Are they using AI, and would anyone know?
Two things are worth separating: how many students actually use these tools, which Australia now measures properly, and whether a school can tell, which is far less settled than either schools or detector vendors suggest. Both answers below come from documents you can open.
How common it actually is
ACARA’s national sample assessment of ICT literacy asked. This is a nationally representative Australian sample — 5,498 Year 6 students and 4,753 Year 10 students across 641 schools — rather than a vendor survey or an overseas figure.
One in four Year 10 students, and about one in ten Year 6 students. If your teenager is using these tools, they are in the ordinary middle of their cohort, not at its edge.
One caution that belongs with the number, and it is ACARA’s own: the 2025 cycle ran in May rather than October for the first time, and every trend table in the report carries a note that comparison with earlier cycles should be interpreted with caution. So this is a good picture of now, and not evidence of a rise.
Whether a school can tell — the part that is not settled
The intuitive assumption is that detection software resolves this. The evidence says it does not, and it fails in both directions at once.
light ‘refine abstract only’ edits, a proxy for guideline-compliant AI assistance, are flagged at 38 to 80% … unmodified 2023 to 2025 originals are flagged at 9 to 15%
Karr, Khvatskii, Hua & Chawla (2026), arXiv:2608.11256. Registered as FR-12.
Two implications, and they cut opposite ways. A student who used AI the way many school policies expressly permit — to refine something they wrote — can be flagged most of the time. And a student using commercial humanising software can pass. A flag is therefore not proof, and a pass is not innocence.
The limit on that study matters and should not be skipped. Its corpus is published academic abstracts, not school essays, and it is a preprint that has not been peer-reviewed. The false-positive rate on a Year 10 English response is genuinely unknown. What can be said is narrower than either a worried parent or a detector vendor would like: the case for detector reliability is weak, and it rests on a population that is not your child.
The more useful question
Not “did they use it”, which is increasingly unanswerable, but “can they do it without it”. Those come apart quickly, and only one of them is a problem.
A student who drafts an argument, then asks a model to tighten a paragraph, has done the thinking. A student who asks for the argument has not, and will discover the difference in an exam room with no device — which for VCE English is three hours, handwritten. The gap between those two students is invisible in a submitted document and obvious in a timed one.
What to do instead of policing it
- Ask them to explain the piece, not defend it. “Why is this your second paragraph rather than your fourth?” A student who wrote it can answer instantly. This is a better instrument than any detector and it does not accuse anyone.
- Watch what happens under time, on paper. One handwritten piece to time, occasionally, tells you what a folder of polished submissions cannot.
- Read your school’s actual policy.Most now permit some assistance and prohibit substitution. The distinction your child is being asked to observe is usually more specific than “don’t use AI”.
- Do not accuse on a detector result alone. On the best available evidence a flag carries far less information than its confidence score implies.
For the wider question of what these tools are doing in Australian schools, and what the research does and does not establish, AI in Australian schools is the longer treatment, and the AI entries in the register carry the sources.
The test is what they can write unaided.
A twenty-minute discovery call — an honest read of what your child can actually do on the page, and what would move it. Free, and without obligation.
Weekly one-to-one lessons, Years 3 to 12 — VCE English tutoring →