Can I Use ChatGPT to Study for the Leaving Cert?

Yes, for some parts of study, and you should be careful with the rest. A general chatbot is genuinely good at explaining, rephrasing and drilling you. It is unreliable on the two things the Leaving Cert turns on, which are the real past papers and the official marking schemes.

Where a general chatbot helps

Explaining something a different way. A textbook explains a topic once, in one register. If that explanation did not land, you can ask a chatbot for another one, then ask again for a simpler version, then ask it for a worked example. That back-and-forth is the thing a static page cannot do.

Turning your own words into the subject's words. You can describe what you think is going on in ordinary language and ask which curriculum terms match. This is useful in Biology, Geography and Business, where a lot of marks are attached to naming the right thing.

Quizzing you. Ask it to test you on a topic, one question at a time, and to wait for your answer before giving the next. Answering from memory is what makes a topic stick.

Language practice. For French, German, Spanish and Irish, a chatbot will hold a conversation at whatever level you set. It corrects your grammar as you go, and it keeps going for as long as you want. Conversation practice is hard to get on your own outside class, and a chatbot will do it on demand.

Company when you are stuck. When you cannot start a question there is often nobody to ask at that hour. Having something to ask keeps you at the desk.

Where it falls down for this exam

It does not have the marking scheme in front of it

The State Examinations Commission publishes an official marking scheme beside every paper, free. The scheme states what earns each mark.

Take the composing section of Leaving Cert English 2024 Higher Level Paper 1. A composition there is marked out of 100, split into Clarity of Purpose 30, Coherence of Delivery 30, Efficiency of Language Use 30 and Accuracy of Mechanics 10 (SEC marking scheme). Thirty of those marks are for shaping and sustaining the piece, and ten are for spelling and punctuation alone.

Ask a chatbot for "a good answer" and it writes something that reads well. Unless you paste the scheme in, or it searches out the document itself, it is answering from a general sense of the subject. It cannot tell you which of those four criteria your attempt is short on.

It can invent questions that were never on a paper

Language models produce confident wrong answers. A September 2025 paper by researchers at OpenAI and Georgia Tech puts the cause plainly. It argues that "language models hallucinate because the training and evaluation procedures reward guessing over acknowledging uncertainty" (Kalai and others, 2025). A model with no record of a specific 2019 Biology question can produce a plausible one anyway.

For exam practice this matters, because an invented question reads exactly like a real one, and no marking scheme exists behind it for you to check yourself against.

The same problem shows up in predicted papers. Ahead of the 2026 exams, the SEC warned students about AI-generated predictions. A spokeswoman said candidates "should be wary of all sources which claim to have knowledge of an examination paper in advance" (Tuesday 26 May 2026, Irish Examiner).

It can do the work you needed to do yourself

This one has been measured. A field experiment with nearly 1,000 secondary school maths students, published in PNAS in July 2025, gave one group a standard ChatGPT interface during practice sessions. Their practice grades came in 48% higher than the grades of the group given no AI. On a later exam sat without any AI access, that group scored 17% lower than students who had never been given the tool (PNAS).

A second group used a version prompted to give hints and withhold the final answer. That group did not show the drop. The researchers describe the first group as using the model as a crutch during practice.

The lesson is about how you use it. The tool does what you ask, so if you ask for finished answers during practice you will get finished answers, and the exam in June is sat without it.

There are rules on coursework

The SEC's Coursework Rules & Procedures 2025-2026 states that "AI tools should not be used to generate coursework content". The same document allows that they "may assist in generating ideas, conducting research, or enhancing clarity, provided its use is clearly documented". You sign an authentication form confirming your coursework is free from any unacknowledged AI assistance. These rules cover coursework you submit, and they do not cover private study.

How to use any AI tool well

Ask it to quiz you rather than to answer you. Tell it to ask one question, wait, then tell you what was missing from your answer. This flips it from a source into a testing partner.

Give it the real question. Open a genuine past paper and paste the question in, so the wording and the mark allocation are the ones the SEC used. The past paper finder gets you the paper and its marking scheme together.

Paste the marking scheme in before you ask it to mark you. Without the scheme it invents a grading standard. With the scheme in front of it, it can tell you which specific marks your answer did not reach.

Ask for a hint first, then a bigger hint. The PNAS result is about answers handed over too early. You can impose the guardrail yourself by asking for the next step only.

Never take its word on what is on the papers. Anything it tells you about frequency, past questions or likely topics needs checking against the actual papers. The Leaving Cert Biology past papers and every other subject hub link straight to the SEC PDFs.

What we do differently

ExamAI is built around those failure modes.

The tutor searches a database of real past exam questions and their official SEC marking schemes across Leaving Cert subjects. When it pulls a question, the real question text appears in the chat, with the official marking scheme behind a reveal and a link to the original SEC PDF. You can always see the source. When it marks your attempt it uses the scheme's real mark allocations. When a paper is not in the database it tells you, and it will not present an invented question as a real one.

It also defaults to Socratic. The first response to a stuck student is a question or a hint, and it works down to a full solution only when the hints are not landing. That is deliberate, and it is the same guardrail the PNAS study found to protect learning.

We wrote separately about how this works on maths past papers, and about what to do with a past paper once you have one.

FAQ

Is ChatGPT good for the Leaving Cert? It is good for explanation, rephrasing, quizzing and language practice. It is weak on the marking schemes and on the contents of specific past papers, so check anything it says about the exam itself against the SEC documents.

Can I use AI for my Leaving Cert coursework? Not to generate the content. The SEC coursework rules allow AI tools for generating ideas, research and clarity, provided the use is clearly documented. You also sign a form confirming there is no unacknowledged AI assistance in the work.

Can AI predict what will come up on the paper? No source can. The SEC told students in May 2026 to be wary of anything claiming advance knowledge of a paper. Past-paper frequency is history you can check yourself, and it is not a prediction.

Is it cheating to use AI to study? No. The written examinations are what is assessed, and how you prepare for them is your own business. The rules apply to coursework you submit.

Will using AI make me lazier? It can. The PNAS experiment found students who had answers handed to them during practice did worse afterwards without the tool. The group who got hints and no final answer did not show that drop.

Open the ExamAI tutor and ask it about a topic you are stuck on. It works from real past papers and their marking schemes, and it is free to start.