{"course":{"slug":"ai-for-teachers","title":"AI for Teachers","url":"https://addaly.com/learn/ai-for-teachers","level":"beginner","track":"work"},"lesson":{"slug":"writing-the-exam-paper","title":"Writing the exam paper","url":"https://addaly.com/learn/ai-for-teachers/writing-the-exam-paper","position":32,"of":57,"minutes":9,"answer":"Draft the paper from a coverage grid (topic against type against marks) rather than from a prompt for \"a test\", and then sit it yourself against the clock.","keyIdea":"Draft the paper from a coverage grid (topic against type against marks) rather than from a prompt for \"a test\", and then sit it yourself against the clock, because the model does not know your board's structure until shown and cannot feel how long forty marks take."},"body":{"format":"markdown","text":"## The test is a design, not a request\n\n\"Write me a test\" gets a test. It is not the test you need, because a test has a structure that a prompt for \"a test\" does not carry: which topics, in what proportion, tested how, worth how much, in what order, in the format the students will meet in the real exam. Experienced teachers hold that structure in their heads. Writing it down is the step that makes the model useful and the paper defensible.\n\n## The coverage grid\n\nBefore any prompt, a grid. Rows are the topics of the term, from the syllabus list. Columns are question types: recall, application, multi-step, extended. Cells are marks. A 60-mark paper might put 8 marks on topic one (6 recall, 2 application), 15 on topic two (5, 5, 5), and so on, until the column totals show the paper is a third recall and two-thirds thinking, or whatever balance the board's papers actually have.\n\nThe grid takes fifteen minutes and it is the exam. Everything else is filling it. It is also the document that answers \"why was there nothing on topic four?\" from a head of department, and \"why were there so many calculations?\" from a parent.\n\n## The prompt is the grid plus the format\n\nPaste the grid. Paste one real past paper from your board, or its section structure and command words if the paper is long. Then:\n\n```text\nWrite the questions to fill this grid exactly, in the format and command words of the pasted paper. For each question give the marks in brackets and a mark scheme with one line per mark awarded. Number the questions in order of increasing difficulty within each section.\nDo not add questions the grid does not specify. Do not write an introduction.\n```\n\nThe question bank lesson's two-pass rule applies: get the mark schemes in a second pass, from the questions alone, and compare. The calculator rule applies to every number.\n\n> Figure: Two rows of a coverage grid. A two by two grid, rows Rivers: erosion and deposition and Flooding: causes and management, columns Recall and describe and Apply and explain: Rivers: erosion and deposition and Recall and describe: 8 marks, 4 questions; Rivers: erosion and deposition and Apply and explain: 7 marks, 2 questions; Flooding: causes and management and Recall and describe: 6 marks, 3 questions; Flooding: causes and management and Apply and explain: 15 marks, 3 questions. A slice of a 60-mark Class 9 geography paper. A real grid has every topic of the term down the side and four or five question types across the top, and the column totals are what tell you the paper is a third recall rather than two-thirds. The grid takes fifteen minutes and it is the exam; the prompt is only the filling.\n\n## Sit it yourself\n\nThe check question is not a joke. Every teacher who has used a generated paper has been caught by the timing. The model produced sixty marks of questions, and sixty marks of questions is not sixty minutes; it is however long those particular questions take, and the model cannot know, because it has never sat a test.\n\nSo sit it. Against the clock, writing full answers, as a good student would. Multiply your time by about one and a half for the class. If that is over the exam length, cut, and cut from the grid, so the balance survives, not from the end of the paper, which is where the hardest questions are.\n\nSitting the paper also finds the question that cannot be answered from the information given, the diagram the question needs and does not have, and the question that gives away the answer to the next one. A paper nobody has sat is a draft.\n\n## Difficulty and order\n\nAsk for a spread: a few marks every student will get, a middle most will, and a few that separate the top. The model produces a flat paper unless told, and a flat paper tells you less. Order within a section from easiest to hardest so that weaker students bank marks before they meet the wall.\n\nThe bank's data helps here. If you have recorded which questions students found hard last time, you know what a \"hard\" question is for your class, and you can show the model an example rather than describe one.\n\n## Two papers from one grid\n\nOnce the grid is right, a parallel paper (same grid, different questions) is one prompt, and the reconciliation check catches the arithmetic that changed with the numbers. That gives you a resit paper, a paper for the class next door, and next year's mock, all from one afternoon's design.\n\n## The paper as a teaching record\n\nKeep the grid with the results. After the test, fill in the class average per cell. A topic that scored badly across all types was not learned; a topic that scored well on recall and badly on application was learned as facts and not as understanding. That is a different reteaching problem, and the grid shows you which one you have.\n\n## The honest limit\n\nThe model can fill a grid well. It cannot tell you whether the grid is right: whether a third recall is too much for this class, whether the syllabus weighting you chose matches what the board will examine next year. That is assessment expertise, and it comes from marking real papers and reading examiner reports. Use the model to produce the paper faster, and spend the time you saved reading the examiner's report from last year, which tells you what a paper should look like better than any prompt."},"check":{"question":"A teacher asks for \"a 60-mark end-of-term physics test for Class 10\" and gets a well-formatted paper. When she sits it herself it takes 85 minutes; students have 60. Which change to her process fixes the problem at source?","options":[{"text":"Ask for a 45-mark test instead of 60, since experience shows the model overestimates what fits by roughly a third every time.","correct":false,"tempting":"A fixed correction factor treats a symptom; the model has no sense of time at all, and the overrun varies with question type."},{"text":"Add \"this must take exactly 60 minutes\" to the prompt so that the model calibrates the number and length of the questions.","correct":false,"tempting":"The model cannot feel duration; a time instruction changes the number of questions loosely and does not calibrate to real students."},{"text":"Ask the model to estimate the time for each question and sum the estimates before finalising the paper.","correct":false,"tempting":"Model estimates of time are guesses from text patterns, not measurements, and summing guesses does not produce a measurement."},{"text":"Build the paper from a marks grid and sit it herself against the clock before it is set.","correct":true}]},"prev":{"slug":"marking-handwritten-work-with-a-phone","title":"Marking handwritten work with a phone","url":"https://addaly.com/learn/ai-for-teachers/marking-handwritten-work-with-a-phone"},"next":{"slug":"peer-assessment-with-a-third-marker","title":"Peer assessment with a third marker","url":"https://addaly.com/learn/ai-for-teachers/peer-assessment-with-a-third-marker"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'Writing the exam paper', in AI for Teachers. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-for-teachers/writing-the-exam-paper (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"Writing the exam paper","courseTitle":"AI for Teachers","module":{"index":4,"slug":"assessment-and-feedback","title":"Assessment, feedback and marking"},"answer":null,"keyIdea":"Draft the paper from a coverage grid (topic against type against marks) rather than from a prompt for “a test”, and then sit it yourself against the clock, because the model does not know your board’s structure until shown and cannot feel how long forty marks take.","hero":{"kind":"figure","figure":{"t":"figure","label":"Figure","figure":{"cols":["Recall and describe","Apply and explain"],"rows":["Rivers: erosion and deposition","Flooding: causes and management"],"type":"matrix","cells":[["8 marks, 4 questions","7 marks, 2 questions"],["6 marks, 3 questions","15 marks, 3 questions"]],"title":"Two rows of a coverage grid","caption":"A slice of a 60-mark Class 9 geography paper. A real grid has every topic of the term down the side and four or five question types across the top, and the column totals are what tell you the paper is a third recall rather than two-thirds. The grid takes fifteen minutes and it is the exam; the prompt is only the filling.","cellLabels":[["banked early","the middle of the paper"],["banked early","where the paper separates"]]},"alt":"Figure: Two rows of a coverage grid. A two by two grid, rows Rivers: erosion and deposition and Flooding: causes and management, columns Recall and describe and Apply and explain: Rivers: erosion and deposition and Recall and describe: 8 marks, 4 questions; Rivers: erosion and deposition and Apply and explain: 7 marks, 2 questions; Flooding: causes and management and Recall and describe: 6 marks, 3 questions; Flooding: causes and management and Apply and explain: 15 marks, 3 questions. A slice of a 60-mark Class 9 geography paper. A real grid has every topic of the term down the side and four or five question types across the top, and the column totals are what tell you the paper is a third recall rather than two-thirds. The grid takes fifteen minutes and it is the exam; the prompt is only the filling."}},"blocks":[{"t":"h2","id":"the-test-is-a-design-not-a-request","runs":[{"text":"The test is a design, not a request"}]},{"t":"p","runs":[{"text":"“Write me a test” gets a test. It is not the test you need, because a test has a structure that a prompt for “a test” does not carry: which topics, in what proportion, tested how, worth how much, in what order, in the format the students will meet in the real exam. Experienced teachers hold that structure in their heads. Writing it down is the step that makes the model useful and the paper defensible."}]},{"t":"h2","id":"the-coverage-grid","runs":[{"text":"The coverage grid"}]},{"t":"p","runs":[{"text":"Before any prompt, a grid. Rows are the topics of the term, from the syllabus list. Columns are question types: recall, application, multi-step, extended. Cells are marks. A 60-mark paper might put 8 marks on topic one (6 recall, 2 application), 15 on topic two (5, 5, 5), and so on, until the column totals show the paper is a third recall and two-thirds thinking, or whatever balance the board’s papers actually have."}]},{"t":"p","runs":[{"text":"The grid takes fifteen minutes and it is the exam. Everything else is filling it. It is also the document that answers “why was there nothing on topic four?” from a head of department, and “why were there so many calculations?” from a parent."}]},{"t":"h2","id":"the-prompt-is-the-grid-plus-the-format","runs":[{"text":"The prompt is the grid plus the format"}]},{"t":"p","runs":[{"text":"Paste the grid. Paste one real past paper from your board, or its section structure and command words if the paper is long. Then:"}]},{"t":"code","lang":"text","label":"Text","text":"Write the questions to fill this grid exactly, in the format and command words of the pasted paper. For each question give the marks in brackets and a mark scheme with one line per mark awarded. Number the questions in order of increasing difficulty within each section.\nDo not add questions the grid does not specify. Do not write an introduction."},{"t":"p","runs":[{"text":"The question bank lesson’s two-pass rule applies: get the mark schemes in a second pass, from the questions alone, and compare. The calculator rule applies to every number."}]},{"t":"h2","id":"sit-it-yourself","runs":[{"text":"Sit it yourself"}]},{"t":"p","runs":[{"text":"The check question is not a joke. Every teacher who has used a generated paper has been caught by the timing. The model produced sixty marks of questions, and sixty marks of questions is not sixty minutes; it is however long those particular questions take, and the model cannot know, because it has never sat a test."}]},{"t":"p","runs":[{"text":"So sit it. Against the clock, writing full answers, as a good student would. Multiply your time by about one and a half for the class. If that is over the exam length, cut, and cut from the grid, so the balance survives, not from the end of the paper, which is where the hardest questions are."}]},{"t":"p","runs":[{"text":"Sitting the paper also finds the question that cannot be answered from the information given, the diagram the question needs and does not have, and the question that gives away the answer to the next one. A paper nobody has sat is a draft."}]},{"t":"h2","id":"difficulty-and-order","runs":[{"text":"Difficulty and order"}]},{"t":"p","runs":[{"text":"Ask for a spread: a few marks every student will get, a middle most will, and a few that separate the top. The model produces a flat paper unless told, and a flat paper tells you less. Order within a section from easiest to hardest so that weaker students bank marks before they meet the wall."}]},{"t":"p","runs":[{"text":"The bank’s data helps here. If you have recorded which questions students found hard last time, you know what a “hard” question is for your class, and you can show the model an example rather than describe one."}]},{"t":"h2","id":"two-papers-from-one-grid","runs":[{"text":"Two papers from one grid"}]},{"t":"p","runs":[{"text":"Once the grid is right, a parallel paper (same grid, different questions) is one prompt, and the reconciliation check catches the arithmetic that changed with the numbers. That gives you a resit paper, a paper for the class next door, and next year’s mock, all from one afternoon’s design."}]},{"t":"h2","id":"the-paper-as-a-teaching-record","runs":[{"text":"The paper as a teaching record"}]},{"t":"p","runs":[{"text":"Keep the grid with the results. After the test, fill in the class average per cell. A topic that scored badly across all types was not learned; a topic that scored well on recall and badly on application was learned as facts and not as understanding. That is a different reteaching problem, and the grid shows you which one you have."}]},{"t":"h2","id":"the-honest-limit","runs":[{"text":"The honest limit"}],"role":"limit"},{"t":"p","runs":[{"text":"The model can fill a grid well. It cannot tell you whether the grid is right: whether a third recall is too much for this class, whether the syllabus weighting you chose matches what the board will examine next year. That is assessment expertise, and it comes from marking real papers and reading examiner reports. Use the model to produce the paper faster, and spend the time you saved reading the examiner’s report from last year, which tells you what a paper should look like better than any prompt."}]}],"check":{"question":[{"text":"A teacher asks for “a 60-mark end-of-term physics test for Class 10” and gets a well-formatted paper. When she sits it herself it takes 85 minutes; students have 60. Which change to her process fixes the problem at source?"}],"options":[{"text":[{"text":"Ask for a 45-mark test instead of 60, since experience shows the model overestimates what fits by roughly a third every time."}],"correct":false,"tempting":[{"text":"A fixed correction factor treats a symptom; the model has no sense of time at all, and the overrun varies with question type."}]},{"text":[{"text":"Add “this must take exactly 60 minutes” to the prompt so that the model calibrates the number and length of the questions."}],"correct":false,"tempting":[{"text":"The model cannot feel duration; a time instruction changes the number of questions loosely and does not calibrate to real students."}]},{"text":[{"text":"Ask the model to estimate the time for each question and sum the estimates before finalising the paper."}],"correct":false,"tempting":[{"text":"Model estimates of time are guesses from text patterns, not measurements, and summing guesses does not produce a measurement."}]},{"text":[{"text":"Build the paper from a marks grid and sit it herself against the clock before it is set."}],"correct":true,"tempting":null}],"order":[1,0,2,3],"why":null},"prev":{"slug":"marking-handwritten-work-with-a-phone","title":"Marking handwritten work with a phone","minutes":8},"next":{"slug":"peer-assessment-with-a-third-marker","title":"Peer assessment with a third marker","minutes":8}}}