{"course":{"slug":"ai-actually-explained","title":"AI, Actually Explained","url":"https://addaly.com/learn/ai-actually-explained","level":"beginner","track":"start"},"lesson":{"slug":"giving-it-what-it-needs","title":"Give it the material, not the question","url":"https://addaly.com/learn/ai-actually-explained/giving-it-what-it-needs","position":63,"of":92,"minutes":8,"answer":"The model can only work on what is in its context window, so the reliable way to use it is to hand it the document, the draft, the examples and the audience.","keyIdea":"The model can only work on what is in its context window, so the reliable way to use it is to hand it the document, the draft, the examples and the audience, and ask for work on those rather than facts from memory."},"body":{"format":"markdown","text":"## The empty desk\n\nThe chatbot module described the context window as the desk the model works on. Most people use it with an empty desk. They type a question and receive an answer assembled from memory, which is the least reliable thing the model does. The fix is mechanical: put the material on the desk first.\n\nThis lesson is the beginner's version of what the prompting course in this catalogue does in depth. It covers the four things that make the largest difference, and one that most people get backwards.\n\n## Give it the document\n\nIf the task is about a document, paste the document. A contract, an email thread, a page of notes, a spreadsheet exported as text, a transcript. A modern context window holds a hundred pages or more, and the model reads all of it before answering. \"Summarise the attached\" with the text present is a reliable task; \"summarise the report on X\" with nothing present is an invitation to invent a report.\n\nThe same applies to code, to a form, to a product description, to the syllabus you are asking about. If it is not on the desk, the model does not have it, however confidently it proceeds.\n\n## Give it the draft\n\nAsk for a rewrite of something you wrote rather than a first draft from nothing. The draft carries your facts, your structure and your voice, and the model's job becomes transformation: the reliable column from the previous lesson. A first draft from nothing is memory, and it will be generic for the reason the lesson on where the text came from explained: generic is the likeliest continuation.\n\n## Give it examples\n\nTwo or three examples of what you want are worth more than a paragraph describing it. Show it three product descriptions in your house style and ask for a fourth. Show it two emails you sent that landed well and ask for one to a new person. The model infers the pattern from the examples in the same way it learned everything else: from instances. This is called few-shot prompting and it is the single most effective technique that costs nothing to learn.\n\n## Tell it who it is for, and how long\n\n\"Explain this to a customer who has never used the product, in under a hundred words\" produces something usable. \"Explain this\" produces the median explanation on the web, at the median length, for the median reader. The model has no idea who you are writing for unless you say, and the audience changes everything about a good answer.\n\n## Then argue with it\n\nThe first answer is a draft. Say what is wrong with it. \"Too formal.\" \"You dropped the second point.\" \"Shorter, and lead with the deadline.\" Each turn keeps the previous ones on the desk, so the model revises rather than starts over. Most people stop after one exchange, and most of the value is in the second and third.\n\n## The thing people get backwards\n\nDo not ask the model for facts and then work on them. Get the facts yourself, from a source you trust, put them on the desk, and ask the model to work on them. The reason is the entire fourth module of this course: facts from memory are invented in the specifics, and once an invented specific is in the conversation, every subsequent turn treats it as true. The model is a superb writer and an unreliable witness. Use it as the first and not the second.\n\n## A worked example\n\nInstead of: *\"Write a letter to my landlord about the leak.\"*\n\nTry: *\"Here is the clause of my tenancy agreement about repairs. [paste] Here is the email I sent on 3 March and the reply I got. [paste] Write a firm, polite letter, under two hundred words, citing the clause, giving a deadline of fourteen days, and asking for written confirmation. Do not invent any dates or amounts that are not in the material above.\"*\n\nThe second version is longer to type and takes the model thirty seconds. The first produces a letter with a plausible clause number, a plausible date and a plausible deadline, none of which is yours.\n\n## What context does not fix\n\nPutting material on the desk makes the model's reading reliable. It does not make its judgement reliable, and it does not fix the jagged frontier. A model with your whole contract in view will still occasionally misread a clause, and it will do so fluently. The material makes checking possible; it does not make checking unnecessary."},"check":{"question":"You want a summary of a twenty-page report. Which request is most likely to produce a reliable result, and why?","options":[{"text":"\"Summarise the 2024 annual report of [company]\": the model has read many annual reports and will know this one.","correct":false,"tempting":"Leaves the desk empty; the model will assemble a plausible report from memory rather than the real one."},{"text":"Paste the full report and ask for a summary of under two hundred words for a named reader.","correct":true},{"text":"Paste the report's title and table of contents so the model knows the structure.","correct":false,"tempting":"Gives structure without content; the sections will be filled with the likeliest generic material."},{"text":"Ask the model what it knows about the company first, then ask for the summary.","correct":false,"tempting":"Front-loads memory into the conversation, so any invented detail is then carried into the summary as fact."}]},"prev":{"slug":"the-shape-of-a-good-task","title":"The shape of a task it is good at","url":"https://addaly.com/learn/ai-actually-explained/the-shape-of-a-good-task"},"next":{"slug":"where-your-words-go","title":"Where your words go","url":"https://addaly.com/learn/ai-actually-explained/where-your-words-go"},"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) 'Give it the material, not the question', in AI, Actually Explained. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-actually-explained/giving-it-what-it-needs (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":"Give it the material, not the question","courseTitle":"AI, Actually Explained","module":{"index":7,"slug":"putting-it-to-work","title":"Putting it to work without getting burned"},"answer":null,"keyIdea":"The model can only work on what is in its context window, so the reliable way to use it is to hand it the document, the draft, the examples and the audience, and ask for work on those rather than facts from memory.","hero":{"kind":"plate","image":"https://addaly.com/art/modules/ai-actually-explained/putting-it-to-work.webp","alt":"A coaching-centre desk with a stack of books, a phone, an open folder of notes, a steel tumbler and a plain wall calendar above.","caption":"Using these tools well is mostly a matter of what you put on the desk beside them and what you decide never to hand over."},"blocks":[{"t":"h2","id":"the-empty-desk","runs":[{"text":"The empty desk"}]},{"t":"p","runs":[{"text":"The chatbot module described the context window as the desk the model works on. Most people use it with an empty desk. They type a question and receive an answer assembled from memory, which is the least reliable thing the model does. The fix is mechanical: put the material on the desk first."}]},{"t":"p","runs":[{"text":"This lesson is the beginner’s version of what the prompting course in this catalogue does in depth. It covers the four things that make the largest difference, and one that most people get backwards."}]},{"t":"h2","id":"give-it-the-document","runs":[{"text":"Give it the document"}]},{"t":"p","runs":[{"text":"If the task is about a document, paste the document. A contract, an email thread, a page of notes, a spreadsheet exported as text, a transcript. A modern context window holds a hundred pages or more, and the model reads all of it before answering. “Summarise the attached” with the text present is a reliable task; “summarise the report on X” with nothing present is an invitation to invent a report."}]},{"t":"p","runs":[{"text":"The same applies to code, to a form, to a product description, to the syllabus you are asking about. If it is not on the desk, the model does not have it, however confidently it proceeds."}]},{"t":"h2","id":"give-it-the-draft","runs":[{"text":"Give it the draft"}]},{"t":"p","runs":[{"text":"Ask for a rewrite of something you wrote rather than a first draft from nothing. The draft carries your facts, your structure and your voice, and the model’s job becomes transformation: the reliable column from the previous lesson. A first draft from nothing is memory, and it will be generic for the reason the lesson on where the text came from explained: generic is the likeliest continuation."}]},{"t":"h2","id":"give-it-examples","runs":[{"text":"Give it examples"}]},{"t":"p","runs":[{"text":"Two or three examples of what you want are worth more than a paragraph describing it. Show it three product descriptions in your house style and ask for a fourth. Show it two emails you sent that landed well and ask for one to a new person. The model infers the pattern from the examples in the same way it learned everything else: from instances. This is called few-shot prompting and it is the single most effective technique that costs nothing to learn."}]},{"t":"h2","id":"tell-it-who-it-is-for-and-how-long","runs":[{"text":"Tell it who it is for, and how long"}]},{"t":"p","runs":[{"text":"“Explain this to a customer who has never used the product, in under a hundred words” produces something usable. “Explain this” produces the median explanation on the web, at the median length, for the median reader. The model has no idea who you are writing for unless you say, and the audience changes everything about a good answer."}]},{"t":"h2","id":"then-argue-with-it","runs":[{"text":"Then argue with it"}]},{"t":"p","runs":[{"text":"The first answer is a draft. Say what is wrong with it. “Too formal.” “You dropped the second point.” “Shorter, and lead with the deadline.” Each turn keeps the previous ones on the desk, so the model revises rather than starts over. Most people stop after one exchange, and most of the value is in the second and third."}]},{"t":"h2","id":"the-thing-people-get-backwards","runs":[{"text":"The thing people get backwards"}]},{"t":"p","runs":[{"text":"Do not ask the model for facts and then work on them. Get the facts yourself, from a source you trust, put them on the desk, and ask the model to work on them. The reason is the entire fourth module of this course: facts from memory are invented in the specifics, and once an invented specific is in the conversation, every subsequent turn treats it as true. The model is a superb writer and an unreliable witness. Use it as the first and not the second."}]},{"t":"h2","id":"a-worked-example","runs":[{"text":"A worked example"}]},{"t":"p","runs":[{"text":"Instead of: "},{"text":"”Write a letter to my landlord about the leak.”","i":true}]},{"t":"p","runs":[{"text":"Try: "},{"text":"”Here is the clause of my tenancy agreement about repairs. [paste] Here is the email I sent on 3 March and the reply I got. [paste] Write a firm, polite letter, under two hundred words, citing the clause, giving a deadline of fourteen days, and asking for written confirmation. Do not invent any dates or amounts that are not in the material above.”","i":true}]},{"t":"p","runs":[{"text":"The second version is longer to type and takes the model thirty seconds. The first produces a letter with a plausible clause number, a plausible date and a plausible deadline, none of which is yours."}]},{"t":"h2","id":"what-context-does-not-fix","runs":[{"text":"What context does not fix"}]},{"t":"p","runs":[{"text":"Putting material on the desk makes the model’s reading reliable. It does not make its judgement reliable, and it does not fix the jagged frontier. A model with your whole contract in view will still occasionally misread a clause, and it will do so fluently. The material makes checking possible; it does not make checking unnecessary."}]}],"check":{"question":[{"text":"You want a summary of a twenty-page report. Which request is most likely to produce a reliable result, and why?"}],"options":[{"text":[{"text":"“Summarise the 2024 annual report of [company]”: the model has read many annual reports and will know this one."}],"correct":false,"tempting":[{"text":"Leaves the desk empty; the model will assemble a plausible report from memory rather than the real one."}]},{"text":[{"text":"Paste the full report and ask for a summary of under two hundred words for a named reader."}],"correct":true,"tempting":null},{"text":[{"text":"Paste the report’s title and table of contents so the model knows the structure."}],"correct":false,"tempting":[{"text":"Gives structure without content; the sections will be filled with the likeliest generic material."}]},{"text":[{"text":"Ask the model what it knows about the company first, then ask for the summary."}],"correct":false,"tempting":[{"text":"Front-loads memory into the conversation, so any invented detail is then carried into the summary as fact."}]}],"order":[0,1,3,2],"why":null},"prev":{"slug":"the-shape-of-a-good-task","title":"The shape of a task it is good at","minutes":9},"next":{"slug":"where-your-words-go","title":"Where your words go","minutes":10}}}