{"course":{"slug":"prompting","title":"Getting Real Work out of a Model","url":"https://addaly.com/learn/prompting","level":"beginner","track":"start"},"lesson":{"slug":"role-vs-task","title":"A role is not a task","url":"https://addaly.com/learn/prompting/role-vs-task","position":5,"of":72,"minutes":6,"answer":"Naming a persona sets tone; only naming the verb, the object, the constraints and the finished state sets the work.","keyIdea":"Naming a persona sets tone; only naming the verb, the object, the constraints and the finished state sets the work."},"body":{"format":"markdown","text":"## The most repeated advice in prompting\n\n> You are a world-class marketing expert with 20 years of experience.\n\nMillions of prompts start like this. It does less than people believe, and the belief underneath it costs real quality, so it is worth being precise about what a role does and does not do.\n\n## What a role changes\n\nA role sets register. Write \"you are a paediatric nurse explaining this to a worried parent\" and the vocabulary, the reading level, the pacing and the amount of reassurance all shift. That is a real effect, and sometimes it is exactly what you wanted.\n\nWhat a role cannot do is add competence. There is no expert mode that switches on. The knowledge is the same either way; you have only indicated which kind of text to produce. \"20 years of experience\" adds no experience. Neither does \"world-class\", \"senior\", or \"PhD-level\". Turning it up to 30 years does nothing at all.\n\nBe aware that most role-prompt advice online was written in 2023 against much weaker models, where it helped more. It has aged. It is repeated anyway.\n\n## The useful version is audience, not costume\n\nThe part of role prompting that still earns its place is naming the reader:\n\n> Explain this to me as if I am the plumber who has to install it, not the architect who designed it.\n\n> Write this for a school governor who is not a lawyer and has ten minutes.\n\nThat is not a costume for the model. It is information about who the output is for, which is a constraint, which is part of the task.\n\n## What a task looks like\n\nA task names five things:\n\n- **the verb**: rewrite, compare, find, translate, check, shorten\n- **the object**: this contract, these three quotes, the second paragraph\n- **the constraints**: under 120 words, in Spanish, without mentioning price\n- **the finished state**: so that a reader can tell in one read whether they qualify\n- **what to do when stuck**: if the document does not say, write \"not stated\"\n\nNotice that \"review\" fails this test. Review has no finished state. Neither does \"improve\", \"look at\", or \"give me feedback on\". They are the verbs we reach for when we have not decided what we want.\n\n## Before and after\n\nBefore:\n\n> You are an expert lawyer. Review my contract.\n\nAfter:\n\n> I am a freelance illustrator in Manila. The contract is pasted below. Find every clause that lets the client keep using the artwork if they never pay the final instalment, and every clause that assigns copyright rather than licensing it. Quote each clause exactly, then say in one plain sentence what it means for me. I am not asking for legal advice. I will take this to a lawyer. I want to know what to ask her.\n\nThe second prompt has no role in it and will produce far better work, because it says what \"review\" was standing in for.\n\nThat last line matters as well. Naming what the output is for changes what a good output is. A list of questions for a lawyer looks nothing like a legal opinion, and only one of those is useful to you.\n\n## When you still might open with a role\n\nWhen the register is the whole point and hard to describe otherwise. \"Write this as a hospital discharge nurse would say it out loud\" carries tone, pace and word choice in eight words. That is a fair trade.\n\nWhat you should not do is write the costume and then stop, believing the hard part is handled. The costume is the cheap part. The task is the work."},"check":{"question":"Two prompts aim at the same badly worded refund policy. Prompt A: \"You are a senior policy writer with 20 years of experience. Rewrite this policy.\" Prompt B: \"Rewrite this policy so that a customer who bought a phone charger and wants to return it after 40 days can tell in one read whether they can. Under 120 words. Say 'you', not 'the customer'.\" Why does B usually produce the more useful policy?","options":[{"text":"A is weaker because 20 years is arbitrary. Saying \"30 years\" or \"world-renowned\" would have strengthened it.","correct":false,"tempting":"Seniority reads like a dial you can turn up, so people tune the number instead of noticing that the number was never doing anything."},{"text":"A fails because the model has no idea what a policy writer is.","correct":false,"tempting":"If the persona does not help, assuming the concept is missing is a natural guess, but the model knows the concept perfectly well; the concept simply does not specify the job."},{"text":"Role prompts are harmful, so A was always going to produce something worse than no instruction at all.","correct":false,"tempting":"Overcorrecting after learning that personas are oversold: a role does set register and occasionally helps, so \"does little by itself\" is the honest claim, not \"damages the output\"."},{"text":"B names the reader, the situation and the finished state, so there is a standard the output can be measured against. A only names a costume.","correct":true}]},"prev":{"slug":"instruction-following-is-trained","title":"Why it obeys you at all","url":"https://addaly.com/learn/prompting/instruction-following-is-trained"},"next":{"slug":"anatomy-of-a-prompt","title":"The six things a working prompt contains","url":"https://addaly.com/learn/prompting/anatomy-of-a-prompt"},"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) 'A role is not a task', in Getting Real Work out of a Model. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/prompting/role-vs-task (Accessed: 31 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"A role is not a task","courseTitle":"Getting Real Work out of a Model","module":{"index":1,"slug":"what-a-prompt-is","title":"What a prompt actually is"},"answer":null,"keyIdea":"Naming a persona sets tone; only naming the verb, the object, the constraints and the finished state sets the work.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"the-most-repeated-advice-in-prompting","runs":[{"text":"The most repeated advice in prompting"}]},{"t":"quote","runs":[{"text":"You are a world-class marketing expert with 20 years of experience."}]},{"t":"p","runs":[{"text":"Millions of prompts start like this. It does less than people believe, and the belief underneath it costs real quality, so it is worth being precise about what a role does and does not do."}]},{"t":"h2","id":"what-a-role-changes","runs":[{"text":"What a role changes"}]},{"t":"p","runs":[{"text":"A role sets register. Write “you are a paediatric nurse explaining this to a worried parent” and the vocabulary, the reading level, the pacing and the amount of reassurance all shift. That is a real effect, and sometimes it is exactly what you wanted."}]},{"t":"p","runs":[{"text":"What a role cannot do is add competence. There is no expert mode that switches on. The knowledge is the same either way; you have only indicated which kind of text to produce. “20 years of experience” adds no experience. Neither does “world-class”, “senior”, or “PhD-level”. Turning it up to 30 years does nothing at all."}]},{"t":"p","runs":[{"text":"Be aware that most role-prompt advice online was written in 2023 against much weaker models, where it helped more. It has aged. It is repeated anyway."}]},{"t":"h2","id":"the-useful-version-is-audience-not-costume","runs":[{"text":"The useful version is audience, not costume"}]},{"t":"p","runs":[{"text":"The part of role prompting that still earns its place is naming the reader:"}]},{"t":"quote","runs":[{"text":"Explain this to me as if I am the plumber who has to install it, not the architect who designed it."}]},{"t":"quote","runs":[{"text":"Write this for a school governor who is not a lawyer and has ten minutes."}]},{"t":"p","runs":[{"text":"That is not a costume for the model. It is information about who the output is for, which is a constraint, which is part of the task."}]},{"t":"h2","id":"what-a-task-looks-like","runs":[{"text":"What a task looks like"}]},{"t":"p","runs":[{"text":"A task names five things:"}]},{"t":"ul","items":[[{"text":"the verb","b":true},{"text":": rewrite, compare, find, translate, check, shorten"}],[{"text":"the object","b":true},{"text":": this contract, these three quotes, the second paragraph"}],[{"text":"the constraints","b":true},{"text":": under 120 words, in Spanish, without mentioning price"}],[{"text":"the finished state","b":true},{"text":": so that a reader can tell in one read whether they qualify"}],[{"text":"what to do when stuck","b":true},{"text":": if the document does not say, write “not stated”"}]]},{"t":"p","runs":[{"text":"Notice that “review” fails this test. Review has no finished state. Neither does “improve”, “look at”, or “give me feedback on”. They are the verbs we reach for when we have not decided what we want."}]},{"t":"h2","id":"before-and-after","runs":[{"text":"Before and after"}]},{"t":"p","runs":[{"text":"Before:"}]},{"t":"quote","runs":[{"text":"You are an expert lawyer. Review my contract."}]},{"t":"p","runs":[{"text":"After:"}]},{"t":"quote","runs":[{"text":"I am a freelance illustrator in Manila. The contract is pasted below. Find every clause that lets the client keep using the artwork if they never pay the final instalment, and every clause that assigns copyright rather than licensing it. Quote each clause exactly, then say in one plain sentence what it means for me. I am not asking for legal advice. I will take this to a lawyer. I want to know what to ask her."}]},{"t":"p","runs":[{"text":"The second prompt has no role in it and will produce far better work, because it says what “review” was standing in for."}]},{"t":"p","runs":[{"text":"That last line matters as well. Naming what the output is for changes what a good output is. A list of questions for a lawyer looks nothing like a legal opinion, and only one of those is useful to you."}]},{"t":"h2","id":"when-you-still-might-open-with-a-role","runs":[{"text":"When you still might open with a role"}]},{"t":"p","runs":[{"text":"When the register is the whole point and hard to describe otherwise. “Write this as a hospital discharge nurse would say it out loud” carries tone, pace and word choice in eight words. That is a fair trade."}]},{"t":"p","runs":[{"text":"What you should not do is write the costume and then stop, believing the hard part is handled. The costume is the cheap part. The task is the work."}]}],"check":{"question":[{"text":"Two prompts aim at the same badly worded refund policy. Prompt A: “You are a senior policy writer with 20 years of experience. Rewrite this policy.” Prompt B: “Rewrite this policy so that a customer who bought a phone charger and wants to return it after 40 days can tell in one read whether they can. Under 120 words. Say ‘you’, not ‘the customer’.” Why does B usually produce the more useful policy?"}],"options":[{"text":[{"text":"A is weaker because 20 years is arbitrary. Saying “30 years” or “world-renowned” would have strengthened it."}],"correct":false,"tempting":[{"text":"Seniority reads like a dial you can turn up, so people tune the number instead of noticing that the number was never doing anything."}]},{"text":[{"text":"A fails because the model has no idea what a policy writer is."}],"correct":false,"tempting":[{"text":"If the persona does not help, assuming the concept is missing is a natural guess, but the model knows the concept perfectly well; the concept simply does not specify the job."}]},{"text":[{"text":"Role prompts are harmful, so A was always going to produce something worse than no instruction at all."}],"correct":false,"tempting":[{"text":"Overcorrecting after learning that personas are oversold: a role does set register and occasionally helps, so “does little by itself” is the honest claim, not “damages the output”."}]},{"text":[{"text":"B names the reader, the situation and the finished state, so there is a standard the output can be measured against. A only names a costume."}],"correct":true,"tempting":null}],"order":[2,0,3,1],"why":null},"prev":{"slug":"instruction-following-is-trained","title":"Why it obeys you at all","minutes":9},"next":{"slug":"anatomy-of-a-prompt","title":"The six things a working prompt contains","minutes":8}}}