How to Predict Exam Questions: The Four Signals

To predict what will be on an exam, treat it as a research job. Four signals do almost all the work: the syllabus and its published weightings, the pattern across past papers, what your instructor says out loud in class, and how much teaching time each topic got. Rank topics by how many signals point at them.

Most advice on this stops at the first line of that paragraph. Look at past papers, it says, and then moves on to something else. That skips the part that carries the marks: knowing what each signal is reliable for, what it is useless for, and what to do when two of them disagree. A syllabus is excellent evidence about question format and nearly silent about which topic gets the eight-mark question. A throwaway line in a lecture is the opposite. Read all four, weight them properly, and the paper stops feeling like a lottery.

Can you actually predict what will be on an exam?

At the level of topics and question types, yes. At the level of individual questions, no, and anyone promising otherwise is selling something. The reason the first half works comes down to how exams get built: your paper was designed backwards from a document you are allowed to read.

The name for that design principle is constructive alignment. Queen Mary University of London describes it as an approach to curriculum design which is focused on closely aligning teaching and assessment to intended learning outcomes, coined by John Biggs, and it puts the sequence in a specific order: first the intended learning outcomes are written, then the assessment is designed to verify those outcomes, then teaching activities are built to get students there. The same page quotes Biggs on the detail that matters most to you as a student: the learning activity in the intended outcomes, expressed as a verb, needs to be activated in the teaching and in the assessment task.

Read that as a student rather than as a course designer and it says something useful. The person setting your paper started from a list you can look up, and they were under professional pressure to make the questions match it. The list is a partial answer key for the shape of the exam. What it cannot tell you is the wording, the data set, the case study, or the numbers, and that is the residue of genuine uncertainty you should plan around instead of pretending away.

Signal one: read the syllabus for verbs and weightings

Open the syllabus, specification, or course outline and read it twice. On the first pass, underline every verb attached to an outcome. Describe, define, and state are recall questions. Calculate, apply, and derive are worked questions. Compare, evaluate, justify, and assess are extended-answer questions carrying the highest marks. An outcome phrased as evaluate will almost never be examined by a definition question, so a topic you can only define is a topic you have half prepared.

On the second pass, hunt for numbers. A surprising amount of exam weighting is published outright, and most students never look. AQA states the assessment objectives for A-level Biology and how the marks split between them: roughly 30 to 35 percent for demonstrating knowledge and understanding, 40 to 45 percent for applying it, and 25 to 30 percent for analysing, interpreting, and evaluating. It then breaks that down per paper, so paper 1 leans on recall at 44 to 48 percent while paper 2 runs recall down to 23 to 27 percent and pushes application up to 52 to 56 percent. A student who revises by memorising definitions has prepared for about a third of that qualification.

The US equivalent is just as blunt. College Board publishes a percentage of the exam score against every unit of AP Biology: Natural Selection at 13 to 20 percent of the exam, Cellular Energetics and Gene Expression at 12 to 16 percent each, Chemistry of Life and Heredity at 8 to 11 percent. The assessment page adds the structure: 60 multiple-choice questions and 6 free-response questions, ninety minutes each, worth 50 percent of the score apiece. Two topics on that list are worth double two others. That is a ranking handed to you for free, and it took four minutes to find.

Turn the second pass into a table with one row per topic and the published weighting beside it. Then add one honest column: how confident you are on that topic today. You now have the skeleton of a revision order before any other signal arrives. One rule keeps this clean: anything the syllabus never mentions is very unlikely to be examined, however many happy hours your lecturer spent on it.

Signal two: what repeats across past papers

Past papers are where question spotting earns its reputation, both the good half and the bad half. The good half is that the architecture of a paper repeats with remarkable stubbornness. The bad half is that students look for repeated questions, find none, and conclude the exercise failed.

What actually repeats is the frame. Section layout. The number of marks attached to each kind of task. The command words. The two or three topics that appear in every single sitting because the course cannot examine itself without them. Sometimes the exact structure of a long question, with a definition part, a calculation part, and an evaluation part in that order, year after year, with only the context swapped.

Mine that properly with a grid. One row per past paper, one column per syllabus topic. Work through three or four papers marking which topics appeared and how many marks each carried, and the pattern surfaces without any interpretation on your part. Count the command words separately while you are there. If evaluate shows up nine times across four papers and derive shows up twice, you have learned something about what the examiners think this course is for.

Two cautions. First, a reformed syllabus makes old papers actively misleading, and reforms happen more often than students realise. English A-levels went linear in the 2015 reform, for instance, so papers written under the old modular system test a structure that no longer exists. Check the specification version on the front of any paper before you trust it. Second, a topic that has appeared every year for four years counts as evidence and stops well short of a promise. Examiners know that students count, and a well-run assessment team occasionally rotates the heavy question specifically to catch people who bet on the pattern.

Your own mock is the freshest past paper you will ever get, and it was usually written or chosen by the people closest to your real one. Sit it honestly and it doubles as a prediction instrument, which is one of the strongest arguments for keeping your pre-mock revision light rather than cramming for it. A mock you crammed for tells you about your cramming.

Signal three: the things your instructor says out loud

This is the strongest predictor available to you and the one almost nobody records. The person who writes your exam spends several hours a week telling a room full of students what they care about, and most of that room writes down only the content.

Start a separate log, one line per hint, with the date and the exact words rather than your summary of them. The phrases worth catching are unmistakable once you are listening for them:

• You will need to know this. The unambiguous version. Also: this comes up a lot, this is the sort of thing we assess, make sure you can do this one.
• Anything written on the board that is not on the slide. If someone prepared a deck and still stopped to write something by hand, that thing survived their own editing pass twice.
• A worked example done a second time. Repetition inside a single course is expensive in teaching time. Nobody spends it accidentally.
• The topic they are visibly pleased about. Enthusiasm is a leaky signal. Instructors write questions about the material they find interesting, and their own research area is a standing favourite.
• Everything said in a review session. Revision lectures exist to steer you. The topics chosen for the session, and the order they came in, are a compressed version of the person deciding what matters.
• Corrections to a past cohort. Lines like: last year almost everybody got this wrong. That is somebody describing a question they intend to ask again.

If your lectures are recorded, this becomes a search problem instead of a memory problem. Scrubbing a recording for the moments around each slide transition costs far less than rewatching it, and the technique for mining a recording without losing an afternoon is worth having on its own, which we cover in how to study from recorded lectures.

You can also just ask, and students consistently underrate how well this works. Asking what is on the exam gets a vague answer because it is a vague question. Narrow it and the same instructor will often be specific: which of these two chapters carries more marks, is this derivation something we reproduce or only recognise, what kind of question should we expect on this topic. Those are questions about preparation, so answering them is squarely part of the job. Ask at office hours, write the reply down verbatim, and treat it as strong evidence rather than a guarantee.

Signal four: teaching time is a rough proxy for exam weight

Count the room. How many lectures did each topic get, how many slides, how many pages of the handbook, which topics carried a problem set or a lab or a graded assignment. Courses are built to a timetable, and the timetable is a budget: time spent is an approximation of importance held by the same person who sets the paper.

Slide decks make this measurable in about ten minutes, because a deck is already sorted into topics and already counted for you. Pull the slide count per topic alongside your syllabus weightings and the two usually agree. Where they disagree, you have found something worth a question at office hours. Decks repay this kind of attention in general, and there is a specific way to work them that avoids drowning in bullet points, which we set out in how to study from PowerPoint slides.

Rank this signal below the syllabus, though, because teaching time gets distorted in two directions. A topic can eat three lectures because it is hard to teach rather than because it is heavily assessed. And a topic can be covered in twenty minutes because it is straightforward while still carrying a fifteen-percent weighting in the specification. When teaching time and published weighting disagree, believe the published weighting.

How do you turn the signals into an actual question list?

By writing the questions out in full, in the language the exam uses. The gap between a revision list that says photosynthesis and a question that says evaluate the evidence that light intensity limits the rate of photosynthesis, twelve marks, is the entire difference between recognising a topic and being able to answer on it.

Sort your ranked table by how many signals agree, which is deliberately a different sorting key from the one you would use under time pressure. With the exam tomorrow, you rank by marks available against your own weakness and accept that some topics get dropped. With the exam four weeks out, signal agreement is the better key, because the job is allocating a month of attention across a whole course rather than triaging a single night.

Writing the questions is doing more work than it looks. Composing a question forces you to decide what the main idea of a section actually is, which is why question generation has been studied as a comprehension strategy in its own right. Rosenshine, Meister and Chapman reviewed the intervention studies in 1996 and reported that teaching students to generate questions about material they had read produced gains in comprehension, with an overall median effect size of 0.36 on standardized tests and 0.86 on experimenter-developed comprehension tests, all measured on new material. Be careful about how far you carry that. Those were reading-comprehension interventions rather than studies of exam prediction, so the honest reading stays narrow: composing questions counts as studying in its own right. Which means the hours you spend building a predicted paper come out of your revision budget and go straight back into it.

Then sit the thing. A predicted paper you have written and never attempted is a list of topics wearing a costume. Give yourself the real time limit, work closed-book, and treat the result exactly as you would a real practice test, which means sorting every miss by the reason you missed it. The method for that is in how to review a practice test with an error log, and it applies to a paper you invented just as cleanly as to an official one.

Where question spotting goes wrong

The technique has a bad name among teachers, and the reasons are worth taking seriously, because every one of them is a way students actually lose marks.

• Betting the whole revision on three topics. The single most expensive mistake. Prediction is a tool for ordering your effort. The moment it becomes a tool for deleting topics, one rotated question wipes out a term.
• Treating one hint as a contract. An instructor saying this is important raises the probability. It does not make the topic certain, and it says nothing about the other 80 percent of the paper.
• Predicting content and ignoring format. Students who guess the topics correctly and still underperform have usually prepared recall for a paper that wanted application. The verbs in the syllabus exist to stop exactly this.
• A new instructor resets the pattern. If the person teaching the course changed this year, past papers describe somebody else. Weight the spoken signal much higher and the historical signal much lower.
• Losing the connective tissue. Revising ten isolated predicted answers leaves you helpless in front of a question that spans two of them, and synthesis questions are exactly where the top marks live.

The format you are sitting also changes what prediction even means. In a written paper you are guessing at somebody else, and the four signals are external. In an oral exam, a viva, or a defence, the primary source document is usually your own work, so the questions come out of the weakest paragraph you wrote and the method you chose to skip. That is a different research problem with a different answer, laid out in our guide to preparing for an oral exam.

Where software can and cannot do this for you

Start with what no tool can do, because it is most of the job. Software cannot sit in your lecture and hear the sentence that begins with you will need to know. It cannot attend your review session, read the room, notice that your instructor got animated about one paper in particular, or ask a follow-up at office hours. It cannot read an exam that has not been written. Signal three, the strongest of the four, is entirely human work and always will be.

What software removes is the transcription step at the end. Once the ranking exists and you know your top two or three topics, turning the underlying material into questions you can actually sit is mechanical work that eats an evening. Upload the relevant file to GeniusPal, the lecture deck or the syllabus PDF or your own notes on the topics that scored highest, and it returns a set of questions drawn from that file. Those questions can be worked three ways: as a multiple-choice quiz, as flashcards, or in active recall, which asks you to write an answer out from memory and then shows you the source to check it against.

The boundaries are worth stating plainly. GeniusPal reads text-based files you upload, so a spoken hint reaches it only if you typed it into your notes first, and it has no idea what your exam format or command words look like unless the file you gave it says so. It takes one file and returns one set of questions in a single pass, with no chat and no follow-up questions. Your predicted paper, the one written in the exam language with the right mark allocations, is still something you write. A new account comes with two study-set generations for the life of that account, so the sensible way to spend the first one is on whichever topic your ranking put at number one.