Start by finding out what actually happened, before you write to anyone. That is not disloyalty. It is the order that works, because the evidence that clears your child is the same evidence that would convict them: the drafts, the notes, the version history, and whether they can talk you through their own argument tonight. A detector score settles nothing on its own.
I spent years in secondary schools, including as an assistant headteacher, so I have sat on the school's side of this conversation. What follows is what I would do at home, in order, and what I would ask the school for.
Can an AI detector prove my child used AI?
No. It can produce a probability about a piece of text, and that is a different thing from evidence about a child.
The company behind the most widely used tool says so itself. In a post published in March 2023 and still on its site as I write, Turnitin describes a false positive rate of "less than 1%" in its own testing, and then says plainly that "Turnitin does not make a determination of misconduct", and that educators must "apply your professional judgment, knowledge of your students, and the specific context surrounding the assignment". That is the vendor telling schools the score is not the verdict.
Other detectors have done worse, and the failure is not random. In a Stanford study by Weixin Liang and colleagues, published in the journal Patterns in July 2023, seven widely used GPT detectors were run over 91 human-written TOEFL essays by non-native English speakers and 88 essays by US eighth graders. The detectors were close to perfect on the eighth-grade essays. On the TOEFL essays they were wrong more often than they were right, with an average false positive rate of 61.22%, and all seven agreed in wrongly flagging 18 of the 91. Those detectors were not Turnitin, and the study is now three years old, so do not carry the 61% figure into a conversation about a specific tool. Carry the shape of it: these systems key on how predictable the writing is, which the authors link to limited linguistic variability, so the children who write in a second language, or write plainly, are the ones who pay for it.
In England, the exam boards reached the same conclusion from the other direction. The Joint Council for Qualifications guidance on AI use in assessments, in the version dated April 2025, tells teachers that detection tools "vary in accuracy depending on the AI tool and version used, the proportion of AI to human content, prompt types and other factors", and that they should establish authenticity by comparing the work against what the student has produced before. It is worth reading the list of indicators it gives teachers, because one of them is "a default use of American spelling, currency, terms". A British child who reads a lot of American writing is not cheating. That is the level of certainty the system is actually working at.
An AI detector does not know your child. It has read one piece of writing and nothing else. It cannot see the two evenings at the kitchen table, and it cannot see the part where they gave up and started again.
What to do tonight, before you email anyone
Ask one specific question, and ask it without the accusation in your voice. "Did you cheat?" gets a denial from an honest child and a denial from a dishonest one, so it tells you nothing.
You could say: "The school thinks part of this was written by AI. I'm not angry and I'm not assuming. I need to know exactly what you typed into anything, and what came back, so I can help."
Then run the check that will tell you more than any score can.
The Show Me Again Test is what I would do before I wrote a single word to the school: ask your child to talk you through the piece of work, out loud, without the document open. Not the sentences. The choices. Why that example, why that order, what they cut and why. A detector produces a probability about a text. A child explaining their own thinking produces something a school can actually weigh.
A child who did the work will be able to do this, sometimes badly, usually with a lot of shrugging. A child who did not will run out of road in about ninety seconds, and you will both know it. This is the same thing good teachers have always done, and it is the third of the three checks I describe in the piece on where help ends and cheating begins: what did my child produce, how did they produce it, and what can they now do.
While you are there, gather the working. If your child wrote in Google Docs, the drafting history is already saved: open the file and click the "Last edit" link at the top, and Google will show the earlier versions and who made each change. Microsoft Word has an equivalent. Two honest limits, because version history is offered all over the internet as the magic answer and it is not. Google says you need edit permission on the file to browse earlier versions, so if the work sits in a school account your child may not be able to open the history themselves. And a document that appears in three large pastes is not proof of anything either, because plenty of children draft on paper, or on their phone, and paste it in.
One thing you will probably try tonight and should know about first: you cannot simply open your child's chatbot history and read it the way you are imagining. On an account you set up and linked, you get some visibility. On one your teenager made alone, you get none, and that is a limit of the product rather than a sign of anything. I have set out what a parent can and cannot see separately, because families keep discovering it in the middle of exactly this argument.
What to send the school, and what to ask for
Keep it short, unemotional and specific. You are not arguing that AI detection is flawed in general. You are asking one school to show its working on one piece of work.
You could write: "Thank you for telling me. I have talked to him and gathered what we have. Before we go further, could you tell me what the concern is based on, besides the detector score, and what the school's process is from here? I can send his version history and his notes, and he is happy to talk the essay through with you."
Three things that email does. It accepts the school's right to ask. It separates the score from the rest of the evidence. And it puts your child forward for the conversation rather than hiding him from it, which is the single move most likely to end the whole thing that week.
Ask for the written policy too, if you have not seen it. Under the exam boards' guidance a suspected case in UK coursework or non-exam assessment is handled differently depending on whether the student has already signed the declaration of authentication, and after that point the school has to report the case to the awarding organization. In the US it will be your district's academic integrity policy and, where a grade or a disciplinary record is at stake, whatever appeal route sits behind it. Ask which one you are in before you assume you are in trouble. Most parents find out afterward that they were never in a formal process at all. You can also ask what the school holds on your child, which is the same request I set out in the piece on what you can actually ask a school for.
What I See in Schools
The pattern I see most often is that the accusation does not start with the detector. It starts with a teacher reading something that does not sound like the child they teach. The score gets run afterward, to confirm a feeling that was already there. That order is not unreasonable, and it is much closer to what the exam boards actually ask for than a school that runs everything through a checker. But it does mean the number is doing confirmation work rather than detection work, and a number fetched to confirm a judgment is a poor thing to rest that judgment on.
The second pattern, working on AI strategy with schools, trusts and districts, is that the policy stops exactly where a parent needs it to start. Almost every AI policy I read now says something like "misuse of AI will be treated as academic malpractice". Very few say who investigates, what counts as evidence, what the student may submit in their own defense, how long it takes, who tells the parents, or who you go to if you disagree. The sentence is there. The process underneath it is often not written down anywhere, which is why two children in the same school can get very different outcomes from the same accusation. This is the same gap I describe when a school announces a rule about AI: there is the policy on the website, and there is whatever happens when thirty children open a laptop and one adult is in the room, and it is the second one your child lives in.
The third thing, and the one that surprises parents most, is who is actually handling it. Not a panel. Usually one subject teacher and a head of year, working it out between lessons, with no training in this specific problem, trying to be fair to a child they mostly like. They are not building a case against your family. Most of these end in a conversation, and the conversation is the school doing it properly rather than the school closing in.
Where it goes wrong is almost always the same moment: the score gets passed to the parent as a number, on its own, with no explanation of what it is. A line like "the system says 87% AI" sounds like the reading off of a fact. It is a model's guess about a text, and once a family has been handed it as a fact, everybody starts arguing about the wrong thing.
What if your child did use it?
Then the useful conversation is a different one, and it is better to have it early than to win an argument you should have lost. This is common ground, not a rare disgrace: Common Sense Media's census of 1,204 US children aged 9 to 17, with fieldwork from March 18 to 26, 2026, found that 85% of those who use AI use it for schoolwork, and that 73% said their school had told them what they should and should not use it for. The rules exist. Whether your child could state them tonight is a different question.
Find out which part the machine did, because that decides how serious this is. A child who asked a chatbot to fix their punctuation and a child who submitted its essay have done two different things, and most school policies, if you read them, are trying to say so. Ask your child to tell you, then tell the school before they find the rest of it. A family that comes forward almost always gets a smaller outcome than one that is caught twice.
Two things not to do. Do not write the apology for them. And do not let a teenager confess to something they did not do in order to end an uncomfortable evening, which is a real risk once an adult has made it clear that admitting it is the fast way out. If your child keeps saying they did not do it, keep believing that is possible, and go back to the working.
If they are badly shaken by this, and some children are, tell the school's pastoral lead or counselor rather than only the subject teacher. An accusation of dishonesty lands harder on a child who has never had one than most adults expect.
What changes with age
Under 8. This is rare, and when it happens it is usually a misunderstanding about who helped, not an accusation of cheating. Treat it as a question, not a charge. You could say: "Your teacher wasn't sure which bits were yours. Let's tell her together which bits I helped with."
8 to 12. At this age children often cannot explain the difference between asking a chatbot to check something and asking it to write something, because nobody has drawn that line for them. Do not ask whether they used AI. Ask what they typed, word for word, and what came back. You could say: "Show me the actual thing you typed. I'm not cross about you using it. I need to know what you asked it to do."
13 to 15. The stakes here are trust more than grades, and the risk runs in both directions: a wrongly accused teenager who feels you doubted them, or a rightly accused one who learns that denial works. Keep the conversation about the work rather than their character. You could say: "I'll go in and ask them to show me what this is based on. In return I need you to be straight with me tonight, because I can't help you if I find out something later."
16 to 18. The penalties become formal, and they attach to coursework, GCSE and A level non-exam assessment, graded assignments and college applications. Your teenager should read the actual rule themselves and lead the reply, with you beside them. You could say: "Find the exact wording in the policy, and we'll write the response together. It's your work and your name, so it should be your answer."
When the school's process comes first
Almost always, at the start. The school sets the rules on its own assessment, and pushing past that on day one usually costs you the goodwill you will need later. Four situations to treat differently:
- If it touches a formal qualification, such as UK coursework once the authentication declaration has been signed, the school may have no discretion left and the awarding organization decides. Ask early which stage you are at.
- If your child has an education, health and care plan, an IEP or a 504 plan, raise it through that route as well. Assistive tools that read, predict or correct text can change how a piece of writing looks, and the people who agreed those tools should be in the conversation.
- If a sanction is proposed that follows your child, a grade of zero, a record, a removal from a course, ask for the appeal route in writing before you respond to the substance.
- If English is an additional language for your child, say so plainly and early, because that is the group the research above shows these tools misjudge most.
None of this is a reason to treat the school as an opponent. In my experience the schools that handle these cases best are not the ones with the best detector. They are the ones that decided in advance what evidence they would accept, and told families what it was.
The next step
If your child's school has an AI line in its policy and nothing written underneath it about how a case is actually handled, that gap is the thing to raise, and it is the work I do with schools: helping them decide what counts as evidence, what a student may put forward, and what parents get told. It is worth sending to your head of year or principal, and it is a far better conversation to have before a case than during one.
If it helps to know what is coming before it arrives at your kitchen table, I write a Sunday newsletter that keeps more than 44,000 teachers and parents up to date without the panic. Parents are very welcome.
Dan Fitzpatrick is a former secondary teacher and assistant headteacher who now works with schools on their AI strategy. Read Dan's story.
