{"course":{"slug":"ai-for-small-business","title":"AI for a Small Business","url":"https://addaly.com/learn/ai-for-small-business","level":"beginner","track":"work"},"lesson":{"slug":"the-five-part-request","title":"Five parts a request should always have","url":"https://addaly.com/learn/ai-for-small-business/the-five-part-request","position":11,"of":88,"minutes":8,"answer":"A good request states situation, facts, one task, constraints and format, and the part most people skimp, the raw facts, is exactly the part whose absence produces the empty prose that reads as generated.","keyIdea":"A good request states situation, facts, one task, constraints and format, and the part most people skimp, the raw facts, is exactly the part whose absence produces the empty prose that reads as generated."},"body":{"format":"markdown","text":"## Why the first attempt disappoints\n\nMost disappointing output comes from an underspecified request rather than a weak model. \"Write a post about our new opening hours\" leaves five decisions to the machine (length, audience, tone, format, what to emphasise) and it will make all five, reasonably and generically.\n\nMake the five decisions yourself and the same model produces something usable. The structure below is not a formula to memorise; it is a checklist for what you forgot.\n\n### 1. Situation\n\nWho is writing and to whom. One line. \"I run a two-chair barbershop in Leeds. This is going to regular customers who have been coming for years.\"\n\nThis does more than set tone. It tells the model what can be assumed and what must be explained, which is the difference between a message that sounds like you and one that sounds like a company.\n\n### 2. Facts\n\nThe raw material. Your facts sheet, plus whatever is specific to this job. Messy is fine: notes, a pasted message, bullet points, a voice note transcript. It does not need to be tidy going in; tidying is the job you are handing over.\n\nThis is the part people skimp, and skimping here is what produces the empty prose everyone recognises as generated.\n\n### 3. Task\n\nOne verb. Write, rewrite, shorten, sort, extract, translate, list, compare. If your request has three verbs it is three requests, and you will get a mediocre attempt at all three. Split them.\n\n### 4. Constraints\n\nThe boundaries. Length in words rather than \"short\", because short means nothing. Things to avoid. Things that must appear. And the honesty rule from the next lesson but one.\n\n```text\nUnder 70 words. No exclamation marks. Do not mention price.\nMust include the Tuesday closure. If anything is missing\nfrom my notes, write [CHECK] rather than guessing.\n```\n\n### 5. Format\n\nWhat you want handed back. A paragraph, five bullets, a table of two columns, a subject line and a body, three options to choose between.\n\nAsking for three options is underused. It costs the same, and choosing between three is faster and produces better results than editing one, because you can see the range rather than judging a single sample in a vacuum.\n\n## The same request, twice\n\nBefore:\n\n```text\nWrite something about our summer sale.\n```\n\nAfter:\n\n```text\nI run a small fabric shop in Ahmedabad. This is for our\nWhatsApp broadcast list — mostly tailors and regular\ncustomers, people who know us.\n\nFacts: 20% off all cotton and linen, 5 to 19 June.\nNot silk, not the imported wools. Shop closed 12 June\nfor a wedding. New consignment of block prints arrived\nlast week — that is the real news.\n\nTask: write the broadcast message.\n\nConstraints: under 80 words. Plain Hindi and English\nmixed the way we speak. No emoji. Lead with the block\nprints, not the discount. If a date or figure is missing,\nwrite [CHECK].\n\nFormat: the message only, no preamble.\n```\n\nThe second takes ninety seconds to write, and once written it becomes a template where only the facts change each time.\n\n## Two habits worth forming\n\n**Say what to leave out.** Negative constraints are strong and people rarely use them. \"Do not mention the price.\" \"No closing question.\" \"Do not apologise.\" That last one is useful in complaint replies, where the default behaviour is to concede fault you have not established.\n\n**Ask for the format you will actually paste.** \"Just the message, no preamble\" saves you deleting \"Here's a draft for your consideration!\" forty times a month. Small, but you will do this hundreds of times.\n\n## What this is not\n\nThere is a genre of content promising magic phrases: the prompt that unlocks hidden capability. Ignore it. There is no secret sentence. Specific input, clear constraints and real examples account for nearly all of the difference between good and bad output, and all three are things you already know about your own business.\n\n## Turn it into a template once\n\nWrite the five parts properly one time, then replace only the facts. Situation, constraints and format almost never change for a given task; only the raw material does. Keep the skeleton with a gap in it:\n\n```text\n[situation - fixed]\nFACTS: [paste the sheet - fixed]\nNEW: <<< paste today's message here >>>\n[task - fixed]\n[constraints - fixed]\n[format - fixed]\n```\n\nAfter a fortnight you will be pasting one thing into one slot, which takes four seconds and gets you the output of a ninety-second brief.\n\n## When you do not know what to specify\n\nFor an unfamiliar task, invert it. Describe the situation and end with: *before writing anything, ask me the five questions you most need answered.*\n\nThe questions that come back are usually the specification you could not write. Most people find two of the five are things they had not decided: who the audience actually is, what the piece is supposed to achieve. Answer them, and only then ask for the draft. It costs one extra turn and it removes the most common cause of a disappointing first attempt, which is that the request was underspecified because the requester was."},"check":{"question":"An owner writes a detailed prompt that asks the model to summarise a customer complaint, draft a reply, and suggest a policy change to prevent it recurring. The output is weak on all three. What is the most likely cause?","options":[{"text":"The prompt lacks sufficient context about the business, so the model had nothing specific to work with.","correct":false,"tempting":"Names a real failure but not this one; the prompt is described as detailed, and adding more context to a three-task request produces three slightly better mediocre answers."},{"text":"Complaint handling requires judgement, so this task should not have been given to a model at all.","correct":false,"tempting":"Overcorrects into a boundary rule; drafting a reply from stated facts is ordinary text work, and it is the bundling rather than the subject that failed here."},{"text":"Three verbs in one request means three tasks competing for one response, so each gets a fraction of the attention and none is done properly.","correct":true},{"text":"The model handles analysis better than generation, so summarising and drafting should not be mixed in one request without splitting the analytical work into its own conversation first.","correct":false,"tempting":"Invents a capability split that does not exist; both are generation, and the same request would fail with three summaries just as it fails with a summary and two drafts."}]},"prev":{"slug":"the-facts-sheet","title":"The one page that travels with every request","url":"https://addaly.com/learn/ai-for-small-business/the-facts-sheet"},"next":{"slug":"examples-beat-instructions","title":"Five examples teach what adjectives cannot","url":"https://addaly.com/learn/ai-for-small-business/examples-beat-instructions"},"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) 'Five parts a request should always have', in AI for a Small Business. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-for-small-business/the-five-part-request (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":"Five parts a request should always have","courseTitle":"AI for a Small Business","module":{"index":2,"slug":"asking-properly","title":"Asking properly: prompting for business work"},"answer":null,"keyIdea":"A good request states situation, facts, one task, constraints and format, and the part most people skimp, the raw facts, is exactly the part whose absence produces the empty prose that reads as generated.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"why-the-first-attempt-disappoints","runs":[{"text":"Why the first attempt disappoints"}]},{"t":"p","runs":[{"text":"Most disappointing output comes from an underspecified request rather than a weak model. “Write a post about our new opening hours” leaves five decisions to the machine (length, audience, tone, format, what to emphasise) and it will make all five, reasonably and generically."}]},{"t":"p","runs":[{"text":"Make the five decisions yourself and the same model produces something usable. The structure below is not a formula to memorise; it is a checklist for what you forgot."}]},{"t":"h3","id":"1-situation","runs":[{"text":"1. Situation"}]},{"t":"p","runs":[{"text":"Who is writing and to whom. One line. “I run a two-chair barbershop in Leeds. This is going to regular customers who have been coming for years.”"}]},{"t":"p","runs":[{"text":"This does more than set tone. It tells the model what can be assumed and what must be explained, which is the difference between a message that sounds like you and one that sounds like a company."}]},{"t":"h3","id":"2-facts","runs":[{"text":"2. Facts"}]},{"t":"p","runs":[{"text":"The raw material. Your facts sheet, plus whatever is specific to this job. Messy is fine: notes, a pasted message, bullet points, a voice note transcript. It does not need to be tidy going in; tidying is the job you are handing over."}]},{"t":"p","runs":[{"text":"This is the part people skimp, and skimping here is what produces the empty prose everyone recognises as generated."}]},{"t":"h3","id":"3-task","runs":[{"text":"3. Task"}]},{"t":"p","runs":[{"text":"One verb. Write, rewrite, shorten, sort, extract, translate, list, compare. If your request has three verbs it is three requests, and you will get a mediocre attempt at all three. Split them."}]},{"t":"h3","id":"4-constraints","runs":[{"text":"4. Constraints"}]},{"t":"p","runs":[{"text":"The boundaries. Length in words rather than “short”, because short means nothing. Things to avoid. Things that must appear. And the honesty rule from the next lesson but one."}]},{"t":"code","lang":"text","label":"Text","text":"Under 70 words. No exclamation marks. Do not mention price.\nMust include the Tuesday closure. If anything is missing\nfrom my notes, write [CHECK] rather than guessing."},{"t":"h3","id":"5-format","runs":[{"text":"5. Format"}]},{"t":"p","runs":[{"text":"What you want handed back. A paragraph, five bullets, a table of two columns, a subject line and a body, three options to choose between."}]},{"t":"p","runs":[{"text":"Asking for three options is underused. It costs the same, and choosing between three is faster and produces better results than editing one, because you can see the range rather than judging a single sample in a vacuum."}]},{"t":"h2","id":"the-same-request-twice","runs":[{"text":"The same request, twice"}]},{"t":"p","runs":[{"text":"Before:"}]},{"t":"code","lang":"text","label":"Text","text":"Write something about our summer sale."},{"t":"p","runs":[{"text":"After:"}]},{"t":"code","lang":"text","label":"Text","text":"I run a small fabric shop in Ahmedabad. This is for our\nWhatsApp broadcast list — mostly tailors and regular\ncustomers, people who know us.\n\nFacts: 20% off all cotton and linen, 5 to 19 June.\nNot silk, not the imported wools. Shop closed 12 June\nfor a wedding. New consignment of block prints arrived\nlast week — that is the real news.\n\nTask: write the broadcast message.\n\nConstraints: under 80 words. Plain Hindi and English\nmixed the way we speak. No emoji. Lead with the block\nprints, not the discount. If a date or figure is missing,\nwrite [CHECK].\n\nFormat: the message only, no preamble."},{"t":"p","runs":[{"text":"The second takes ninety seconds to write, and once written it becomes a template where only the facts change each time."}]},{"t":"h2","id":"two-habits-worth-forming","runs":[{"text":"Two habits worth forming"}]},{"t":"p","runs":[{"text":"Say what to leave out.","b":true},{"text":" Negative constraints are strong and people rarely use them. “Do not mention the price.” “No closing question.” “Do not apologise.” That last one is useful in complaint replies, where the default behaviour is to concede fault you have not established."}]},{"t":"p","runs":[{"text":"Ask for the format you will actually paste.","b":true},{"text":" “Just the message, no preamble” saves you deleting “Here’s a draft for your consideration!” forty times a month. Small, but you will do this hundreds of times."}]},{"t":"h2","id":"what-this-is-not","runs":[{"text":"What this is not"}]},{"t":"p","runs":[{"text":"There is a genre of content promising magic phrases: the prompt that unlocks hidden capability. Ignore it. There is no secret sentence. Specific input, clear constraints and real examples account for nearly all of the difference between good and bad output, and all three are things you already know about your own business."}]},{"t":"h2","id":"turn-it-into-a-template-once","runs":[{"text":"Turn it into a template once"}]},{"t":"p","runs":[{"text":"Write the five parts properly one time, then replace only the facts. Situation, constraints and format almost never change for a given task; only the raw material does. Keep the skeleton with a gap in it:"}]},{"t":"code","lang":"text","label":"Text","text":"[situation - fixed]\nFACTS: [paste the sheet - fixed]\nNEW: <<< paste today's message here >>>\n[task - fixed]\n[constraints - fixed]\n[format - fixed]"},{"t":"p","runs":[{"text":"After a fortnight you will be pasting one thing into one slot, which takes four seconds and gets you the output of a ninety-second brief."}]},{"t":"h2","id":"when-you-do-not-know-what-to-specify","runs":[{"text":"When you do not know what to specify"}]},{"t":"p","runs":[{"text":"For an unfamiliar task, invert it. Describe the situation and end with: "},{"text":"before writing anything, ask me the five questions you most need answered.","i":true}]},{"t":"p","runs":[{"text":"The questions that come back are usually the specification you could not write. Most people find two of the five are things they had not decided: who the audience actually is, what the piece is supposed to achieve. Answer them, and only then ask for the draft. It costs one extra turn and it removes the most common cause of a disappointing first attempt, which is that the request was underspecified because the requester was."}]}],"check":{"question":[{"text":"An owner writes a detailed prompt that asks the model to summarise a customer complaint, draft a reply, and suggest a policy change to prevent it recurring. The output is weak on all three. What is the most likely cause?"}],"options":[{"text":[{"text":"The prompt lacks sufficient context about the business, so the model had nothing specific to work with."}],"correct":false,"tempting":[{"text":"Names a real failure but not this one; the prompt is described as detailed, and adding more context to a three-task request produces three slightly better mediocre answers."}]},{"text":[{"text":"Complaint handling requires judgement, so this task should not have been given to a model at all."}],"correct":false,"tempting":[{"text":"Overcorrects into a boundary rule; drafting a reply from stated facts is ordinary text work, and it is the bundling rather than the subject that failed here."}]},{"text":[{"text":"Three verbs in one request means three tasks competing for one response, so each gets a fraction of the attention and none is done properly."}],"correct":true,"tempting":null},{"text":[{"text":"The model handles analysis better than generation, so summarising and drafting should not be mixed in one request without splitting the analytical work into its own conversation first."}],"correct":false,"tempting":[{"text":"Invents a capability split that does not exist; both are generation, and the same request would fail with three summaries just as it fails with a summary and two drafts."}]}],"order":[3,0,1,2],"why":null},"prev":{"slug":"the-facts-sheet","title":"The one page that travels with every request","minutes":9},"next":{"slug":"examples-beat-instructions","title":"Five examples teach what adjectives cannot","minutes":9}}}