The first time I went looking for an honest answer to “should I study with AI or just grind through past papers,” every result was a product page trying to sell me an AI exam tool. None of them spoke to a uni student. They were all built for Year 12 kids chasing an ATAR (the Australian Tertiary Admission Rank, the score that gets school leavers into university). I was a mature-age student with two subjects, a full-time sales job, and kids who needed dinner. I needed a strategy, not a subscription.

So here is the honest answer I wish I had found: it is not AI versus past papers. It is both, sequenced correctly. Get the order wrong and you waste the little study time you have. Get it right and you walk into the exam knowing what is coming.

Let me explain what each one actually does, where the academic integrity line sits, and the sequence I now use every exam block.

What past papers actually do

Past papers do one thing no AI tool can fully fake: they put you under exam conditions. Real question formats. Real time pressure. The exact phrasing your faculty likes to use. The way marks get carved up across a question.

When you sit a past paper with a timer running, you find out fast whether you can actually produce an answer in the eight minutes the marker has allowed for it, or whether you only “know” the content when you have unlimited time and your notes open. Those are two very different states, and most of us overestimate the first one.

Australian university exams also carry their own quirks. Many subjects use a heavily weighted final exam, sometimes with a hurdle requirement (a minimum mark you must hit on that one assessment to pass the subject, regardless of your overall score). Past papers show you how that weighting plays out in practice. They reveal whether the exam rewards breadth across the whole unit or depth on three big topics.

There is one limit worth naming. Plenty of Australian unis do not release past papers, or release them without answers. Check your subject’s learning management system, the library exam database, and your subject coordinator before assuming none exist. A paper without a marking guide is still useful for format and timing, even if you have to reconstruct what a strong answer looks like.

What AI tools actually do

AI tools are brilliant at the part of study that happens between exam sittings: building understanding.

Tools like ChatGPT, Claude, NotebookLM, and AI-generated Anki decks are very good at explaining a concept on demand, surfacing the gaps in what you know, and turning a dense reading into something you can actually hold in your head. I went from ChatGPT to Copilot to Claude over the last couple of years, and the thing they all do well is compress. NotebookLM, as of its early design, worked only from sources you uploaded, which made it well suited to grounding answers in your own lecture notes rather than the open internet. It has since expanded to include optional web-sourcing and Deep Research features, so its capabilities continue to change.

What AI does not do well is simulate exam pressure. An AI quiz will happily give you a hint, accept a half-formed answer, and let you try again. A real exam does none of that. So if your whole prep is “chat with an AI about the topic,” you will feel ready and then meet a blank exam booklet and a clock.

Australia’s own student guidance is comfortable with this study use. Study Australia advises international students that AI tools can be helpful for understanding course materials, generating ideas, and fixing grammar, while warning that AI-generated content must not replace original thinking or research (Study Australia, n.d.). The University of Sydney similarly frames AI as something students can use to support their learning, provided the work they submit is their own (University of Sydney, n.d.).

The academic integrity line is clearer than people think

Here is the boundary, and it is not blurry. Using AI to help you understand content is permitted at nearly every Australian university. Submitting AI-generated answers as your own work is not.

The ANU’s student guide on generative AI lays out best-practice habits: be transparent, check your unit’s specific rules, and never pass off AI output as your own thinking (Australian National University, n.d.). The sector regulator, TEQSA, has flagged that generative AI poses a real and evolving risk to academic integrity, which is exactly why individual unit rules now vary so much (Tertiary Education Quality and Standards Agency, 2024).

For exam prep specifically, this matters less than it does for assignments, because in a closed exam you are not submitting anything you prepared at home. But it still shapes how you use AI. Asking AI to explain why an answer is wrong is study. Memorising an AI-written model answer to reproduce verbatim is the kind of habit that gets you in trouble the moment the assessment changes shape. If you want the full breakdown of where that line sits, I wrote a longer piece on using AI to study without crossing the integrity line.

The sequence that actually works

Stop treating this as an either/or. Treat it as a timeline.

Early in the study period, lean on AI. This is when you have the most ground to cover and the least time pressure on any single concept. Use AI to explain the topics you skated over during semester, to generate practice questions that expose what you do not know, and to build flashcards for spaced repetition (reviewing material at increasing intervals so it sticks). I cover the scheduling side of that in my guide to spaced repetition for busy students.

In the final week, switch to past papers. Once the gaps are mostly closed, you need exam-condition practice, not more explanation. Sit the paper with a timer. Mark it honestly against whatever criteria you can find. Notice where you ran out of time and which topics you fumbled, then go back to AI for one more targeted pass on just those weak spots.

That last loop is the whole trick. Past papers diagnose; AI treats. You sit a timed paper to find the problem, then use AI to fix the specific thing the paper exposed. Doing it in that order means your scarce study hours land where they actually move your mark.

What this looks like when you are time-poor

If you are a working or mature-age student, your real constraint is not motivation. It is blocks of time. My own study sessions run on roughly 42-minute focus cycles, not the textbook 25, because that is where my attention actually drifts. You should find your own number rather than forcing someone else’s.

AI’s genuine strength here is compression. It can turn a two-hour lecture recording into a 20-minute question-and-answer session you can do on your phone between meetings. That is real, and on a tight week it is the difference between covering a topic and skipping it.

But do not mistake that efficiency for exam readiness. Compressing the content is step one. You still owe yourself at least one or two full timed papers before the exam, even if you have to claw the time out of a weekend. The students I have spoken to who got burnt were almost always the ones who did all their prep in efficient little AI sessions and never once practised under the clock.

The real risks of over-relying on AI

Two failure modes show up again and again.

First, AI explanations can be confidently wrong. The output reads as authoritative even when it has misstated a formula, invented a case, or fudged a definition. If you are learning a topic for the first time from AI alone, you have no way to catch the error, because you do not yet know enough to spot it. Cross-check anything that will be on the exam against your lecture notes, the textbook, or your teaching staff.

Second, students who skip past papers consistently misjudge the exam. They get the difficulty wrong, the format wrong, and the time budget wrong. AI can describe what an exam might look like, but it does not have your unit coordinator’s actual paper, and it does not know how your faculty distributes marks. The official open-book exam guidance from universities like Melbourne makes the same point in a different way: preparation has to match the real format of the assessment, not a generic idea of it (University of Melbourne, n.d.).

The layer both methods are missing

Whether you are drilling AI flashcards or working through a past paper, both methods get sharper when you already know what the exam is testing and how the marks are split. That is the bit neither tool gives you on its own.

This is the gap I am building GradeMap to fill. GradeMap is designed to help you decode assessment criteria and rubrics before you sit down to study, so you walk into either method knowing which concepts the marker actually cares about most. I am building it because I needed it myself, juggling a Graduate Diploma at QUT on top of full-time work. If you want a head start on that skill now, my walkthrough on using AI to decode a university rubric covers the manual version.

Rubric clarity first, then AI to close the gaps, then past papers under the clock. That is the sequence. Get the order right and you stop studying everything and start studying the right things.

References

Australian National University. (n.d.). Guide for students: Best practice when using generative AI. https://www.anu.edu.au/students/academic-skills/referencing-and-academic-integrity/academic-integrity-best-practice/guide

Tertiary Education Quality and Standards Agency. (2024). The evolving risk to academic integrity posed by generative artificial intelligence. https://www.teqsa.gov.au/sites/default/files/2024-08/evolving-risk-to-academic-integrity-posed-by-generative-artificial-intelligence.pdf

University of Melbourne. (n.d.). Open-book exams. https://students.unimelb.edu.au/academic-skills/resources/study-skills/exam-preparation/open-book-exams

University of Sydney. (n.d.). Artificial intelligence. https://www.sydney.edu.au/students/academic-integrity/artificial-intelligence.html

FAQ

Is it cheating to use ChatGPT or Claude to study for an exam?

No, using AI to understand content, generate practice questions, or explain concepts is permitted at nearly every Australian university, as long as the work you submit is your own. The line is crossed when you present AI-generated answers as your own thinking. For closed exams this is rarely an issue, but always check your specific unit’s rules, since they vary across subjects and institutions (Australian National University, n.d.).

Should I use past papers or AI tools first?

Use AI first, past papers second. Early in your study period, when you have the most ground to cover, AI is efficient for closing knowledge gaps and explaining tricky concepts. In the final week, switch to past papers for timed, exam-condition practice. Then go back to AI for one targeted pass on whatever the paper exposed as weak.

Can AI replace past papers entirely?

No. AI cannot reliably simulate exam pressure, and it does not have your unit coordinator’s actual paper or your faculty’s real marking weights. Students who skip past papers tend to misjudge the exam’s difficulty, format, and timing. AI is for building understanding between sessions; past papers are for rehearsing the real thing.

What if my university does not release past papers?

Check the learning management system, the library’s exam database, and ask your subject coordinator before assuming none exist. If you can only find papers without answers, they are still valuable for practising the format and timing. You can use AI to help you reconstruct what a strong answer might look like, then verify it against your lecture notes and textbook rather than trusting the AI output alone.

How much study time should a working student spend on each method?

There is no fixed split, but a rough guide is to spend the bulk of your early prep on AI-assisted understanding and reserve the final stretch for at least one or two full timed past papers. Work in focus blocks that match your own attention span rather than a prescribed length, and protect the timed practice even on a busy week, because that is the part most time-poor students skip and regret.