{"course":{"slug":"prompting","title":"Getting Real Work out of a Model","url":"https://addaly.com/learn/prompting","level":"beginner","track":"start"},"lesson":{"slug":"what-to-leave-out","title":"Choosing what not to include","url":"https://addaly.com/learn/prompting/what-to-leave-out","position":17,"of":72,"minutes":8,"answer":"Include what feeds a line of the answer you expect and cut the rest.","keyIdea":"Include what feeds a line of the answer you expect and cut the rest, because irrelevant material does not sit harmlessly in the context: it dilutes attention and supplies wrong facts with the same confidence as right ones."},"body":{"format":"markdown","text":"## The skill nobody teaches\n\nEveryone tells you to give more context. Almost nobody tells you what to cut, and by the second month of using these tools seriously that becomes the binding constraint. You have three reports, a spreadsheet, an email thread and a policy document, and the honest question is which two pages of that belong in the prompt.\n\nThree reasons cutting matters, and only the first is about limits.\n\n**The window is finite.** Obvious, and the least important, since windows are now large.\n\n**Accuracy falls as the pile grows.** From the position lesson: relevant material is used less reliably when it is buried among irrelevant material. A model given ten documents to answer a question about one of them does measurably worse than a model given the one.\n\n**Irrelevant context actively misleads.** This is the underrated part. Include last year's pricing sheet \"for background\" and some of last year's prices will appear in this year's answer, presented as current. The model has no way to know which document you consider authoritative unless you say so.\n\n> Figure: Answering a question about one document. Bars (% correct): The one document 92; Plus two related documents 86; Plus nine related documents 74; Plus nineteen, 'for background' 63. Accuracy falls as the pile grows, and the extra documents do not sit harmlessly: last year's pricing sheet, included for background, supplies last year's prices with this year's confidence.\n\n## A method that takes two minutes\n\nBefore assembling a long prompt, write the answer's skeleton. Not the answer: the shape of it. \"A recommendation between supplier A and B, based on unit price, lead time and minimum order.\"\n\nNow go through your material and ask of each piece: **which line of the skeleton does this feed?** Anything that feeds nothing comes out.\n\nApplied to the supplier decision: the two quotes go in. The email thread negotiating a different contract does not. Your notes from the site visit go in if they mention lead time. The company handbook does not, however relevant it feels.\n\nYou will cut about two thirds. The answer gets better, not worse, and you will have found that you were including things because you had them, not because they mattered.\n\n## When you cannot cut, do it in two passes\n\nSome questions need the whole corpus: finding a contradiction, checking whether anything covers a topic, summarising. Do not choose in advance; narrow in the open.\n\n**Pass one, enumeration.** \"List every section that mentions payment terms, with its heading and clause number. Do not analyse.\" Cheap and hard to get wrong.\n\n**Pass two, the real question**, with only the sections pass one found.\n\nThis is the same move a competent researcher makes with a paper index, and it works for the same reason: locating and reasoning are different jobs, and doing them at once does neither well.\n\nIt also gives you something to check. If pass one lists four sections and you know there are six, you have caught the failure before it reached your conclusion, which you could not have done from a fluent final answer that quietly used four.\n\n## Rank what you do include\n\nWhen several documents are in the prompt and they disagree, say which wins.\n\n> `<policy2026>` is authoritative. `<policy2024>` is included only so you can tell me what changed. Where they conflict, follow 2026 and note the difference.\n\nWithout that line, you get a blend, and a blend of two policies is a policy that does not exist. Nobody wrote it, nobody approved it, and it will read as authoritative as either source.\n\n## The signs you have included too much\n\n- The answer mixes facts that come from different documents without saying which.\n- Figures appear that are correct for some other year, region or version.\n- The response gets longer and vaguer as you add material.\n- The model starts hedging: \"depending on which policy applies…\". That sentence is a symptom, and it means you left the disambiguation to it.\n\n## The counterweight\n\nNone of this licenses going back to describing your problem instead of showing it. The failure this lesson describes is rarer than under-supplying context, and it appears later, once someone has learned the first lesson well.\n\nThe target is not minimal and not maximal. It is **everything that bears on the answer, and nothing that does not**, with a stated order of authority when two pieces disagree. Getting there requires knowing what the answer looks like before you ask, which is why the two-minute skeleton is the whole method."},"check":{"question":"An analyst pastes this year's and last year's price lists \"for completeness\" and asks for a quote for a customer. The quote comes back with two prices that match last year's list. Why?","options":[{"text":"Both lists were equally valid material as far as the model could tell, because no line stated which one governs.","correct":true},{"text":"The model preferred the older document because it appeared first in the prompt.","correct":false,"tempting":"Position affects retrieval, but the deeper problem is that nothing told the model which document was authoritative."},{"text":"Price lists are tabular, and tables are unreliable when converted to text.","correct":false,"tempting":"A real hazard, but it would produce garbled figures rather than figures that are precisely correct for the wrong year."},{"text":"The question was ambiguous about which customer segment the quote was for.","correct":false,"tempting":"Segment ambiguity would produce a hedge or a question, not a systematic pull from the superseded list."}]},"prev":{"slug":"what-happens-to-an-attachment","title":"What actually happens when you attach a file","url":"https://addaly.com/learn/prompting/what-happens-to-an-attachment"},"next":{"slug":"saying-what-not-to-do","title":"Negative instructions, and why they half-work","url":"https://addaly.com/learn/prompting/saying-what-not-to-do"},"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) 'Choosing what not to include', in Getting Real Work out of a Model. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/prompting/what-to-leave-out (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":"Choosing what not to include","courseTitle":"Getting Real Work out of a Model","module":{"index":2,"slug":"the-material-you-give-it","title":"The material you give it"},"answer":null,"keyIdea":"Include what feeds a line of the answer you expect and cut the rest, because irrelevant material does not sit harmlessly in the context: it dilutes attention and supplies wrong facts with the same confidence as right ones.","hero":{"kind":"figure","figure":{"t":"figure","label":"Figure","figure":{"type":"bars","unit":"% correct","title":"Answering a question about one document","series":[{"label":"The one document","value":92,"highlight":true},{"label":"Plus two related documents","value":86},{"label":"Plus nine related documents","value":74},{"label":"Plus nineteen, ‘for background’","value":63}],"caption":"Accuracy falls as the pile grows, and the extra documents do not sit harmlessly: last year’s pricing sheet, included for background, supplies last year’s prices with this year’s confidence."},"alt":"Figure: Answering a question about one document. Bars (% correct): The one document 92; Plus two related documents 86; Plus nine related documents 74; Plus nineteen, ‘for background’ 63. Accuracy falls as the pile grows, and the extra documents do not sit harmlessly: last year’s pricing sheet, included for background, supplies last year’s prices with this year’s confidence."}},"blocks":[{"t":"h2","id":"the-skill-nobody-teaches","runs":[{"text":"The skill nobody teaches"}]},{"t":"p","runs":[{"text":"Everyone tells you to give more context. Almost nobody tells you what to cut, and by the second month of using these tools seriously that becomes the binding constraint. You have three reports, a spreadsheet, an email thread and a policy document, and the honest question is which two pages of that belong in the prompt."}]},{"t":"p","runs":[{"text":"Three reasons cutting matters, and only the first is about limits."}]},{"t":"p","runs":[{"text":"The window is finite.","b":true},{"text":" Obvious, and the least important, since windows are now large."}]},{"t":"p","runs":[{"text":"Accuracy falls as the pile grows.","b":true},{"text":" From the position lesson: relevant material is used less reliably when it is buried among irrelevant material. A model given ten documents to answer a question about one of them does measurably worse than a model given the one."}]},{"t":"p","runs":[{"text":"Irrelevant context actively misleads.","b":true},{"text":" This is the underrated part. Include last year’s pricing sheet “for background” and some of last year’s prices will appear in this year’s answer, presented as current. The model has no way to know which document you consider authoritative unless you say so."}]},{"t":"h2","id":"a-method-that-takes-two-minutes","runs":[{"text":"A method that takes two minutes"}]},{"t":"p","runs":[{"text":"Before assembling a long prompt, write the answer’s skeleton. Not the answer: the shape of it. “A recommendation between supplier A and B, based on unit price, lead time and minimum order.”"}]},{"t":"p","runs":[{"text":"Now go through your material and ask of each piece: "},{"text":"which line of the skeleton does this feed?","b":true},{"text":" Anything that feeds nothing comes out."}]},{"t":"p","runs":[{"text":"Applied to the supplier decision: the two quotes go in. The email thread negotiating a different contract does not. Your notes from the site visit go in if they mention lead time. The company handbook does not, however relevant it feels."}]},{"t":"p","runs":[{"text":"You will cut about two thirds. The answer gets better, not worse, and you will have found that you were including things because you had them, not because they mattered."}]},{"t":"h2","id":"when-you-cannot-cut-do-it-in-two-passes","runs":[{"text":"When you cannot cut, do it in two passes"}]},{"t":"p","runs":[{"text":"Some questions need the whole corpus: finding a contradiction, checking whether anything covers a topic, summarising. Do not choose in advance; narrow in the open."}]},{"t":"p","runs":[{"text":"Pass one, enumeration.","b":true},{"text":" “List every section that mentions payment terms, with its heading and clause number. Do not analyse.” Cheap and hard to get wrong."}]},{"t":"p","runs":[{"text":"Pass two, the real question","b":true},{"text":", with only the sections pass one found."}]},{"t":"p","runs":[{"text":"This is the same move a competent researcher makes with a paper index, and it works for the same reason: locating and reasoning are different jobs, and doing them at once does neither well."}]},{"t":"p","runs":[{"text":"It also gives you something to check. If pass one lists four sections and you know there are six, you have caught the failure before it reached your conclusion, which you could not have done from a fluent final answer that quietly used four."}]},{"t":"h2","id":"rank-what-you-do-include","runs":[{"text":"Rank what you do include"}]},{"t":"p","runs":[{"text":"When several documents are in the prompt and they disagree, say which wins."}]},{"t":"quote","runs":[{"text":"<policy2026>","code":true},{"text":" is authoritative. "},{"text":"<policy2024>","code":true},{"text":" is included only so you can tell me what changed. Where they conflict, follow 2026 and note the difference."}]},{"t":"p","runs":[{"text":"Without that line, you get a blend, and a blend of two policies is a policy that does not exist. Nobody wrote it, nobody approved it, and it will read as authoritative as either source."}]},{"t":"h2","id":"the-signs-you-have-included-too-much","runs":[{"text":"The signs you have included too much"}]},{"t":"ul","items":[[{"text":"The answer mixes facts that come from different documents without saying which."}],[{"text":"Figures appear that are correct for some other year, region or version."}],[{"text":"The response gets longer and vaguer as you add material."}],[{"text":"The model starts hedging: “depending on which policy applies…”. That sentence is a symptom, and it means you left the disambiguation to it."}]]},{"t":"h2","id":"the-counterweight","runs":[{"text":"The counterweight"}]},{"t":"p","runs":[{"text":"None of this licenses going back to describing your problem instead of showing it. The failure this lesson describes is rarer than under-supplying context, and it appears later, once someone has learned the first lesson well."}]},{"t":"p","runs":[{"text":"The target is not minimal and not maximal. It is "},{"text":"everything that bears on the answer, and nothing that does not","b":true},{"text":", with a stated order of authority when two pieces disagree. Getting there requires knowing what the answer looks like before you ask, which is why the two-minute skeleton is the whole method."}]}],"check":{"question":[{"text":"An analyst pastes this year’s and last year’s price lists “for completeness” and asks for a quote for a customer. The quote comes back with two prices that match last year’s list. Why?"}],"options":[{"text":[{"text":"Both lists were equally valid material as far as the model could tell, because no line stated which one governs."}],"correct":true,"tempting":null},{"text":[{"text":"The model preferred the older document because it appeared first in the prompt."}],"correct":false,"tempting":[{"text":"Position affects retrieval, but the deeper problem is that nothing told the model which document was authoritative."}]},{"text":[{"text":"Price lists are tabular, and tables are unreliable when converted to text."}],"correct":false,"tempting":[{"text":"A real hazard, but it would produce garbled figures rather than figures that are precisely correct for the wrong year."}]},{"text":[{"text":"The question was ambiguous about which customer segment the quote was for."}],"correct":false,"tempting":[{"text":"Segment ambiguity would produce a hedge or a question, not a systematic pull from the superseded list."}]}],"order":[2,0,1,3],"why":null},"prev":{"slug":"what-happens-to-an-attachment","title":"What actually happens when you attach a file","minutes":9},"next":{"slug":"saying-what-not-to-do","title":"Negative instructions, and why they half-work","minutes":8}}}