips for Teachers
- Add short oral checks to written work. Even two minutes with a few students per lesson, chosen at random, changes the whole class's behaviour.
- Mark the explanation, not only the answer. Ask students to present their working, or record a short voice note explaining one question.
- Use presentations and demos. When students must present and answer questions about what they made, there's nowhere to hide.
How We Do It at DotCode
Every one of our courses ends with a demo day, where each student presents the project they built and answers questions about it in front of the class. You can't present a game, a program or an AI model you don't understand, because the first question from the audience will find out.
In our AI & Python course, students go further: they build their own language model in Python, line by line, so they understand exactly what AI can and can't do, including why it can produce a perfect answer without any understanding at all.
Every course starts with a free trial lesson. Book one and see how your child explains what they built.
