Using AI feedback on a GAMSAT Section 2 essay without handing over the writing

Using AI after you have written a Section 2 essay is a defensible practice habit. Using it before, to supply angles, paragraphs or sentences, is where the risk sits, and the reason comes from how the Australian Council for Educational Research (ACER) describes the section.

The framework below has five parts: write the full attempt yourself under time, ask the tool questions rather than for text, sort every comment against ACER's published criteria, revise from your own head, and find a second reader where you can. Each part exists for a reason, and the reasons are set out in turn.

What ACER's booklet says that bears on AI

ACER's GAMSAT information booklet for 2026 (last updated 20 November 2025) describes Written Communication as two 30-minute writing tasks. Both tasks give four statements, tied together by a theme. The booklet says the section assesses whether a candidate can produce ideas and build them out on the page. Marking considers two broad things: how good the thinking about the topic is, and how well the language carries it. Language control is assessed only as far as it contributes to the effectiveness of the whole response, not in isolation, and candidates are not marked on whether their ideas or attitudes are correct.

On originality, the booklet states that every response goes through plagiarism checks as part of marking, that the wording of those checks involves artificial intelligence, and that responses are also compared with ones submitted at earlier sittings. It does not describe how any of this works. It adds that an essay counts as original only if it answers the topics set for that sitting. Text or statements taken from a prepared essay can be flagged as plagiarised even when the candidate has never seen the original, and misconduct clauses may apply. ACER advises against memorised essays, and says responses prepared in advance or unrelated to the topic receive a low score.

The booklet does not mention AI-written model essays by name, and the passage concerns the response you submit on the day rather than practice. Our reading is that a sentence you absorbed from a tool and then reproduce in the test is borrowed text in the sense the booklet describes, and that rehearsing a tool-written essay for a theme you hope will appear runs against the requirement to answer the topics actually set. That is an inference from ACER's wording, not an ACER statement about AI tools. If the point matters to you, ask ACER.

Write the whole attempt first, alone and timed

If the section assesses whether you can produce and develop ideas, then the practice that matters is producing and developing them. A tool that offers angles before you have drafted changes the exercise into choosing among someone else's ideas. There is a second reason. Feedback is only information about your writing if the writing was yours, so an essay assembled with help gives you comments about the help.

Time the attempt to match the real structure: 30 minutes per task, one theme, four statements. A relaxed untimed draft will show you how you write with no time limit at all, which is a different thing from the response ACER marks.

Five questions that return the work to you

With a finished draft in hand, these are the uses we would treat as safe. Each one asks the tool to point at something while leaving you to decide what to do about it.

• Where could a reader take this sentence in two ways? You can check each flagged sentence yourself and rule on whether it was ambiguous, so the tool points and you judge.
• What might a reader misunderstand about my position? ACER's criteria look at the thought you offer in response to the task, and this question tests whether the thought lands as you meant it.
• Which steps in my argument have no reason attached? Finding a gap and filling it from your own thinking is the development of an idea, which is what the booklet says the section is looking for.
• What does each of these two outlines leave unexamined? Write both outlines yourself first. The generating is already done, and the tool is only asked to compare.
• What would a sceptical reader ask at this point? A question has to be answered in your own words, whereas a suggested rewrite can be pasted in.

The uses to avoid are the mirror image: asking for a stronger opening, a model essay on the same theme, a better version of a paragraph, or a decision about which side to argue. Each returns text or ideas that you would then have to disown, and each is the kind of material the originality wording is about.

Sort every comment against ACER's two criteria

Treat each piece of feedback as a claim to check. ACER publishes two groupings: thought and content (what is made of the task, and the kinds of thought offered) and organisation and expression (the shape of the piece, and how effective and fluent the language is). Put each comment into one of three piles: speaks to thought and content, speaks to organisation and expression, or speaks to neither.

The third pile is where the published wording does useful work. A comment telling you your view is the wrong one addresses something the booklet says is not assessed. A line edit that changes nothing about how clearly an idea lands has no stated link to the criteria either, since language is assessed through the effectiveness of the whole response. You can still decide to take such advice, but you would be doing so for reasons other than the criteria.

What the competitor pages do and do not tell you

Candidates now meet AI offered as a product. Fraser's markets a Section 2 AI tutor, saying it was built for GAMSAT writing and judges essays on three named qualities: how deep the thinking goes, plus structure and expression. One course provider's page lists an internally developed AI system, described in terms of answering questions about content and about how a topic relates to the examination, with tutors available at the student's request. The same page lists tutor-marked essays as a separate item.

Those are vendors describing their own products. Nothing we have read establishes how reliably any AI tool judges a Section 2 essay, in either direction, so the sorting step above is how you supply the check yourself. Translating a tool's own categories into ACER's two before acting on them is part of that. ACER also states that three independent scorings are applied to each response, so a single run from any tool is one reading. Asking the same question twice and seeing whether the same issue is raised is a cheap way to test whether a comment is stable. That is our suggestion, not a validated method.

Revise from a blank page, then find a second reader

Once a comment has survived the sort, act on it by rewriting the relevant part from your own head, without the tool's wording in view. Pasting or lightly editing suggested phrasing carries tool-written text into your own habits, and that text is what the originality wording makes risky.

Where you can, put a human reader's view next to the tool's. The booklet's preparation section lists a question on whether candidates can have feedback on practice essay responses. The answer was not among the pages we read, so check it directly in the booklet before relying on anything we say about ACER's own feedback.

Where this approach falls short

It is slower than asking a tool to improve your essay, because you write first and rewrite afterwards. It does not tell you what an attempt would score, and nothing in it makes AI feedback accurate. It only makes you check. The booklet's AI wording leaves open how the checks work and how text from a tool would be treated, so our reading above is cautious rather than settled. The booklet is the 2026 edition, so confirm the current wording on ACER's GAMSAT site before you sit.

What the sorting step cannot show you on its own

Once you have several comments across several essays, it becomes hard to see which criterion each one belongs to and whether revising to a comment changed anything in the next attempt. Verantrix can give structured feedback on a Section 2 essay against a published rubric, so each comment is attached to a named criterion you can inspect. It does not mark as ACER would, and the same checking applies to it. Whatever tool you use, write down the single change you are testing before your next timed attempt, then compare the two drafts against that change alone.