{"course":{"slug":"prompting","title":"Getting Real Work out of a Model","url":"https://addaly.com/learn/prompting","level":"beginner","track":"start"},"lesson":{"slug":"choosing-the-container","title":"Choosing the container","url":"https://addaly.com/learn/prompting/choosing-the-container","position":30,"of":72,"minutes":8,"answer":"Choose the container by what you will do with the answer next, and counteract the bias of whichever you chose: tables invent comparability, lists flatten importance, prose hides omissions.","keyIdea":"Choose the container by what you will do with the answer next, and counteract the bias of whichever you chose: tables invent comparability, lists flatten importance, prose hides omissions."},"body":{"format":"markdown","text":"## The container is a decision about what you will do next\n\nAsk for prose and you get an argument you have to read. Ask for a table and you get facts you can scan. Neither is better; they serve different next actions, and choosing badly is why a good answer still costs you twenty minutes.\n\nWork backwards from what happens after you read it.\n\n## The five containers you actually need\n\n**Prose** when you need reasoning you will weigh. Advice, an explanation, an argument about a judgement call. Prose has room for qualification, and qualification is the point.\n\n**A table** when you will compare like against like, and the comparison is the answer. Name the columns. Every cell must then be filled with something specific about that row, which is where a table earns its keep: vagueness has nowhere to sit.\n\n**A numbered list** when order matters or you will act on the items one at a time. Steps, priorities, a checklist. Specify the count (\"exactly five\") if you want to force selection rather than a dump.\n\n**A bulleted list** when the items are peers and order does not matter. Underused correctly and overused generally: most of the bulleted lists people receive should have been two sentences of prose, and the model produces them because bullets pattern-match to helpfulness.\n\n**A machine format** (CSV, JSON, a fixed line format) when the output goes into a spreadsheet, a script, or another system. Next lesson.\n\nThere is a sixth worth naming: **the thing itself**. The email, ready to send. The message, ready to paste. No commentary before or after. \"Give me only the message\" is a format instruction and people forget to write it.\n\n## The instruction that makes a table useful\n\nA table with unnamed columns is a table the model invents, and it will invent columns that flatter the answer. Name them, and name what goes in the empty ones:\n\n> Columns: supplier, unit price (KES), minimum order, lead time in days, whether they deliver to Mombasa. One row per supplier. If a quote does not state something, write \"not stated\". Do not estimate. After the table, one line: which supplier is cheapest for an order of 200 units, and why.\n\nFour instructions: the columns, the row rule, the missing-data rule, and the one thing you want said in prose. That is a complete format spec and it takes fifteen seconds.\n\n## Where each container distorts\n\nEvery container has a bias, and knowing them lets you counteract.\n\n**Tables force false comparability.** If two of your four candidates do not have a value for a column, the table pressures the model to produce one. This is why \"not stated\" must be an explicit instruction rather than an assumption.\n\n**Lists force parallel importance.** Five bullets look equally weighted. If one of them matters ten times more than the rest, a list has hidden that, and you have to say so: \"put the most important first and say why it is first.\"\n\n**Prose hides omissions.** A fluent paragraph about three of your four questions reads exactly like a fluent paragraph about four. This is the strongest argument for structure on anything with a fixed set of questions.\n\n**Machine formats suppress reasoning**, which is next lesson's whole subject.\n\n## Say the destination, not only the shape\n\n\"This goes into a slide, so no sentence longer than eight words.\" \"This goes into a spreadsheet, so no commas inside fields.\" \"This is a reply to a customer who has already complained twice.\" \"This goes into a printed leaflet for people who may not have finished school.\"\n\nThe destination is doing several jobs at once: it sets the register, the length, the vocabulary, and the format details you would otherwise discover by getting them wrong. It is the highest-value sentence in most format specifications, and it takes ten words.\n\n## The reflex to build\n\nBefore sending, finish this sentence: *when this answer arrives, the very next thing I will do with it is ___.*\n\nPaste it into an email. Put it in a spreadsheet. Read it and decide. Send it to my manager. Feed it into a script.\n\nEach of those implies a different container, and once you have said it out loud the format instruction usually writes itself."},"check":{"question":"A procurement officer asks for a prose recommendation between three quotes and gets a confident, well-argued paragraph. She later finds it never addressed delivery terms, which she had asked about. What would have prevented this most reliably?","options":[{"text":"Asking for a longer answer, so there was room to cover every question.","correct":false,"tempting":"More prose gives more room for the same omission; length does not force coverage."},{"text":"Asking for a table with one column per question, so an unaddressed question would appear as an empty or \"not stated\" cell.","correct":true},{"text":"Asking the model to confirm at the end that it addressed all her questions.","correct":false,"tempting":"Better than nothing, but a self-report about coverage is as unreliable as the coverage itself."},{"text":"Splitting her request into three separate prompts, one per quote.","correct":false,"tempting":"Solves comparison by removing it, and each answer could still omit delivery terms."}]},"prev":{"slug":"length-and-what-it-buys","title":"Length, and what a word limit actually costs","url":"https://addaly.com/learn/prompting/length-and-what-it-buys"},"next":{"slug":"machine-readable-output","title":"Output a spreadsheet or a script can read","url":"https://addaly.com/learn/prompting/machine-readable-output"},"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 the container', in Getting Real Work out of a Model. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/prompting/choosing-the-container (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 the container","courseTitle":"Getting Real Work out of a Model","module":{"index":4,"slug":"the-shape-of-the-answer","title":"The shape of the answer"},"answer":null,"keyIdea":"Choose the container by what you will do with the answer next, and counteract the bias of whichever you chose: tables invent comparability, lists flatten importance, prose hides omissions.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"the-container-is-a-decision-about-what-you-will-do-next","runs":[{"text":"The container is a decision about what you will do next"}]},{"t":"p","runs":[{"text":"Ask for prose and you get an argument you have to read. Ask for a table and you get facts you can scan. Neither is better; they serve different next actions, and choosing badly is why a good answer still costs you twenty minutes."}]},{"t":"p","runs":[{"text":"Work backwards from what happens after you read it."}]},{"t":"h2","id":"the-five-containers-you-actually-need","runs":[{"text":"The five containers you actually need"}]},{"t":"p","runs":[{"text":"Prose","b":true},{"text":" when you need reasoning you will weigh. Advice, an explanation, an argument about a judgement call. Prose has room for qualification, and qualification is the point."}]},{"t":"p","runs":[{"text":"A table","b":true},{"text":" when you will compare like against like, and the comparison is the answer. Name the columns. Every cell must then be filled with something specific about that row, which is where a table earns its keep: vagueness has nowhere to sit."}]},{"t":"p","runs":[{"text":"A numbered list","b":true},{"text":" when order matters or you will act on the items one at a time. Steps, priorities, a checklist. Specify the count (“exactly five”) if you want to force selection rather than a dump."}]},{"t":"p","runs":[{"text":"A bulleted list","b":true},{"text":" when the items are peers and order does not matter. Underused correctly and overused generally: most of the bulleted lists people receive should have been two sentences of prose, and the model produces them because bullets pattern-match to helpfulness."}]},{"t":"p","runs":[{"text":"A machine format","b":true},{"text":" (CSV, JSON, a fixed line format) when the output goes into a spreadsheet, a script, or another system. Next lesson."}]},{"t":"p","runs":[{"text":"There is a sixth worth naming: "},{"text":"the thing itself","b":true},{"text":". The email, ready to send. The message, ready to paste. No commentary before or after. “Give me only the message” is a format instruction and people forget to write it."}]},{"t":"h2","id":"the-instruction-that-makes-a-table-useful","runs":[{"text":"The instruction that makes a table useful"}]},{"t":"p","runs":[{"text":"A table with unnamed columns is a table the model invents, and it will invent columns that flatter the answer. Name them, and name what goes in the empty ones:"}]},{"t":"quote","runs":[{"text":"Columns: supplier, unit price (KES), minimum order, lead time in days, whether they deliver to Mombasa. One row per supplier. If a quote does not state something, write “not stated”. Do not estimate. After the table, one line: which supplier is cheapest for an order of 200 units, and why."}]},{"t":"p","runs":[{"text":"Four instructions: the columns, the row rule, the missing-data rule, and the one thing you want said in prose. That is a complete format spec and it takes fifteen seconds."}]},{"t":"h2","id":"where-each-container-distorts","runs":[{"text":"Where each container distorts"}]},{"t":"p","runs":[{"text":"Every container has a bias, and knowing them lets you counteract."}]},{"t":"p","runs":[{"text":"Tables force false comparability.","b":true},{"text":" If two of your four candidates do not have a value for a column, the table pressures the model to produce one. This is why “not stated” must be an explicit instruction rather than an assumption."}]},{"t":"p","runs":[{"text":"Lists force parallel importance.","b":true},{"text":" Five bullets look equally weighted. If one of them matters ten times more than the rest, a list has hidden that, and you have to say so: “put the most important first and say why it is first.”"}]},{"t":"p","runs":[{"text":"Prose hides omissions.","b":true},{"text":" A fluent paragraph about three of your four questions reads exactly like a fluent paragraph about four. This is the strongest argument for structure on anything with a fixed set of questions."}]},{"t":"p","runs":[{"text":"Machine formats suppress reasoning","b":true},{"text":", which is next lesson’s whole subject."}]},{"t":"h2","id":"say-the-destination-not-only-the-shape","runs":[{"text":"Say the destination, not only the shape"}]},{"t":"p","runs":[{"text":"“This goes into a slide, so no sentence longer than eight words.” “This goes into a spreadsheet, so no commas inside fields.” “This is a reply to a customer who has already complained twice.” “This goes into a printed leaflet for people who may not have finished school.”"}]},{"t":"p","runs":[{"text":"The destination is doing several jobs at once: it sets the register, the length, the vocabulary, and the format details you would otherwise discover by getting them wrong. It is the highest-value sentence in most format specifications, and it takes ten words."}]},{"t":"h2","id":"the-reflex-to-build","runs":[{"text":"The reflex to build"}]},{"t":"p","runs":[{"text":"Before sending, finish this sentence: "},{"text":"when this answer arrives, the very next thing I will do with it is ___.","i":true}]},{"t":"p","runs":[{"text":"Paste it into an email. Put it in a spreadsheet. Read it and decide. Send it to my manager. Feed it into a script."}]},{"t":"p","runs":[{"text":"Each of those implies a different container, and once you have said it out loud the format instruction usually writes itself."}]}],"check":{"question":[{"text":"A procurement officer asks for a prose recommendation between three quotes and gets a confident, well-argued paragraph. She later finds it never addressed delivery terms, which she had asked about. What would have prevented this most reliably?"}],"options":[{"text":[{"text":"Asking for a longer answer, so there was room to cover every question."}],"correct":false,"tempting":[{"text":"More prose gives more room for the same omission; length does not force coverage."}]},{"text":[{"text":"Asking for a table with one column per question, so an unaddressed question would appear as an empty or “not stated” cell."}],"correct":true,"tempting":null},{"text":[{"text":"Asking the model to confirm at the end that it addressed all her questions."}],"correct":false,"tempting":[{"text":"Better than nothing, but a self-report about coverage is as unreliable as the coverage itself."}]},{"text":[{"text":"Splitting her request into three separate prompts, one per quote."}],"correct":false,"tempting":[{"text":"Solves comparison by removing it, and each answer could still omit delivery terms."}]}],"order":[3,2,0,1],"why":null},"prev":{"slug":"length-and-what-it-buys","title":"Length, and what a word limit actually costs","minutes":8},"next":{"slug":"machine-readable-output","title":"Output a spreadsheet or a script can read","minutes":9}}}