{"course":{"slug":"ai-for-students","title":"AI for Students","url":"https://addaly.com/learn/ai-for-students","level":"beginner","track":"work"},"lesson":{"slug":"prompts-worth-keeping","title":"Prompts worth keeping","url":"https://addaly.com/learn/ai-for-students/prompts-worth-keeping","position":18,"of":73,"minutes":7,"answer":"Keep six to ten prompts you actually reuse in one plain file, and note that the clauses doing the real work are the restrictions: one at a time, do not rewrite, do not help me.","keyIdea":"Keep six to ten prompts you actually reuse in one plain file, and note that the clauses doing the real work are the restrictions: one at a time, do not rewrite, do not help me."},"body":{"format":"markdown","text":"## A small file beats a large library\n\nThere is a market in prompt packs. Almost all of it is filler. What actually helps is a plain text file with six to ten prompts you have personally found useful, kept where you can copy from it in two seconds. Obsidian, Joplin, a note on your phone, a `prompts.md` file: the tool does not matter.\n\nWhat follows are the ones worth starting that file with. They are written to be pasted and edited, not admired.\n\n## The quiz\n\n> Here is the material, between the markers.\n>\n> ---\n> [paste]\n> ---\n>\n> Ask me one question at a time on this material. Do not give me the answer with the question. Wait for my answer, then tell me whether it is right, what I missed, and ask the next one. Make roughly a third of the questions require applying the idea to a case not in the text. Keep going until I say stop, then tell me which topics I was weakest on.\n\nThe *one at a time* and *wait* clauses are what make this work. Without them you get a list of ten questions and ten answers, which is reading, not retrieval.\n\n## The explain-back\n\n> I am going to explain [topic] to you from memory. Do not help me and do not fill gaps. When I finish, tell me: what was vague, what was actually wrong, what I omitted entirely, and what I clearly do not understand as opposed to merely mis-stated.\n\nYou do the production; the model does the marking. This inverts the usual arrangement and is worth more than any explanation it could give you.\n\n## The marker\n\n> Here is the assessment brief and mark scheme, and here is my draft.\n>\n> ---\n> [brief and mark scheme]\n> ---\n> [draft]\n> ---\n>\n> Act as the marker. For each criterion, say which band this currently sits in and quote the sentence that decides it. Then give the three changes that would move the most marks, in order. Do not rewrite anything and do not be encouraging.\n\n*Do not rewrite anything* is the essential clause. Without it you get an improved draft that is no longer yours.\n\n## The hint ladder\n\n> I am stuck on this problem. Give me hints one at a time, starting with the smallest possible nudge: a question about what I should be looking at, not a step. Only give the next hint when I ask. Never give the full solution unless I type SOLVE.\n\n## The sceptic\n\n> Below is an argument. Act as a hostile examiner. Identify the weakest step, the strongest counterargument, and any claim that is doing more work than the evidence supports. Do not soften anything and do not list strengths.\n\nUse this in a fresh chat with the argument pasted in, not at the end of the thread that produced it.\n\n## The level ladder\n\n> Explain [topic] three times: first as you would to someone a level below me, then at my level using standard terminology, then telling me what the version at level two simplifies or hides and what a specialist would object to.\n\n## The reading foothold\n\n> Here is a paragraph I cannot follow, and here is the sentence before and after it. Restate it in plain language without adding anything that is not there, then list the assumed background knowledge I appear to be missing. I will go back and read the original.\n\n## Standing instructions\n\nAnything you would paste every time belongs in the tool's custom-instructions field instead. Level, spelling convention, tone, and your standing preference for directness. Then your saved prompts can stay short.\n\n## Where a saved prompt goes wrong\n\nA saved prompt is a fixed instruction meeting a changing situation, so it fails in one predictable way: it keeps working after it should have stopped. The quiz prompt written when you knew nothing about a topic will still be asking recall questions in week nine, when you need application. The marker prompt aimed at a 1,500-word essay will happily be applied to a lab report with entirely different criteria.\n\nSo read the prompt before you paste it, every time, and change the one clause that no longer fits. That takes five seconds and is the difference between a tool and a ritual.\n\n## The maintenance rule\n\nDelete prompts you have not used in a month. A file of six live prompts gets used; a file of forty becomes a thing you scroll past. The best test of a prompt is whether you reached for it unprompted, and the best source of new ones is noticing the sentence you just typed that made an answer much better, and saving that."},"check":{"question":"In the quiz prompt, why does \"ask me one question at a time and wait for my answer\" matter so much?","options":[{"text":"It reduces token use, so a free tier's allowance stretches further across a revision session.","correct":false,"tempting":"True as a side effect and irrelevant to the learning mechanism; the same tokens spent on a question list would be equally cheap and far less useful."},{"text":"Without it you receive questions and answers together, which is reading rather than retrieval.","correct":true},{"text":"It keeps the conversation short enough to avoid drift and truncation over a long revision session.","correct":false,"tempting":"Names a real hazard of long threads but the wrong one here; an interactive quiz produces a longer conversation than a single list, not a shorter one."},{"text":"It lets the model adapt each question's difficulty to your previous answer, which is what makes the practice effective.","correct":false,"tempting":"Adaptivity is a genuine bonus, but the technique works even with fixed questions; the effect comes from attempting recall before seeing the answer."}]},"prev":{"slug":"diagnosing-a-bad-answer","title":"Diagnosing a bad answer","url":"https://addaly.com/learn/ai-for-students/diagnosing-a-bad-answer"},"next":{"slug":"getting-unstuck-not-finished","title":"Getting Unstuck, Not Getting Finished","url":"https://addaly.com/learn/ai-for-students/getting-unstuck-not-finished"},"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) 'Prompts worth keeping', in AI for Students. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-for-students/prompts-worth-keeping (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":"Prompts worth keeping","courseTitle":"AI for Students","module":{"index":2,"slug":"asking-well","title":"Asking well"},"answer":null,"keyIdea":"Keep six to ten prompts you actually reuse in one plain file, and note that the clauses doing the real work are the restrictions: one at a time, do not rewrite, do not help me.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"a-small-file-beats-a-large-library","runs":[{"text":"A small file beats a large library"}]},{"t":"p","runs":[{"text":"There is a market in prompt packs. Almost all of it is filler. What actually helps is a plain text file with six to ten prompts you have personally found useful, kept where you can copy from it in two seconds. Obsidian, Joplin, a note on your phone, a "},{"text":"prompts.md","code":true},{"text":" file: the tool does not matter."}]},{"t":"p","runs":[{"text":"What follows are the ones worth starting that file with. They are written to be pasted and edited, not admired."}]},{"t":"h2","id":"the-quiz","runs":[{"text":"The quiz"}]},{"t":"quote","runs":[{"text":"Here is the material, between the markers."}]},{"t":"quote","runs":[]},{"t":"quote","runs":[{"text":"---"}]},{"t":"quote","runs":[{"text":"[paste]"}]},{"t":"quote","runs":[{"text":"---"}]},{"t":"quote","runs":[]},{"t":"quote","runs":[{"text":"Ask me one question at a time on this material. Do not give me the answer with the question. Wait for my answer, then tell me whether it is right, what I missed, and ask the next one. Make roughly a third of the questions require applying the idea to a case not in the text. Keep going until I say stop, then tell me which topics I was weakest on."}]},{"t":"p","runs":[{"text":"The "},{"text":"one at a time","i":true},{"text":" and "},{"text":"wait","i":true},{"text":" clauses are what make this work. Without them you get a list of ten questions and ten answers, which is reading, not retrieval."}]},{"t":"h2","id":"the-explain-back","runs":[{"text":"The explain-back"}]},{"t":"quote","runs":[{"text":"I am going to explain [topic] to you from memory. Do not help me and do not fill gaps. When I finish, tell me: what was vague, what was actually wrong, what I omitted entirely, and what I clearly do not understand as opposed to merely mis-stated."}]},{"t":"p","runs":[{"text":"You do the production; the model does the marking. This inverts the usual arrangement and is worth more than any explanation it could give you."}]},{"t":"h2","id":"the-marker","runs":[{"text":"The marker"}]},{"t":"quote","runs":[{"text":"Here is the assessment brief and mark scheme, and here is my draft."}]},{"t":"quote","runs":[]},{"t":"quote","runs":[{"text":"---"}]},{"t":"quote","runs":[{"text":"[brief and mark scheme]"}]},{"t":"quote","runs":[{"text":"---"}]},{"t":"quote","runs":[{"text":"[draft]"}]},{"t":"quote","runs":[{"text":"---"}]},{"t":"quote","runs":[]},{"t":"quote","runs":[{"text":"Act as the marker. For each criterion, say which band this currently sits in and quote the sentence that decides it. Then give the three changes that would move the most marks, in order. Do not rewrite anything and do not be encouraging."}]},{"t":"p","runs":[{"text":"Do not rewrite anything","i":true},{"text":" is the essential clause. Without it you get an improved draft that is no longer yours."}]},{"t":"h2","id":"the-hint-ladder","runs":[{"text":"The hint ladder"}]},{"t":"quote","runs":[{"text":"I am stuck on this problem. Give me hints one at a time, starting with the smallest possible nudge: a question about what I should be looking at, not a step. Only give the next hint when I ask. Never give the full solution unless I type SOLVE."}]},{"t":"h2","id":"the-sceptic","runs":[{"text":"The sceptic"}]},{"t":"quote","runs":[{"text":"Below is an argument. Act as a hostile examiner. Identify the weakest step, the strongest counterargument, and any claim that is doing more work than the evidence supports. Do not soften anything and do not list strengths."}]},{"t":"p","runs":[{"text":"Use this in a fresh chat with the argument pasted in, not at the end of the thread that produced it."}]},{"t":"h2","id":"the-level-ladder","runs":[{"text":"The level ladder"}]},{"t":"quote","runs":[{"text":"Explain [topic] three times: first as you would to someone a level below me, then at my level using standard terminology, then telling me what the version at level two simplifies or hides and what a specialist would object to."}]},{"t":"h2","id":"the-reading-foothold","runs":[{"text":"The reading foothold"}]},{"t":"quote","runs":[{"text":"Here is a paragraph I cannot follow, and here is the sentence before and after it. Restate it in plain language without adding anything that is not there, then list the assumed background knowledge I appear to be missing. I will go back and read the original."}]},{"t":"h2","id":"standing-instructions","runs":[{"text":"Standing instructions"}]},{"t":"p","runs":[{"text":"Anything you would paste every time belongs in the tool’s custom-instructions field instead. Level, spelling convention, tone, and your standing preference for directness. Then your saved prompts can stay short."}]},{"t":"h2","id":"where-a-saved-prompt-goes-wrong","runs":[{"text":"Where a saved prompt goes wrong"}]},{"t":"p","runs":[{"text":"A saved prompt is a fixed instruction meeting a changing situation, so it fails in one predictable way: it keeps working after it should have stopped. The quiz prompt written when you knew nothing about a topic will still be asking recall questions in week nine, when you need application. The marker prompt aimed at a 1,500-word essay will happily be applied to a lab report with entirely different criteria."}]},{"t":"p","runs":[{"text":"So read the prompt before you paste it, every time, and change the one clause that no longer fits. That takes five seconds and is the difference between a tool and a ritual."}]},{"t":"h2","id":"the-maintenance-rule","runs":[{"text":"The maintenance rule"}]},{"t":"p","runs":[{"text":"Delete prompts you have not used in a month. A file of six live prompts gets used; a file of forty becomes a thing you scroll past. The best test of a prompt is whether you reached for it unprompted, and the best source of new ones is noticing the sentence you just typed that made an answer much better, and saving that."}]}],"check":{"question":[{"text":"In the quiz prompt, why does “ask me one question at a time and wait for my answer” matter so much?"}],"options":[{"text":[{"text":"It reduces token use, so a free tier’s allowance stretches further across a revision session."}],"correct":false,"tempting":[{"text":"True as a side effect and irrelevant to the learning mechanism; the same tokens spent on a question list would be equally cheap and far less useful."}]},{"text":[{"text":"Without it you receive questions and answers together, which is reading rather than retrieval."}],"correct":true,"tempting":null},{"text":[{"text":"It keeps the conversation short enough to avoid drift and truncation over a long revision session."}],"correct":false,"tempting":[{"text":"Names a real hazard of long threads but the wrong one here; an interactive quiz produces a longer conversation than a single list, not a shorter one."}]},{"text":[{"text":"It lets the model adapt each question’s difficulty to your previous answer, which is what makes the practice effective."}],"correct":false,"tempting":[{"text":"Adaptivity is a genuine bonus, but the technique works even with fixed questions; the effect comes from attempting recall before seeing the answer."}]}],"order":[2,0,3,1],"why":null},"prev":{"slug":"diagnosing-a-bad-answer","title":"Diagnosing a bad answer","minutes":8},"next":{"slug":"getting-unstuck-not-finished","title":"Getting Unstuck, Not Getting Finished","minutes":8}}}