How do teachers know if you used ChatGPT? What to tell your kids

Short answer: not mainly by software. We know because the work does not sound like the child who handed it in, because it contains mistakes a person would not make, and because we ask the child to talk us through it and they cannot. Detection tools exist, and some schools use them, but they are the least reliable part of the picture.

I say "we" because I teach. This is the question my students ask me most often, usually in a slightly nervous voice, and I always give them the same answer. So let me give it to you, and then give you the sentence to say to your teenager tonight.

What the detectors actually do

Turnitin, the plagiarism service most schools already pay for, added an AI-writing detector. On its own pages, the company says the detector is about 98% accurate with fewer than 1% false positives on documents that are more than 20% AI-written. It deliberately shows no score at all when the estimate lands between 1% and 19%, because at that end the false positives multiply. And Turnitin's own guidance is blunt: the score must not be the sole basis for action against a student.

That last line is the one that matters, and it is the company saying it, not me.

Independent researchers have found higher false-positive rates than the marketing numbers suggest. Worse, a 2023 Stanford study found that AI detectors flag writing by non-native English speakers far more often than writing by native speakers. Think about what that means in a South African classroom, where a child may be writing their matric essays in their second or third language.

So if a teacher tells you their software says your child used AI, the honest reply is: that is a signal to have a conversation, not evidence. Ask what else they noticed.

What teachers actually notice

Here is the part your child does not expect. Software is not how most of us catch it.

The voice changes. I have read that child's work all year. I know their sentence rhythms, the words they overuse, the joke they cannot resist, the comma splice they have not shaken since Grade 8. Then a paragraph arrives that is smooth, balanced, slightly corporate, and utterly anonymous, and it sits in the middle of their own writing like a stranger at the dinner table. That contrast is visible from across the room.

The errors are the wrong kind. Human beings and machines fail differently. A struggling student gets the date of the Treaty of Versailles wrong. An AI invents a source. It cites a book that does not exist, attributes a real quotation to the wrong person, describes an experiment that never happened, or gives a beautifully written analysis of a poem it has clearly not read. Generic examples are another tell: when a paragraph about local water scarcity could have been written about anywhere on earth, nobody was actually thinking about the place.

The work does not match the class. We spent three lessons on a framework and the essay uses a completely different one, with vocabulary we have never used. That is not a child growing. That is a child borrowing.

And then we ask. This is the one that settles it, and it takes ninety seconds. "Interesting point in your third paragraph. Tell me more about that." A child who did the work lights up. A child who did not goes quiet, looks at the page, and starts reading their own essay back to me.

The explain-it-back test is the same test we use

This is the good news buried in all of it, and it is the reason I am not particularly frightened of AI in my classroom.

The test a teacher applies is exactly the test a parent can apply at the kitchen table, and it needs no software, no subscription and no surveillance.

It is not an interrogation. Ask it about work you are sure they did themselves, too, so it becomes a normal part of how your family talks about school rather than a trap that springs only when you are suspicious.

It also maps straight onto the V.E.T. rule from Chapter 10: Verify what the AI said against another source, Explain it in your own words, Transform it into your own work. The E is the middle of the rule for a reason. It is the step that turns reading into learning, and it happens to be the step that makes the teacher's question harmless.

What to tell your teenager

Here is the sentence. You can borrow it exactly.

"Use it as a tutor and you will never have to be afraid of the question."

That is the whole thing. If your child uses AI to quiz them, to explain a concept a third way, to find the weakest link in their argument, to check their working, then when a teacher says "talk me through this", they can. They have nothing to manage, nothing to remember, no story to keep straight. They just know the work.

If they used it to produce the work, they are now carrying a small secret around school, and the cost of that is not really the detection. It is the low hum of anxiety, and the fact that they will sit an exam in a hall with no laptop in a few months' time and discover what they did not learn.

I put it to my classes as a practical matter rather than a moral one, because teenagers are allergic to sermons but very responsive to self-interest. Nobody can catch you if there is nothing to catch.

Worth saying to them too: being honest about how you used it is almost always cheaper than being caught hiding it. Most teachers I know respond well to "I used ChatGPT to help me structure this, here is what I changed", and very badly to being lied to. If you want the longer version of where the line actually sits, I wrote about that in is using AI cheating.

Put it in writing before you need it

The Family AI Agreement in Chapter 10 has a section on exactly this, and it is the one I would not skip. Agree, in advance and in your own words, what honesty about schoolwork means in your house. The template puts it like this: we are honest about when AI was used, we do not have to flag every sentence, but we do flag the load-bearing ones. For school, that means telling the teacher when AI helped with the structure of an essay, and never pretending AI-generated text was ours.

Write that line down together, while nobody is in trouble. It takes five minutes with the free agreement builder, and the conversation you have while filling it in is worth more than the document.

Questions parents ask

Can teachers really tell if you used ChatGPT?

Often, yes, though usually not through software. Teachers read a child's writing all year and notice when the voice changes, when the errors are machine-shaped such as invented sources or generic examples, and when the work does not match what was taught in class. The decisive check is asking the student to explain their own work, which takes about ninety seconds.

Do schools use AI detectors, and are they accurate?

Some do, most often the AI detector built into Turnitin. Turnitin claims roughly 98% accuracy with under 1% false positives on documents more than 20% AI-written, and says the score must never be the sole basis for action. Independent research finds higher false-positive rates, and a 2023 Stanford study found detectors flag non-native English speakers far more often.

What happens if my child is wrongly accused of using AI?

Stay calm and ask what evidence exists beyond a detector score, because the detector's own maker says a score is not enough on its own. Offer what actually proves authorship: draft history in Google Docs or Word, notes, and above all your child's ability to discuss the work. Encourage your child to keep drafts from now on. It is the simplest insurance there is.

Is using AI for homework always cheating?

No. Using it to explain a concept, quiz you, check your working or challenge your argument is tutoring, and every school I know permits it. Using it to produce the work you hand in as your own is cheating, whatever the policy says, because you have skipped the learning the task existed to create. The test is whether you can explain what you submitted.

Sources

• Chapter 11 of Raising AI-Ready Children on academic honesty as an ethical choice rather than a rule to be dodged.
• Chapter 10 on the V.E.T. rule and the Explain It Back test; Chapter 17, Concern One, on what the research says about cheating.
• Chapter 10 on the V.E.T. checklist, the 20-Minute Rule and the Family AI Agreement.
• Turnitin's own published guidance on its AI writing detection, including its accuracy claims and its warning that the score must not be the sole basis for action, checked on the date at the top of this post.
• Stanford University, 2023, study finding that AI detectors disproportionately flag writing by non-native English speakers.