The short answer
- AI detectors aren’t the answer — they’re unreliable and produce false accusations. Don’t rely on a score.
- The most reliable sign is a gap: work that’s far more polished than your child’s usual voice, or homework that’s flawless while test scores lag.
- Ask them to explain it. A child who wrote it can talk you through the reasoning; one who didn’t, can’t.
- Look for the tells of unedited AI: generic confidence, no personal voice, invented “facts,” and a suspiciously even, listy structure.
- The goal is understanding, not a trial. You want to know so you can help — not to win a case.
Wanting to know whether AI wrote your child’s homework is completely reasonable. But the internet’s usual answer — “run it through a detector” — is the wrong one, and acting on it can do real harm. Here’s what actually tells you something, and how to use it in the service of helping rather than accusing.
Why not just use an AI detector?
Because they’re wrong often enough to hurt your child. AI detectors don’t identify AI; they estimate how predictable text is, and plenty of honest human writing is predictable. In a landmark Stanford study, detectors falsely flagged human-written essays by non-native English speakers 61% of the time on average. They also miss edited AI text entirely. A detector score is, at best, a weak hint — and at worst, it’s how a good student gets falsely accused. The full evidence is here, and it’s not close.
So put the detectors down. The reliable signals are elsewhere.
What’s the most reliable sign?
A gap between the work and the child. Two gaps in particular:
- Voice. Homework that’s dramatically more polished, formal, or fluent than how your child actually writes and talks. You know their voice better than any algorithm — trust that.
- Homework vs. tests. Spotless homework paired with sagging test scores is the classic tell, because the homework was the tool’s and the test is theirs. This one is especially hard to fake and easy to notice over time.
Neither requires you to analyze a single sentence. They come from knowing your child, which is the one thing no software has.
What are the textual tells?
Unedited AI writing has a signature — but editing erases it. For what it’s worth, raw AI output often shows:
| Tell | What it looks like |
|---|---|
| Generic fluency | Smooth, confident, and completely personality-free |
| No specific detail | No “in class we…”, no personal example, nothing lived |
| Even, listy structure | Suspiciously tidy — three points, three sentences, repeat |
| Hedging | “It’s important to note,” “there are many factors” |
| Invented facts | Confident claims or citations that don’t check out |
The catch: a student who edits the output, or asks for a “more casual” version, removes most of these. That’s exactly why textual analysis is a weak tool and explanation is a strong one — you can’t edit your way into understanding you don’t have.
What should you do if you’re fairly sure?
Treat it as a teaching moment, not a verdict. If the conversation reveals your child skipped the thinking, the useful response isn’t punishment — it’s redirection: show them the honest way to use the tool, which gets them the help without the hollowness. Most kids aren’t trying to be dishonest; they’re trying to be done. Pointing them at “use it to understand, then do it yourself” solves the actual problem.
And keep the stakes in perspective. One AI-written assignment is a habit to correct, not a character verdict. The goal — always — is a child who understands their work and is honest about how they did it. That’s built through conversations, not detectors.
Related: is using AI cheating? · safe AI tools by age · do AI detectors work?
Frequently asked questions
How can I tell if my child used AI to write something?
The most reliable approach isn’t spotting the text — it’s asking your child to explain the work to you. Someone who wrote it can walk you through their argument and choices; someone who generated it can’t. Beyond that, watch for writing far more polished than their usual voice, generic phrasing with no personality, and any facts or sources that turn out to be invented.
Can AI detectors tell me if homework is AI-written?
Not reliably. AI detectors estimate probability, not authorship, and they produce both false negatives and false positives — wrongly flagging human writing, especially from non-native English speakers. A detector score is at best a weak hint and at worst a false accusation. Don’t treat it as proof.
What are the signs of AI-generated writing?
Unedited AI writing often reads as fluent but generic — confident, even-toned, and personality-free. Watch for a lack of specific personal detail, an overly tidy list-like structure, hedging phrases, and factual claims or citations that don’t check out. But edited AI text can hide all of these, which is why explanation beats textual analysis.
Should I confront my child if I think they used AI?
Approach it as a conversation, not a confrontation. Ask them to explain the work and how they produced it. If they used AI to skip the thinking, that emerges naturally and calmly — and it becomes a chance to teach the honest way to use the tool, which is far more valuable than a fight over a single assignment.