Over the last few years, a central question has dominated education discourse in Australia: how do we stop students from using AI to write their essays? It’s a question worth asking, but the wrong one to build a strategy around.
You don’t have to look far to see this instinct in action. The NSW government recently confirmed that HSC students will be limited to a single take-home assessment task per course, worth no more than 15 per cent of their school-based mark, in order to curb AI-assisted cheating. The thinking behind it is that AI is believed to have a massive impact on how children learn and how their assessments are marked. Left unchecked, it offers students a short-cut to complex school work, bypassing the opportunity to build knowledge and learning.
It’s an understandable belief and it’s highlighting a very serious concern the sector already knows – unguided AI use allows students to skip the learning process. But responding by forcing every student back into exam conditions is a blunt instrument. Although it shrinks the space where misuse can happen, it pushes proof of learning into a single high-stakes task, and risks sidelining the assessments that build skills an exam can’t capture. It also fails to ask what students were actually looking for when they turned to these tools in the first place.
This approach assumes we already know what students want from these tools. However, when students are given guardrails on AI use, the results are surprising.
When institutions deploy AI chat tools they control – rather than leaving students to find ones they can exploit – the dominant pattern of interaction isn’t what many would expect.
Students aren’t asking for polished final pieces. They’re not saying “write this for me”. They want feedback and reassurance. They’re asking “is this good enough?”, “am I on track?”, “how can I improve?”.
When you think about it, we actually shouldn’t be surprised. This is nothing new. It’s the same pattern of student behaviour that has always existed, and every academic should recognise it instantly. What’s new is the interface.
Turnitin’s Learning Integrity Insights Report Q2 2026 revealed students were most likely to ask a question in the Turnitin Clarity AI chat when they were in the drafting stage, accounting for 43 per cent of prompts. This was ahead of more content-empty usages, such as follow-up, chit-chat, and technical questions (21 per cent), and proof-reading (18 per cent).
In the drafting phase, 72 per cent of student requests were for feedback, with prompts like “Is my paper good?”, “Am I aligned with the rubric?”, or “How can I improve my paper?”.
Even among content-empty interactions, only 11 per cent of requests were identified as misuse, such as “Write this for me”, while 52 per cent asked follow-up questions, such as “Is this improved?” and “What about now?”.
This kind of formative, in-the-moment reassurance-seeking has always existed, but it’s been largely invisible to institutions. It’s just been hidden in draft folders, study group chats, scribbled notes, and conversations with tutors and peers, leaving no institutional trace. No easy path for educators to jump in and provide the secretly desired support.
But when students turn to governed AI chat instead, that need surfaces in a system the institution owns. It becomes visible, which means it’s actionable.
There’s an understandable instinct in a compliance-first environment to treat this visibility purely as a monitoring opportunity. A way to detect misuse of some nature, but this is incomplete, and risks squandering something far more valuable.
Given a properly governed alternative to unsupervised tools, student usage skews toward coaching-seeking rather than shortcut-seeking. This is signalling where genuine understanding is thin at a scale and speed no unit coordinator, no matter how dedicated, could produce through office hours and forum posts alone.
Not only does a diagnostic layer help an educator’s current cohort, it also informs future assessment design. If a disproportionate share of “am I on track?” questions cluster around a particular brief or task, it could point to a clarity problem in the brief. If students repeatedly ask a governed tool to explain a concept the week before it’s taught in lectures, maybe it’s a curriculum sequencing signal. None of it is available when all you receive is a final product.
Students and educators alike shouldn’t see a governed AI platform as simply giving institutions more control, or an audit trail, or even defensibility for a misconduct case. It gives visibility, which, when applied correctly, provides insight.
If students are pushed toward unsupervised external tools, the institution loses both the ability to shape the interaction and learn from it, and then no one wins. Every one of those “is this good enough?” questions end up happening somewhere institutions can never see.
TEQSA’s guidance is pushing institutions toward assessment redesign rather than detection-only strategies. Redesign needs a feedback source, and the richest one may be sitting inside the very AI interactions that have been a governance headache.
Getting this right won’t be simple, and neither the strictest AI bans nor handing students permission to use an unmonitored chatbot will help. Building a governed middle path, then having the openness and curiosity to look at what students are asking and redesigning around the answer, is how institutions will succeed.
Students aren’t hiding from feedback; they’re asking for more of it. Whether it’s face-to-face with a lecturer or at 2am in a chat window when no one else is around, the insights are all valuable. The question for Australian education institutions is whether they’re willing to build the kind of environment where that question gets answered, and learned from, rather than blocked.
James Thorley is Regional Vice President – APAC at Turnitin
