{"course":{"slug":"ai-for-teachers","title":"AI for Teachers","url":"https://addaly.com/learn/ai-for-teachers","level":"beginner","track":"work"},"lesson":{"slug":"show-it-one-of-yours","title":"Show it one of yours","url":"https://addaly.com/learn/ai-for-teachers/show-it-one-of-yours","position":10,"of":57,"minutes":8,"answer":"One pasted example of your own work steers the output harder than any description of it, because the model copies the shape of what it is shown, including, if you are not careful, the flaws.","keyIdea":"One pasted example of your own work steers the output harder than any description of it, because the model copies the shape of what it is shown, including, if you are not careful, the flaws."},"body":{"format":"markdown","text":"## Description versus demonstration\n\nYou can spend three paragraphs describing the kind of comprehension question you want: inference-based, one sentence, no yes/no answers, referring to a specific line, pitched at Class 7. Or you can paste one question you wrote last year that was exactly right and say: \"Eight more like this, on the attached passage.\"\n\nThe second wins, nearly every time. The model is a pattern-continuer, and a concrete example is the densest possible statement of a pattern. Your description is your theory of what makes the question good. The example is the question. It carries things you did not know you cared about (the length of the stem, the way you phrase \"according to the text\", where the line reference sits) and the model reproduces all of it.\n\nPractitioners call this few-shot prompting. You do not need the term. You need the habit: when you have a good one, show it.\n\n## What to show\n\n- **A worked example in your style.** For maths, paste a solution laid out the way you lay them out on the board. The generated ones will follow your layout, which means students see one consistent method rather than the model's default.\n- **A question with its mark scheme.** The mark scheme teaches the model what a full answer contains, so the next questions come with usable schemes.\n- **A student-facing instruction.** If your instructions always say \"Show your working. Circle your final answer,\" paste one and they all will.\n- **A parent message you were pleased with.** Tone is nearly impossible to describe and trivial to demonstrate.\n- **A whole worksheet.** Photograph it if it only exists on paper. \"Same format, same difficulty, different numbers and a different context\" produces a parallel version for a second class or a resit.\n\nTwo or three examples are better than one when the examples vary along the dimension you want varied: three questions of increasing difficulty tells the model to produce a spread. Beyond three or four you are spending the memory window on demonstration and getting little back.\n\n## The flaw comes with it\n\nThe mechanism cuts both ways. The model copies the shape it is shown, and it has no way to distinguish the parts you are proud of from the parts you did not notice.\n\nIf your example question has an ambiguous stem, you get eight ambiguous stems. If your worked solution skips a step you always skip on the board, every generated solution skips it. If the worksheet you photographed had a typo in the instructions, the typo is now a feature.\n\nSo choose the example with care. Read it once as though a colleague wrote it. Fix what you find before you paste. A clean example is worth more than three good ones with one flaw between them.\n\n## Telling it what to keep and what to change\n\nAn example on its own leaves the model to guess what \"like this\" means. Say it:\n\n> Keep: the structure, the length, the kind of thinking required, the instruction wording. Change: the passage, the numbers, the context.\n\nWithout the split, the model sometimes keeps the content and changes the structure, which is the opposite of what you wanted. It also helps to name the thing you are showing: \"This is a two-mark inference question\" gives it a label to generalise from, so the eight new ones are two-mark inference questions and not two-mark recall questions in inference clothing.\n\n## The negative example\n\nSometimes the fastest steer is a bad example. \"Here is a question I do not want: it can be answered by copying a sentence from the passage. Here is one I do want.\" Two examples, one labelled bad, one labelled good, and the model has the boundary. This is particularly effective for the things teachers find hardest to describe: the difference between a question that tests reading and one that tests understanding is easier to show than to define.\n\n## A department's worth of examples\n\nIf your department has a shared folder of good questions, worksheets and mark schemes, you have a prompt library already. Each item is a demonstration. A new teacher who pastes the department's best Class 9 test and asks for a parallel version inherits the department's standards in an afternoon. That is a better induction than most schools manage, and it costs nothing but the pasting.\n\n## The honest limit\n\nAn example steers form. It does not steer truth. Eight questions \"like this one\" on a new passage can still ask about a detail the passage does not contain, or carry a wrong answer in the key. The example bought you the right shape. The checking is still yours."},"check":{"question":"A teacher pastes one of her own past exam questions, which happens to have an ambiguous stem, and asks for eight more like it. All eight come back ambiguous in the same way. What has happened?","options":[{"text":"The model recognised the ambiguity as an intended feature of her assessment style and reproduced it deliberately.","correct":false,"tempting":"Attributes judgement to pattern-matching; the model has no view on whether the ambiguity was intended, it simply continues the shape it was given."},{"text":"Eight is too many to generate from one example; with three examples the ambiguity would have been averaged out.","correct":false,"tempting":"More examples with the same flaw reinforce the flaw; the fix is a clean example, not more examples."},{"text":"The example steered the shape, flaw included, because the model copies what is shown rather than what is meant.","correct":true},{"text":"The model cannot write exam questions without a mark scheme, so it fell back on copying the example word for word.","correct":false,"tempting":"Nothing in the outcome suggests copying; the eight were new questions sharing a structural flaw, which is imitation of pattern, not reproduction of text."}]},"prev":{"slug":"the-five-things-only-you-know","title":"The five things only you know","url":"https://addaly.com/learn/ai-for-teachers/the-five-things-only-you-know"},"next":{"slug":"the-second-message-matters-more","title":"The second message matters more than the first","url":"https://addaly.com/learn/ai-for-teachers/the-second-message-matters-more"},"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) 'Show it one of yours', in AI for Teachers. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-for-teachers/show-it-one-of-yours (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":"Show it one of yours","courseTitle":"AI for Teachers","module":{"index":2,"slug":"prompting-as-a-teacher","title":"Prompting as a teacher"},"answer":null,"keyIdea":"One pasted example of your own work steers the output harder than any description of it, because the model copies the shape of what it is shown, including, if you are not careful, the flaws.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"description-versus-demonstration","runs":[{"text":"Description versus demonstration"}]},{"t":"p","runs":[{"text":"You can spend three paragraphs describing the kind of comprehension question you want: inference-based, one sentence, no yes/no answers, referring to a specific line, pitched at Class 7. Or you can paste one question you wrote last year that was exactly right and say: “Eight more like this, on the attached passage.”"}]},{"t":"p","runs":[{"text":"The second wins, nearly every time. The model is a pattern-continuer, and a concrete example is the densest possible statement of a pattern. Your description is your theory of what makes the question good. The example is the question. It carries things you did not know you cared about (the length of the stem, the way you phrase “according to the text”, where the line reference sits) and the model reproduces all of it."}]},{"t":"p","runs":[{"text":"Practitioners call this few-shot prompting. You do not need the term. You need the habit: when you have a good one, show it."}]},{"t":"h2","id":"what-to-show","runs":[{"text":"What to show"}]},{"t":"ul","items":[[{"text":"A worked example in your style.","b":true},{"text":" For maths, paste a solution laid out the way you lay them out on the board. The generated ones will follow your layout, which means students see one consistent method rather than the model’s default."}],[{"text":"A question with its mark scheme.","b":true},{"text":" The mark scheme teaches the model what a full answer contains, so the next questions come with usable schemes."}],[{"text":"A student-facing instruction.","b":true},{"text":" If your instructions always say “Show your working. Circle your final answer,” paste one and they all will."}],[{"text":"A parent message you were pleased with.","b":true},{"text":" Tone is nearly impossible to describe and trivial to demonstrate."}],[{"text":"A whole worksheet.","b":true},{"text":" Photograph it if it only exists on paper. “Same format, same difficulty, different numbers and a different context” produces a parallel version for a second class or a resit."}]]},{"t":"p","runs":[{"text":"Two or three examples are better than one when the examples vary along the dimension you want varied: three questions of increasing difficulty tells the model to produce a spread. Beyond three or four you are spending the memory window on demonstration and getting little back."}]},{"t":"h2","id":"the-flaw-comes-with-it","runs":[{"text":"The flaw comes with it"}]},{"t":"p","runs":[{"text":"The mechanism cuts both ways. The model copies the shape it is shown, and it has no way to distinguish the parts you are proud of from the parts you did not notice."}]},{"t":"p","runs":[{"text":"If your example question has an ambiguous stem, you get eight ambiguous stems. If your worked solution skips a step you always skip on the board, every generated solution skips it. If the worksheet you photographed had a typo in the instructions, the typo is now a feature."}]},{"t":"p","runs":[{"text":"So choose the example with care. Read it once as though a colleague wrote it. Fix what you find before you paste. A clean example is worth more than three good ones with one flaw between them."}]},{"t":"h2","id":"telling-it-what-to-keep-and-what-to-change","runs":[{"text":"Telling it what to keep and what to change"}]},{"t":"p","runs":[{"text":"An example on its own leaves the model to guess what “like this” means. Say it:"}]},{"t":"quote","runs":[{"text":"Keep: the structure, the length, the kind of thinking required, the instruction wording. Change: the passage, the numbers, the context."}]},{"t":"p","runs":[{"text":"Without the split, the model sometimes keeps the content and changes the structure, which is the opposite of what you wanted. It also helps to name the thing you are showing: “This is a two-mark inference question” gives it a label to generalise from, so the eight new ones are two-mark inference questions and not two-mark recall questions in inference clothing."}]},{"t":"h2","id":"the-negative-example","runs":[{"text":"The negative example"}]},{"t":"p","runs":[{"text":"Sometimes the fastest steer is a bad example. “Here is a question I do not want: it can be answered by copying a sentence from the passage. Here is one I do want.” Two examples, one labelled bad, one labelled good, and the model has the boundary. This is particularly effective for the things teachers find hardest to describe: the difference between a question that tests reading and one that tests understanding is easier to show than to define."}]},{"t":"h2","id":"a-department-s-worth-of-examples","runs":[{"text":"A department’s worth of examples"}]},{"t":"p","runs":[{"text":"If your department has a shared folder of good questions, worksheets and mark schemes, you have a prompt library already. Each item is a demonstration. A new teacher who pastes the department’s best Class 9 test and asks for a parallel version inherits the department’s standards in an afternoon. That is a better induction than most schools manage, and it costs nothing but the pasting."}]},{"t":"h2","id":"the-honest-limit","runs":[{"text":"The honest limit"}],"role":"limit"},{"t":"p","runs":[{"text":"An example steers form. It does not steer truth. Eight questions “like this one” on a new passage can still ask about a detail the passage does not contain, or carry a wrong answer in the key. The example bought you the right shape. The checking is still yours."}]}],"check":{"question":[{"text":"A teacher pastes one of her own past exam questions, which happens to have an ambiguous stem, and asks for eight more like it. All eight come back ambiguous in the same way. What has happened?"}],"options":[{"text":[{"text":"The model recognised the ambiguity as an intended feature of her assessment style and reproduced it deliberately."}],"correct":false,"tempting":[{"text":"Attributes judgement to pattern-matching; the model has no view on whether the ambiguity was intended, it simply continues the shape it was given."}]},{"text":[{"text":"Eight is too many to generate from one example; with three examples the ambiguity would have been averaged out."}],"correct":false,"tempting":[{"text":"More examples with the same flaw reinforce the flaw; the fix is a clean example, not more examples."}]},{"text":[{"text":"The example steered the shape, flaw included, because the model copies what is shown rather than what is meant."}],"correct":true,"tempting":null},{"text":[{"text":"The model cannot write exam questions without a mark scheme, so it fell back on copying the example word for word."}],"correct":false,"tempting":[{"text":"Nothing in the outcome suggests copying; the eight were new questions sharing a structural flaw, which is imitation of pattern, not reproduction of text."}]}],"order":[0,2,1,3],"why":null},"prev":{"slug":"the-five-things-only-you-know","title":"The five things only you know","minutes":9},"next":{"slug":"the-second-message-matters-more","title":"The second message matters more than the first","minutes":8}}}