{"course":{"slug":"prompting","title":"Getting Real Work out of a Model","url":"https://addaly.com/learn/prompting","level":"beginner","track":"start"},"lesson":{"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","position":29,"of":72,"minutes":8,"answer":"Strip padding permanently and set length by what the content needs, because a tight limit on a hard question removes the working rather than the waste.","keyIdea":"Strip padding permanently and set length by what the content needs, because a tight limit on a hard question removes the working rather than the waste."},"body":{"format":"markdown","text":"## Two different reasons an answer is too long\n\nThe first is padding: preamble, restatement of your question, a closing summary of what was just said, three sentences of hedging. This is pure waste and you can switch most of it off with one line.\n\n> No preamble. Do not restate the question. No closing summary.\n\nThat single line removes perhaps a fifth of the length of a typical response, and it costs nothing, because none of what it removes was information.\n\n> Figure: Where the words go in a typical 300-word answer. Bars (words): Preamble 15; Restating your question 15; The content 240; Closing summary 30. Sixty words of padding, a fifth of the answer, removed by one standing line and never missed. The 240 that remain are a different question, and cutting them is a trade rather than a saving.\n\nThe second reason is content. The answer is long because the subject has six parts. Cutting here is a trade, and pretending otherwise is how people end up with confident two-sentence summaries of things that needed six.\n\nKnowing which of the two you are dealing with is the whole lesson.\n\n## How to specify length\n\n**Say the number, and say the unit that matters.** \"Under 150 words\" is enforceable and roughly honoured. \"Three bullet points\" is more precisely honoured than a word count, because counting to three is easy and counting to 150 is not. \"Two paragraphs\" works. \"Concise\" does not; it is an adjective and it means whatever the model's training suggests.\n\n**Expect word counts to be approximate.** A model does not count words as it writes; it produces text that is about the right length, because length correlates with things it can sense. Ask for exactly 50 words and you will usually get 44 to 58. If the number matters (a character limit on an SMS, a field in a form) check it yourself, or ask for the count in the output and then check that.\n\n**Give a range rather than a maximum when quality matters.** \"Between 120 and 180 words\" produces better writing than \"under 180\", because a bare maximum is a pressure in one direction and things get dropped to satisfy it.\n\n## The cost of a tight limit on a hard question\n\nThis is the part that catches careful people.\n\nFor anything requiring several dependent steps, the text is where the work happens. A model asked to compare four insurance policies in exactly forty words has almost no room in which to do the comparing, so what you get is the shape of a conclusion with very little behind it. The number will look confident. It has to be, since there is no space for a caveat.\n\nIf accuracy matters more than brevity, split it:\n\n> First, work through the comparison in as much detail as you need. Then, at the end, give me the two-sentence version under the heading SHORT.\n\nYou read the short version. The long version exists, so the short one is a summary of some reasoning rather than a guess wearing a summary's clothes. And if the short version surprises you, the working is right there.\n\nThis is the single most useful move in this module, and it generalises: **reasoning first, tidy output last**.\n\n## When brevity is the actual requirement\n\nSometimes short is not a preference, it is the spec. A push notification. A subject line. A message that has to be read on a phone in a queue.\n\nIn that case, say the destination, not just the number:\n\n> This is a WhatsApp message that will be read on a phone. Under 300 characters. One idea. No greeting, no sign-off: they know who I am.\n\nThe destination implies constraints you would otherwise have to write out one at a time, and it stops the answer arriving in a form that technically meets your word count and is still unusable.\n\n## The direction people get wrong\n\nMost people over-ask for short. They are reacting to the padding problem and applying the fix to the content problem.\n\nThe better default: ban the padding permanently, put it in your standing brief, and then set length by what the answer actually needs. You will find that with the preamble and the closing summary gone, most answers are already about the right length and you have one less thing to specify.\n\n## A worked line you can steal\n\nFor everyday work:\n\n> No preamble, no closing summary, no restatement of my question. Answer in as much space as the content needs, and if that is more than 200 words, give me a two-line version at the top under BOTTOM LINE.\n\nYou get the short version for reading and the long version for checking, with the padding gone from both."},"check":{"question":"An analyst asks for a 40-word verdict comparing four suppliers on price, lead time and reliability. The verdict reads well but turns out to be wrong about lead times. What is the most likely mechanism?","options":[{"text":"The model does not have reliable data about supplier lead times and filled the gap.","correct":false,"tempting":"Possible in general, but here the lead times were supplied in the prompt; the constraint is on the output, not the input."},{"text":"Forty words is enough for a verdict, so the error must come from the source data being misread.","correct":false,"tempting":"Assumes the limit is harmless; the limit is the thing that removed the intermediate comparison where a misreading would have shown."},{"text":"Verdicts are inherently unreliable and should never be requested.","correct":false,"tempting":"Overcorrects into a rule that would remove a useful output rather than fixing how it is produced."},{"text":"The word limit left no room for the comparison itself, so the verdict was produced without the intermediate work that would have made it correct.","correct":true}]},"prev":{"slug":"ask-for-a-usable-shape","title":"Ask for the shape you can actually use","url":"https://addaly.com/learn/prompting/ask-for-a-usable-shape"},"next":{"slug":"choosing-the-container","title":"Choosing the container","url":"https://addaly.com/learn/prompting/choosing-the-container"},"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) 'Length, and what a word limit actually costs', in Getting Real Work out of a Model. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/prompting/length-and-what-it-buys (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":"Length, and what a word limit actually costs","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":"Strip padding permanently and set length by what the content needs, because a tight limit on a hard question removes the working rather than the waste.","hero":{"kind":"figure","figure":{"t":"figure","label":"Figure","figure":{"type":"bars","unit":"words","title":"Where the words go in a typical 300-word answer","series":[{"label":"Preamble","value":15},{"label":"Restating your question","value":15},{"label":"The content","value":240,"highlight":true},{"label":"Closing summary","value":30}],"caption":"Sixty words of padding, a fifth of the answer, removed by one standing line and never missed. The 240 that remain are a different question, and cutting them is a trade rather than a saving."},"alt":"Figure: Where the words go in a typical 300-word answer. Bars (words): Preamble 15; Restating your question 15; The content 240; Closing summary 30. Sixty words of padding, a fifth of the answer, removed by one standing line and never missed. The 240 that remain are a different question, and cutting them is a trade rather than a saving."}},"blocks":[{"t":"h2","id":"two-different-reasons-an-answer-is-too-long","runs":[{"text":"Two different reasons an answer is too long"}]},{"t":"p","runs":[{"text":"The first is padding: preamble, restatement of your question, a closing summary of what was just said, three sentences of hedging. This is pure waste and you can switch most of it off with one line."}]},{"t":"quote","runs":[{"text":"No preamble. Do not restate the question. No closing summary."}]},{"t":"p","runs":[{"text":"That single line removes perhaps a fifth of the length of a typical response, and it costs nothing, because none of what it removes was information."}]},{"t":"p","runs":[{"text":"The second reason is content. The answer is long because the subject has six parts. Cutting here is a trade, and pretending otherwise is how people end up with confident two-sentence summaries of things that needed six."}]},{"t":"p","runs":[{"text":"Knowing which of the two you are dealing with is the whole lesson."}]},{"t":"h2","id":"how-to-specify-length","runs":[{"text":"How to specify length"}]},{"t":"p","runs":[{"text":"Say the number, and say the unit that matters.","b":true},{"text":" “Under 150 words” is enforceable and roughly honoured. “Three bullet points” is more precisely honoured than a word count, because counting to three is easy and counting to 150 is not. “Two paragraphs” works. “Concise” does not; it is an adjective and it means whatever the model’s training suggests."}]},{"t":"p","runs":[{"text":"Expect word counts to be approximate.","b":true},{"text":" A model does not count words as it writes; it produces text that is about the right length, because length correlates with things it can sense. Ask for exactly 50 words and you will usually get 44 to 58. If the number matters (a character limit on an SMS, a field in a form) check it yourself, or ask for the count in the output and then check that."}]},{"t":"p","runs":[{"text":"Give a range rather than a maximum when quality matters.","b":true},{"text":" “Between 120 and 180 words” produces better writing than “under 180”, because a bare maximum is a pressure in one direction and things get dropped to satisfy it."}]},{"t":"h2","id":"the-cost-of-a-tight-limit-on-a-hard-question","runs":[{"text":"The cost of a tight limit on a hard question"}]},{"t":"p","runs":[{"text":"This is the part that catches careful people."}]},{"t":"p","runs":[{"text":"For anything requiring several dependent steps, the text is where the work happens. A model asked to compare four insurance policies in exactly forty words has almost no room in which to do the comparing, so what you get is the shape of a conclusion with very little behind it. The number will look confident. It has to be, since there is no space for a caveat."}]},{"t":"p","runs":[{"text":"If accuracy matters more than brevity, split it:"}]},{"t":"quote","runs":[{"text":"First, work through the comparison in as much detail as you need. Then, at the end, give me the two-sentence version under the heading SHORT."}]},{"t":"p","runs":[{"text":"You read the short version. The long version exists, so the short one is a summary of some reasoning rather than a guess wearing a summary’s clothes. And if the short version surprises you, the working is right there."}]},{"t":"p","runs":[{"text":"This is the single most useful move in this module, and it generalises: "},{"text":"reasoning first, tidy output last","b":true},{"text":"."}]},{"t":"h2","id":"when-brevity-is-the-actual-requirement","runs":[{"text":"When brevity is the actual requirement"}]},{"t":"p","runs":[{"text":"Sometimes short is not a preference, it is the spec. A push notification. A subject line. A message that has to be read on a phone in a queue."}]},{"t":"p","runs":[{"text":"In that case, say the destination, not just the number:"}]},{"t":"quote","runs":[{"text":"This is a WhatsApp message that will be read on a phone. Under 300 characters. One idea. No greeting, no sign-off: they know who I am."}]},{"t":"p","runs":[{"text":"The destination implies constraints you would otherwise have to write out one at a time, and it stops the answer arriving in a form that technically meets your word count and is still unusable."}]},{"t":"h2","id":"the-direction-people-get-wrong","runs":[{"text":"The direction people get wrong"}]},{"t":"p","runs":[{"text":"Most people over-ask for short. They are reacting to the padding problem and applying the fix to the content problem."}]},{"t":"p","runs":[{"text":"The better default: ban the padding permanently, put it in your standing brief, and then set length by what the answer actually needs. You will find that with the preamble and the closing summary gone, most answers are already about the right length and you have one less thing to specify."}]},{"t":"h2","id":"a-worked-line-you-can-steal","runs":[{"text":"A worked line you can steal"}]},{"t":"p","runs":[{"text":"For everyday work:"}]},{"t":"quote","runs":[{"text":"No preamble, no closing summary, no restatement of my question. Answer in as much space as the content needs, and if that is more than 200 words, give me a two-line version at the top under BOTTOM LINE."}]},{"t":"p","runs":[{"text":"You get the short version for reading and the long version for checking, with the padding gone from both."}]}],"check":{"question":[{"text":"An analyst asks for a 40-word verdict comparing four suppliers on price, lead time and reliability. The verdict reads well but turns out to be wrong about lead times. What is the most likely mechanism?"}],"options":[{"text":[{"text":"The model does not have reliable data about supplier lead times and filled the gap."}],"correct":false,"tempting":[{"text":"Possible in general, but here the lead times were supplied in the prompt; the constraint is on the output, not the input."}]},{"text":[{"text":"Forty words is enough for a verdict, so the error must come from the source data being misread."}],"correct":false,"tempting":[{"text":"Assumes the limit is harmless; the limit is the thing that removed the intermediate comparison where a misreading would have shown."}]},{"text":[{"text":"Verdicts are inherently unreliable and should never be requested."}],"correct":false,"tempting":[{"text":"Overcorrects into a rule that would remove a useful output rather than fixing how it is produced."}]},{"text":[{"text":"The word limit left no room for the comparison itself, so the verdict was produced without the intermediate work that would have made it correct."}],"correct":true,"tempting":null}],"order":[0,3,2,1],"why":null},"prev":{"slug":"ask-for-a-usable-shape","title":"Ask for the shape you can actually use","minutes":7},"next":{"slug":"choosing-the-container","title":"Choosing the container","minutes":8}}}