What we use AI for, and what we don't

Here is where the line sits.

AI writes the audio, runs the mock orals and explains the questions. It does not mark anything that counts. Below: what that means in practice, and the five things that went up this week.

Listening audio from a transcript. Listening Studio takes your transcript, you pick the voices, and it renders an HSC-style recording with the student and teacher Word files. No transcript yet? It drafts one by course for you to edit.

Mock orals that answer back. A student speaks, a feedback report comes back a couple of minutes later. It is practice, not a mark.

Every question explained. Each practice set in the study hubs now carries a worked explanation: what the question is testing, where the answer sits, and the usual ways marks go missing.

It does not mark work that counts. The written feedback with the Chinese and Literature sets is marked by a teacher against the criteria.

It does not publish a paper unchecked. Every set is read against the course vocabulary before it ships, and anything outside the course is rewritten.

It does not replace the clock. Sitting a full paper under exam timing is still the part that moves marks.

Give students an AI answer and the marking criteria, and ask them to mark it. They find the missing detail much faster when the mistake is not their own.

More of these live in a shared page you can comment on: AI tips for language teachers. Add one that worked in your classroom.