How Schools Detect AI Writing (And Why It's Complicated)

Quick answer

Schools detect AI writing through three layers: teachers who know a student's voice, detection software that estimates how machine-like a text reads, and - increasingly - the process behind the work: drafts, version history and a few direct questions. None of these is foolproof. Detectors miss AI writing that has been lightly edited, and they sometimes flag honest work, which is the more damaging failure. That is why schools are shifting weight from detection to process evidence, asking students to show how a piece of work came to exist. For students, the reliable strategy is not evasion. It is honesty: follow the task rules, disclose AI help where it is allowed, and keep the drafts that prove the work is yours.

The three layers of detection

The teacher's ear

An English teacher who has read a student's writing all year has a calibrated instrument no software matches. Vocabulary that jumps two year levels overnight, arguments that outrun anything the student says in class, a suddenly flawless command of the semicolon - these are the tells teachers actually notice. It is not proof. But it is usually where the question starts.

Detection software

Detection tools scan text for statistical patterns typical of writing produced by a large language model - the technology behind ChatGPT. Machine text tends to be smooth, evenly weighted and low on surprise, and the software estimates how strongly a piece resembles that pattern. The output is a probability, not a fact. We examine the reliability question properly in how reliable are AI detection tools; the short version is useful signal, dangerous verdict.

Process checks

The newest layer is the most old-fashioned: asking to see how the work was made. Drafts, version history, planning notes, and short conversations where the student talks through their choices. Some schools now build this in formally, requiring draft submissions or running a brief viva - a short spoken defence of the work.

Why detection is imperfect in both directions

Start with the misses. A student who generates an essay and spends twenty minutes roughing it up - swapping words, breaking sentences, adding a personal aside - will often slip under the software's radar. Anyone claiming a detector catches everything is selling one.

The false flags are worse, because they hurt honest students. Careful, formulaic writers - including many students who learned English as an additional language and lean on safe, textbook constructions - can read as machine-like to a detector. An honest student accused on the strength of a score alone learns a bitter lesson: that doing the work does not protect you. Few things corrode a school's integrity culture faster.

Both failure modes point to the same conclusion. A detector score is a reason to look closer, never a verdict on its own. Schools that treat it as a verdict are outsourcing a serious judgement to a tool that explicitly disclaims certainty.

The shift toward process evidence

Because detection cannot carry the weight, assessment design is quietly moving to something fairer: evidence of process.

For honest students, this shift is good news. A detector can wrongly accuse you; your version history cannot. A student with outlines, drafts and the ability to explain their choices is effectively accusation-proof, whatever a score says.

Why honesty is the safe strategy

The students who end up in real trouble are rarely the ones who used AI. They are the ones who concealed it on tasks where it was not allowed, then had no answer when asked to explain their own argument. Concealment compounds; disclosure resolves.

The practical playbook is short. Read the task rules, because they differ assessment by assessment. Ask the teacher before submitting, not after, when something is unclear. Disclose AI help in the format the school expects - our guide to how to reference AI in schoolwork covers the common conventions. And always be able to explain the work without the tool. A student who can do that has nothing to fear from any detector on the market.

What parents and schools can do

Parents: make draft-keeping a habit, not a punishment. Writing in a document that records version history costs nothing and quietly protects your child. And have the cheating conversation early, without theatre - the calm version is laid out in academic integrity in the AI era for parents and schools. For the bigger picture of raising a capable, honest AI user, start with AI education for teenagers in Australia.

Schools: treat detector output as a conversation starter with a defined process behind it, and design assessment so process evidence accumulates by default - drafts submitted along the way, a few minutes of oral questioning where stakes are high. It is fairer, and it is also better teaching.

The recommendation: stop asking whether the school can catch AI writing and start asking whether your child could explain their work aloud tomorrow. Teach them to work as if they will be asked - increasingly, they will be. Honesty plus a visible process beats every evasion strategy, and it builds exactly the habits that matter after school anyway.