QUICK ANSWER
AI detectors are not reliable enough to settle the question, and a wrong accusation costs trust you will want later. A comprehension check takes about ninety seconds and tells you more. Ask your child to explain the argument, define a term they used, and say why they chose that example. A student who understands the work can do all three. One who handed in something they do not understand stalls on the second question.
Something about the paragraph on the screen does not sound like your child. The sentences are longer than usual. A word shows up that you have never heard them say out loud.
Then comes the hard part. You want to know what happened, and you can already picture how it goes if you ask straight out.
There is a better way in, and it starts with what the work says rather than who typed it. Here is what the evidence shows about detection, and what to ask instead.
PART 1: WHAT DETECTION CAN AND CANNOT TELL YOU
Can you tell whether your child used AI on an assignment?
Not reliably. The tools built for this job come with real limits, and the companies behind them say so.
- OpenAI pulled its own detector. The AI Text Classifier went offline on July 20, 2023, and OpenAI said it was “no longer available due to its low rate of accuracy.” At launch, it caught 26 percent of AI-written text and wrongly flagged human writing 9 percent of the time.
- Independent testing found the same pattern. Researchers who tested 14 detection tools in 2023 reported that the tools are “neither accurate nor reliable.” No tool scored above 80 percent accuracy. When the AI text was paraphrased, accuracy fell to 26 percent.
- Turnitin says its score is not proof. The company’s own guidance says the model “may not always be accurate” and “should not be used as the sole basis for adverse actions against a student.” Scores under 20 percent get an asterisk, because false positives are more common down there.
- Some universities stopped using detection. Vanderbilt disabled Turnitin’s AI detector in August 2023, citing false positives, a lack of transparency about how the tool works, and evidence of bias against students who are not native English speakers.
- Newer tools are better at some things. A June 2026 study of four detection tools found false positives were rare, which is real progress. Accuracy still varied widely by tool, and one tool missed every fully AI-written paper in the test. The researchers say detection belongs in a broader review, never as the only evidence.
Put together, a detector gives you a probability. Your question needs an answer.
Should you just ask the teacher?
Sometimes, yes, and it helps to know what you are asking for. A teacher who has read your child’s writing all fall has useful context that no tool has. Ask what they noticed rather than asking them to run a check. A question like “does this read like her other work to you?” gets you further than “can you scan it?”
One more thing worth knowing: if your child drafts in Google Docs or Word online, the file keeps a version history. Watching a draft get built, sentence by sentence, tells you more than any score. Look at it together rather than behind their back, which keeps this a conversation instead of an investigation.
What detectors claim and what testing shows
| The question | What independent testing shows | Source |
|---|---|---|
| Can a tool prove my child used AI? | No. Tool makers and researchers agree a score is a flag, not evidence. | Turnitin guidance; IJEI 2026 |
| Are the tools accurate? | In 2023 testing, no tool topped 80 percent, and accuracy fell to 26 percent on paraphrased text. A 2026 test found accuracy still varies widely by tool. | IJEI 2023; IJEI 2026 |
| Do they flag innocent students? | Older tools flagged 61 percent of essays by non-native English writers as AI. Newer tools flag fewer, but false positives still happen. | Liang et al., Patterns, 2023 |
| Does a low score mean it is fine? | No. Turnitin marks scores under 20 percent as less reliable, and edited AI text is harder to catch. | Turnitin guidance |
Why does a wrong accusation cost more than it catches?
Because the student who did the work pays for it. Accusations are not rare, either.
- In a national survey of teachers published in October 2025, 38 percent said a student at their school had been accused of using AI to cheat when it could not be proven, or when it later turned out the student had not.
- Stanford researchers found that seven detectors flagged 61 percent of practice essays written by non-native English speakers as AI-generated. Essays by U.S. eighth graders were judged almost perfectly. Students still building their English took the hit.
- NPR reported in December 2025 on a high school junior accused three times over work she wrote herself. She now spends an extra half hour per assignment rewriting anything a detector might flag.
At home, the cost lands somewhere specific. A student who did the work and gets accused anyway learns one lesson very quickly, which is to stop showing you their work. That is the thing you need most in high school, and it is hard to get back.
PART 2: THE NINETY-SECOND CHECK
What should you ask instead?
Ask about the work, not about who wrote it. These five questions take about ninety seconds, and they work on any assignment in any subject.
- “Explain the argument to me in two sentences, like you are telling a friend.”
- “Define this word for me.” Pick a term from their own paper.
- “Why did you choose this example?”
- “What did you leave out?”
- “What would you change if you had one more day?”
Three things make this work:
- Tone matters more than wording. Curious sounds different from suspicious, and your child can tell the difference in the first three words.
- Timing matters too. Sunday afternoon goes better than 10 p.m. the night before something is due.
- Their own words are the test. A student who built the argument can talk about it without the paper in front of them.
- Five is enough. Stop when you have your answer. This is a check, not a deposition.
What do their answers actually tell you?
Look at the pattern across all five, not any single answer. Nerves can make a student who did the work sound shaky on the first question.
What their answer tells you
| What you hear | What it usually means | Your next move |
|---|---|---|
| Explains it in their own words, stumbles here and there | They wrote it, or they used help and understood it | Say so. Ask what was hardest, and leave it there. |
| Summarizes well, cannot define a term from their own paper | They may have used AI and skimmed the result | Ask where the word came from. Work through the definition together. |
| Cannot say why an example is in there | They did not build the argument themselves | Go back to what the assignment was asking for, then have them rebuild that section. |
| Explains everything except one section that sounds different | Part of the work may not be theirs | Ask about that section alone. Keep the rest out of it. |
| Gets defensive before you finish the question | Could be guilt, could be a rough week | Stop the quiz. Ask how the class is going instead. |
This is a signal, not proof. A comprehension check tells you whether your child understands what they turned in, which is the part that matters for the next test anyway.
What if the answer is yes?
Separate the tool from the dishonesty. Using AI is not the problem in every class. Turning in work that is not yours, in a class that forbids it, is.
- Start with what the assignment was for. The essay was practice at building an argument. Skipping the practice costs the skill, and the grade is only part of it.
- Check the class rule before you decide anything. Some teachers allow AI for brainstorming or checking work. Others allow none of it.
- If a rule was broken, your child tells the teacher. A short script helps: “I used AI on part of this assignment and I should not have. I would like to redo it.”
- Let them make the call, with you nearby. Handling it themselves rebuilds more than a rewritten essay does.
- Keep it off school nights when you can. Nothing useful gets decided at 10 p.m.
One more decision belongs to you, and it is worth making on purpose: whether this becomes a punishment or a repair. Redoing the work, in their own words, teaches the thing the assignment was for. Losing the laptop for a week teaches something else, and it is usually about you rather than the work.
If this is the second or third time, treat the pattern as information. Something is making the honest version of the work feel impossible, and the last section of this post covers the usual culprits.
PART 3: SETTING YOUR OWN RULE
How do you set a family AI rule that holds when you are not in the room?
Keep it to three lines, short enough that a 13-year-old can repeat them. A rule your child can recite in the moment beats a long policy nobody remembers.
- What AI may do. Explain something after you have tried it, quiz you, or help you plan.
- What AI may not do. Write any part of what you hand in as your own.
- When to say you used it. Whenever the teacher asks, and whenever you are unsure.
Read it out loud together. If your child cannot repeat it back, it is too long.
Grade level changes what belongs in that middle line. A seventh grader and a high school senior need different permissions, and we cover that in part two on October 4.
What if the school has not given you a rule?
Then the syllabus is usually the operative rule, and a short email settles the rest. Most teachers have not been handed a policy either. In a 2026 Gallup survey, only 18 percent of teachers said they had received formal guidance from school leaders on how AI should be used, which we covered in our September 20 post on state tech policy.
Two sentences to a teacher usually does it:
Does your class allow students to use AI tools for any part of the work, and if so, should they say when they used one? I want to make sure we have the rule right at home.
| CONNECTICUT AND NEW YORK FAMILIES
Neither state requires districts to adopt an AI policy, so the rule your child follows is usually set by the individual teacher. Connecticut’s Commission for Educational Technology has published AI guidance for schools, though following it is up to each district. Families in Greenwich, Darien, Westport, New Canaan, Scarsdale, Chappaqua and Armonk should all start in the same place, which is the course syllabus. Our September 20 post covers where both states stand on phones and AI. |
When is AI use a symptom rather than the problem?
A student who reaches for AI every night is usually stuck on something. AI use itself is now ordinary. In a Pew Research Center survey published in February 2026, 54 percent of teens said they had used a chatbot for help with schoolwork, and more than 40 percent had used one for research or math problems.
Nightly use tends to trace back to one of three things:
- A content gap. Math especially. A student who missed a unit in the fall cannot solve the November problem set, and AI turns a two-hour wall into a ten-minute workaround that teaches nothing.
- An overloaded schedule. Five AP classes, a sport and a job leave no room for the slow version of anything.
- Trouble getting started. Staring at a blank page is an executive function problem, and AI is the fastest way to make the blank page go away.
The fix depends on which one you are looking at. A content gap needs academic tutoring to close. Trouble starting, planning and finishing is what executive function coaching is built for. Either way, the conversation goes better once you know which problem you have.
Frequently asked questions
Can teachers really tell if a student used AI?
Not with certainty. Teachers who know a student’s writing often notice when something reads differently, and that instinct is worth something. Tools cannot confirm it. Turnitin’s own guidance says its score should not be the sole basis for action against a student.
Are AI detectors accurate?
They are inconsistent. In 2023, a test of 14 tools found none scored above 80 percent accuracy, and accuracy dropped to 26 percent when the AI text was paraphrased. Newer tools have improved on false positives, according to a 2026 study of four of them, but accuracy still varied widely and the researchers said detection should never be the only evidence.
What should I do if I think my child used AI on homework?
Ask about the work rather than the tool. Have your child explain the argument in two sentences, define a term from their own paper, and say why they chose an example. A student who understands what they turned in can do all three. The pattern across the answers tells you more than a detector score.
Is using ChatGPT for homework cheating?
That depends on the class rule. Some teachers allow AI for brainstorming, studying or checking work. Others allow none of it. Turning in AI-written work as your own is cheating in any class. Start with the course syllabus, then ask the teacher if it says nothing.
How do I ask my child about AI without starting a fight?
Ask about the content, not about their honesty. “Walk me through your argument” lands very differently from “did you write this.” Pick a calm moment rather than the night an assignment is due, and say up front that you are trying to understand the work, not run an investigation.
What should a family AI rule include?
Three lines: what AI may do, what it may not do, and when your child should say they used it. Keep it short enough to repeat from memory. Our October 4 post covers what to allow at each grade level.
Where do I find my child’s school AI policy?
Start with the course syllabus, which is usually more specific than the district handbook. If neither says anything, email the teacher. Neither Connecticut nor New York requires districts to adopt an AI policy, so in most local schools the rule is set class by class.
Next Sunday we turn the question around. Part two on October 4 covers what good AI use looks like at each grade level, and how to give permission that still protects the learning. On October 18, we look at what to do when the first progress report arrives and the numbers surprise you. For more on homework, focus and independence in the middle school years, start with our middle school guide.
