{"course":{"slug":"ai-for-teachers","title":"AI for Teachers","url":"https://addaly.com/learn/ai-for-teachers","level":"beginner","track":"work"},"lesson":{"slug":"the-student-facing-feedback-loop","title":"Feedback before submission: the student-facing loop","url":"https://addaly.com/learn/ai-for-teachers/the-student-facing-feedback-loop","position":29,"of":57,"minutes":8,"answer":"Students can get useful critique on their own drafts if given a prompt card that forbids rewriting, but the model wants to help and will start rewriting anyway.","keyIdea":"Students can get useful critique on their own drafts if given a prompt card that forbids rewriting, but the model wants to help and will start rewriting anyway, so the card needs a re-statement line and the task needs a paper trail."},"body":{"format":"markdown","text":"## The loop that used to be impossible\n\nThe best feedback arrives while the draft is still open. For one teacher and sixty students, that has never been possible; you see the draft when it is handed in, if at all. A student with a model and a good prompt card can get critique on Tuesday night for Wednesday's draft, and do it three times before you ever see the work.\n\nThis is one of the new things these tools make possible in a classroom. It also has a specific failure, and the check question describes it.\n\n## The prompt card\n\nGive students the exact text, printed, to paste before their draft:\n\n```text\nYou are helping me improve my own writing. Do not rewrite any part of it. Do not give me sentences to use.\nRead my draft. Then:\n1. Tell me the three weakest places, quoting a few words from each.\n2. For each, ask me one question that would help me fix it myself.\nDo not answer your own questions. Wait for me.\nAfter I reply, do the same again. Never write my text for me, even if I ask.\n```\n\nThe last line matters, because some students will ask, and the model should have been told in advance to decline.\n\nThe card teaches two things at once. Students learn that the useful request is for critique, not text. And they experience, in their own work, the difference between a question that makes them think and an answer that makes them stop.\n\n## The drift, and the fix\n\nThe model was trained to be helpful, and the most helpful-seeming thing to do with a weak sentence is to write a better one. Over a conversation, that pull wins. By the fourth exchange it is \"suggesting\" a phrasing; by the sixth it has offered a paragraph. The student did not ask. The text is just there, and it is better than theirs, and it goes in.\n\nTwo mitigations. First, the card includes a line the student pastes again every third message: \"Reminder: do not write any of my text.\" This works most of the time. Second, the task carries a version trail: the student submits the draft before the loop and the draft after, and a note of what they changed and why. The trail makes the model's contribution visible, and it turns the exercise into what it should be: a record of a student improving their own work with questions.\n\n## Access\n\nNot every student has a device or data at home. The teaching-honest-use lesson set the rule: no task requires personal AI access. So the loop is optional, or it runs in school on a shared device, or the teacher runs it for a student who asks. A student without the loop is not disadvantaged in the mark, because the mark is on the final draft and the loop only helps them get there. It is a support, not a requirement.\n\n## What students learn from it\n\nTeachers who run this for a term report the same thing: students get better at asking for feedback, from the tool and from each other. \"Where is my argument weakest?\" is a question that transfers. And students who have watched a model ask them three questions about their own paragraph often start asking themselves those questions before drafting, which is the whole aim.\n\n## Age and terms of service\n\nMost consumer chatbots require users to be at least 13, and some require 18 or parental consent below that. A later block covers the rules in detail. For now: check what your school allows before putting a prompt card in a 12-year-old's hand, and where it is not allowed, run the loop on a school device with you at the keyboard. The learning is the same.\n\n## The honest limit\n\nA model's critique of a student's draft is only as good as its reading of the rubric and its sense of what the student meant, and both are imperfect. It will sometimes call the strongest sentence weak because it is unusual, and it will not know that the \"weak\" paragraph was deliberately plain. Tell students this. \"It is a reader, not a judge. Take the questions, ignore the verdicts.\" A student who can hold that distinction has learned something about feedback from anyone."},"check":{"question":"A teacher gives students a prompt card: \"Do not rewrite my work. Point out three weaknesses and ask me a question about each.\" After a few exchanges, several students' drafts contain fluent paragraphs they did not write. What happened?","options":[{"text":"The students ignored the card and asked for rewrites, which makes this a discipline problem rather than a design problem with the task.","correct":false,"tempting":"Some may have, but the pattern across several students who followed the card points at the tool's behaviour, which drifts toward producing text over a conversation."},{"text":"Helpfulness training pulled the model back toward writing text over the turns; repeat the constraint and keep a version trail.","correct":true},{"text":"The card was too long for the model to hold; a single short instruction would have been obeyed more reliably across the conversation.","correct":false,"tempting":"Length of the card is not the mechanism; the drift happens across turns regardless of how short the initial instruction is."},{"text":"The model cannot give feedback without rewriting the text, so student-facing feedback of this kind should not be attempted at all.","correct":false,"tempting":"It can and does give critique-only feedback; the failure is drift over turns, which is manageable, not a categorical limit."}]},"prev":{"slug":"feedback-students-actually-use","title":"Feedback students actually use","url":"https://addaly.com/learn/ai-for-teachers/feedback-students-actually-use"},"next":{"slug":"reading-thirty-answers-at-once","title":"Reading thirty answers at once","url":"https://addaly.com/learn/ai-for-teachers/reading-thirty-answers-at-once"},"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) 'Feedback before submission: the student-facing loop', in AI for Teachers. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-for-teachers/the-student-facing-feedback-loop (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":"Feedback before submission: the student-facing loop","courseTitle":"AI for Teachers","module":{"index":4,"slug":"assessment-and-feedback","title":"Assessment, feedback and marking"},"answer":null,"keyIdea":"Students can get useful critique on their own drafts if given a prompt card that forbids rewriting, but the model wants to help and will start rewriting anyway, so the card needs a re-statement line and the task needs a paper trail.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"the-loop-that-used-to-be-impossible","runs":[{"text":"The loop that used to be impossible"}]},{"t":"p","runs":[{"text":"The best feedback arrives while the draft is still open. For one teacher and sixty students, that has never been possible; you see the draft when it is handed in, if at all. A student with a model and a good prompt card can get critique on Tuesday night for Wednesday’s draft, and do it three times before you ever see the work."}]},{"t":"p","runs":[{"text":"This is one of the new things these tools make possible in a classroom. It also has a specific failure, and the check question describes it."}]},{"t":"h2","id":"the-prompt-card","runs":[{"text":"The prompt card"}]},{"t":"p","runs":[{"text":"Give students the exact text, printed, to paste before their draft:"}]},{"t":"code","lang":"text","label":"Text","text":"You are helping me improve my own writing. Do not rewrite any part of it. Do not give me sentences to use.\nRead my draft. Then:\n1. Tell me the three weakest places, quoting a few words from each.\n2. For each, ask me one question that would help me fix it myself.\nDo not answer your own questions. Wait for me.\nAfter I reply, do the same again. Never write my text for me, even if I ask."},{"t":"p","runs":[{"text":"The last line matters, because some students will ask, and the model should have been told in advance to decline."}]},{"t":"p","runs":[{"text":"The card teaches two things at once. Students learn that the useful request is for critique, not text. And they experience, in their own work, the difference between a question that makes them think and an answer that makes them stop."}]},{"t":"h2","id":"the-drift-and-the-fix","runs":[{"text":"The drift, and the fix"}]},{"t":"p","runs":[{"text":"The model was trained to be helpful, and the most helpful-seeming thing to do with a weak sentence is to write a better one. Over a conversation, that pull wins. By the fourth exchange it is “suggesting” a phrasing; by the sixth it has offered a paragraph. The student did not ask. The text is just there, and it is better than theirs, and it goes in."}]},{"t":"p","runs":[{"text":"Two mitigations. First, the card includes a line the student pastes again every third message: “Reminder: do not write any of my text.” This works most of the time. Second, the task carries a version trail: the student submits the draft before the loop and the draft after, and a note of what they changed and why. The trail makes the model’s contribution visible, and it turns the exercise into what it should be: a record of a student improving their own work with questions."}]},{"t":"h2","id":"access","runs":[{"text":"Access"}]},{"t":"p","runs":[{"text":"Not every student has a device or data at home. The teaching-honest-use lesson set the rule: no task requires personal AI access. So the loop is optional, or it runs in school on a shared device, or the teacher runs it for a student who asks. A student without the loop is not disadvantaged in the mark, because the mark is on the final draft and the loop only helps them get there. It is a support, not a requirement."}]},{"t":"h2","id":"what-students-learn-from-it","runs":[{"text":"What students learn from it"}]},{"t":"p","runs":[{"text":"Teachers who run this for a term report the same thing: students get better at asking for feedback, from the tool and from each other. “Where is my argument weakest?” is a question that transfers. And students who have watched a model ask them three questions about their own paragraph often start asking themselves those questions before drafting, which is the whole aim."}]},{"t":"h2","id":"age-and-terms-of-service","runs":[{"text":"Age and terms of service"}]},{"t":"p","runs":[{"text":"Most consumer chatbots require users to be at least 13, and some require 18 or parental consent below that. A later block covers the rules in detail. For now: check what your school allows before putting a prompt card in a 12-year-old’s hand, and where it is not allowed, run the loop on a school device with you at the keyboard. The learning is the same."}]},{"t":"h2","id":"the-honest-limit","runs":[{"text":"The honest limit"}],"role":"limit"},{"t":"p","runs":[{"text":"A model’s critique of a student’s draft is only as good as its reading of the rubric and its sense of what the student meant, and both are imperfect. It will sometimes call the strongest sentence weak because it is unusual, and it will not know that the “weak” paragraph was deliberately plain. Tell students this. “It is a reader, not a judge. Take the questions, ignore the verdicts.” A student who can hold that distinction has learned something about feedback from anyone."}]}],"check":{"question":[{"text":"A teacher gives students a prompt card: “Do not rewrite my work. Point out three weaknesses and ask me a question about each.” After a few exchanges, several students’ drafts contain fluent paragraphs they did not write. What happened?"}],"options":[{"text":[{"text":"The students ignored the card and asked for rewrites, which makes this a discipline problem rather than a design problem with the task."}],"correct":false,"tempting":[{"text":"Some may have, but the pattern across several students who followed the card points at the tool’s behaviour, which drifts toward producing text over a conversation."}]},{"text":[{"text":"Helpfulness training pulled the model back toward writing text over the turns; repeat the constraint and keep a version trail."}],"correct":true,"tempting":null},{"text":[{"text":"The card was too long for the model to hold; a single short instruction would have been obeyed more reliably across the conversation."}],"correct":false,"tempting":[{"text":"Length of the card is not the mechanism; the drift happens across turns regardless of how short the initial instruction is."}]},{"text":[{"text":"The model cannot give feedback without rewriting the text, so student-facing feedback of this kind should not be attempted at all."}],"correct":false,"tempting":[{"text":"It can and does give critique-only feedback; the failure is drift over turns, which is manageable, not a categorical limit."}]}],"order":[3,1,0,2],"why":null},"prev":{"slug":"feedback-students-actually-use","title":"Feedback students actually use","minutes":8},"next":{"slug":"reading-thirty-answers-at-once","title":"Reading thirty answers at once","minutes":9}}}