AI Exam Question Prediction: Hype, Tools, and the 85% Claim

AI exam question prediction has jumped from a niche revision hack to a viral promise, and it is now one of the hottest corners of AI exam preparation. A group of UK teenagers claims their model forecast last year’s A-level and GCSE papers with roughly 85% accuracy, and a wave of AI study tools now offers to do the same for any student. However, exam boards are not impressed — and the maths behind these predictions is messier than the marketing suggests.

AI exam question prediction uses machine learning to forecast which topics and questions are most likely to appear on an upcoming test. In practice, the tools ingest years of past papers, the syllabus, and examiner patterns, then rank topics by probability. As a result, a student gets a focused list of ‘likely’ questions instead of revising everything. It sits at the sharp end of a wider shift toward AI in education, where AI learning tools increasingly shape how students revise.

Moreover, many platforms now build full practice papers from those predictions. The pitch is simple: study smarter, not harder. Still, a prediction is a probability, not a leak — a distinction that tends to vanish in the hype.

The current buzz traces back to a group of teenagers who say they built an AI model that predicted A-level and GCSE exam papers across all subjects. According to The Independent, they claim last year’s forecasts landed at about 85% accuracy, and they are running the model again this year. Naturally, that figure spread fast online.

However, ‘85% accuracy’ is doing a lot of work here. Predicting that a frequently tested topic will reappear is very different from predicting the exact wording of a question. In addition, no independent body has verified the claim. For now, it is a headline, not a guarantee.

Plenty of apps now compete to be the best AI for predicting exam questions. PredictExam.ai, for example, turns your notes into a practice exam in about 30 seconds and tries to surface the 20% of concepts that drive most of the marks. Meanwhile, PaperPredict analyses past papers, syllabi, and trends to flag likely questions; for India’s NEET 2025, it reportedly matched 72 of 90 biology questions.

Other names — ExamPredict, exam-mate, and a growing list of AI exam generator apps — promise similar results. Because of this crowded field, students increasingly treat an AI exam generator as a standard revision tool rather than a gimmick.

You do not actually need a dedicated app. Many students simply use a ChatGPT exam predictor approach: paste in past papers and the syllabus, then ask the model to rank likely topics. With the right exam prediction prompt — one that asks for high-frequency topics plus sample questions — general chatbots produce surprisingly usable revision lists.

You probably already have what you need, too. The AI writing tools students use for essays — ChatGPT, Claude, Gemini and the like — can rank likely exam topics once you feed them past papers. So instead of paying for a niche predictor, you can lean on a chatbot you already use. However, garbage in means garbage out: weak prompts and thin source material produce vague guesses.

Yes — and this is where the technology is genuinely strong. AI can generate exam questions that mirror a syllabus, complete with mark schemes and model answers. As a result, students get unlimited practice papers tuned to their course.

Prediction and generation, though, are not the same thing. Generating a realistic question is reliable; predicting that this specific question will appear is a gamble. Used well, AI-generated practice sharpens recall and exposes weak spots. Used as a shortcut, it breeds false confidence.

So how accurate is AI exam question prediction in practice? The honest answer is useful, but not psychic. These systems are good at spotting high-frequency topics that recur year after year. Therefore, their predictions often look impressive simply because popular topics tend to come up anyway.

However, exams are designed to surprise, and low-frequency topics still appear. Even the strongest tools, like PaperPredict’s NEET result, miss a meaningful slice of questions. In short, treat any accuracy figure as a probability, not a promise.

Can you get AI exam question prediction free? Partly. Several tools offer a free tier — PredictExam.ai, for instance, is free to start — and a basic chatbot prompt costs nothing. Most students simply want a downloadable predicted paper, often as a PDF, that they can print and revise from.

Paid plans then add full mock papers, mark schemes, and extra subjects. Still, free or paid, the underlying limit is the same: the tool ranks probabilities, it does not see the real paper.

Regulators are not buying the hype. In Ireland, the State Examinations Commission warned students to ignore AI Leaving Cert predictions, calling them ‘spurious’ and ‘unhelpful distractions.’ One platform had even claimed to use 15 years of commission data to build predicted papers with ‘marking-scheme verbatim phrases.’ You can read the commission’s warning in full.

Meanwhile, in the UK, exam boards are reportedly weighing action over viral social-media predictions. The worry is twofold: students may revise the wrong things, and a culture of ‘predicted leaks’ slowly chips away at trust in exams.

Curious where this leaves real learning? Our look at whether AI is making our brains lazy digs into what outsourcing study to algorithms does to memory and skill — the bigger question behind every predicted paper.

*Sources: The Independent, Schools Week, PaperPredict, PredictExam.ai.*