{"author":{"@type":"Person","name":"Alexander Bock","url":"https://abxk.ai/en/autor/alexander-bock/"},"categories":["AI Productivity"],"content":{"markdown":"Most weak AI answers have a simple cause: the prompt left out something the assistant needed. A **prompt** is the request you type into an AI tool. Strong prompts follow a pattern that you can learn in an afternoon. It has four building blocks: role, task, context and format.\n\nThis is part 1 of Level 1 in our AI Productivity learning path. In Level 0, you wrote [your first prompt](/en/ai-productivity/learn/your-first-prompt/). Now you will learn what each block does, when you need it and how to combine the blocks. The exercises at the end help you practice with tasks from your own work.\n\n## Four questions every good prompt answers\n\nThink of a prompt as a short briefing for a new colleague. A good briefing answers four questions. Who should the colleague be for this job? What exactly should they do? What do they need to know? And what should the finished work look like?\n\n![Four cards: role, task, context and format, each with the question it answers and a short example](/images/learn/en/prompting-basics-role-task-context-format-blocks.webp \"The four building blocks of a prompt and the question each one answers.\")\n\nEach block solves a different problem. The table shows what typically goes wrong when one of them is missing:\n\n| Block | Question it answers | Without it, you often get |\n|---|---|---|\n| Role | Who should the AI be? | A neutral, one-size-fits-all voice |\n| Task | What should it do? | An answer to a question you did not ask |\n| Context | What does it need to know? | Generic text that fits nobody |\n| Format | What should the result look like? | The wrong length, layout or tone |\n\nYou do not need all four blocks every time. The task is the only block you always need. The other three are tools you add when the result matters.\n\n## Task: one clear verb and a goal\n\nThe task tells the assistant what to do. Start with a clear verb: *write, rewrite, summarize, list, compare, explain, translate* or *check*. Then name the object and, if you can, the goal.\n\n- Vague: \"Something about our new vacation rule.\"\n- Clear: \"Write an email that explains our new vacation booking rule to the team.\"\n\nGive one task per prompt while you are learning. If a job is big, split it. Ask for an outline first, check it, and then ask for the full text. Small steps are easier to check, and a mistake in step one does not spread through the whole result.\n\n## Context: what the assistant cannot know\n\nThe assistant knows a lot about the world, but nothing about your situation. **Context** is the information it needs for this specific job: who will read the result, why you need it, which facts it must use and what limits apply.\n\n- **Audience:** \"for new team members without a technical background\"\n- **Purpose:** \"they should book their vacation in the new HR tool from next month on\"\n- **Facts:** \"requests now need two weeks' notice; the paper form is no longer used\"\n- **Limits:** \"do not mention overtime rules, they are still under discussion\"\n\nWhen you paste a longer text, separate it from your instructions. Use a **delimiter**: a simple marker such as triple quotes (\"\"\") or a line with ### that shows where the pasted text starts and ends. This helps the assistant tell your instructions apart from the material it should work on.\n\nContext is also where private data slips in. Replace names, customer details and internal figures with placeholders such as \"Customer A\" before you paste anything into a tool your company has not approved.\n\n{{\u003c callout type=\"tip\" title=\"Let the assistant ask\" \u003e}}\nNot sure which context to add? In Level 0, you used this trick as a follow-up. It works just as well at the end of the prompt itself: *\"Before you start, ask me up to three questions about anything that is unclear.\"* The assistant will name the information it is missing, and you can answer in one short message.\n{{\u003c /callout \u003e}}\n\n## Format: describe the finished result\n\nThe format tells the assistant what the result should look like. If you skip it, the assistant guesses, and its guess is often too long. Be specific about:\n\n- **Length:** \"under 150 words\", \"three sentences\", \"one page\"\n- **Structure:** \"bullet points\", \"a table with three columns\", \"a subject line and a short body\"\n- **Tone:** \"friendly but not casual\", \"neutral and factual\"\n- **Language:** \"US English\", \"plain words, no jargon\"\n- **What to leave out:** \"no introduction\", \"no emojis\"\n\nThe strongest format instruction is often an example. Paste a short sample of the style you want and add: \"Follow the structure and tone of this example, but not its content.\" One good example can replace many words of description. Level 2 shows how to work with several examples and other [prompt patterns that work](/en/ai-productivity/learn/prompt-patterns-that-work/).\n\n## Role: set the point of view\n\nA **role** tells the assistant from which point of view to work: \"You are an experienced editor\", \"Act as a patient teacher for beginners\" or \"Act as a skeptical customer who reads this offer.\" The role changes the choice of words, the tone and what the assistant pays attention to.\n\nA role does not add knowledge, though. Calling the assistant a lawyer does not make its legal statements correct. Use roles to set a perspective, not to borrow authority. They help most when you want feedback: \"Act as a busy manager and tell me what is unclear in this report.\"\n\n## Putting the blocks together\n\nEach block narrows down what the assistant can give you. The task alone produces a usable but generic text. Every block you add makes the answer fit your situation better.\n\n![Four rising steps: task only, plus context, plus format, plus role; each step makes the answer fit better](/images/learn/en/prompting-basics-role-task-context-format-ladder.webp \"From vague to precise: each block you add removes one kind of guesswork.\")\n\nHere is a complete prompt with all four blocks and an example. The order is a habit, not a rule. What matters is that nothing important is missing.\n\n![A complete prompt split into labeled parts: role, task, context, format and example](/images/learn/en/prompting-basics-role-task-context-format-anatomy.webp \"Anatomy of a complete prompt for an everyday email.\")\n\nDo all prompts have to be this long? No. A short prompt is fine for quick, general questions, such as \"Explain what a pivot table is in two sentences.\" Longer prompts pay off when the result goes to other people, when details matter or when the task comes back every week. Save your best prompts for those repeating tasks. Level 3 shows how to turn them into [reusable prompt templates](/en/ai-productivity/learn/reusable-prompt-templates-and-custom-assistants/).\n\n![Two columns: when a short prompt is enough and when a long prompt pays off](/images/learn/en/prompting-basics-role-task-context-format-short-long.webp \"Short or long? The more the result matters, the more blocks you add.\")\n\nThe four blocks are not tied to one product. You can use them in ChatGPT, Gemini or any other chat assistant, whichever model runs behind it. Our [LLM database](/en/llm/) lists {{\u003c stat \"models\" \u003e}} language models if you want to see what is inside. Specialized tools often turn some blocks into buttons: in a writing tool, you may pick the tone from a menu instead of typing it.\n\n{{\u003c tools \"chatgpt google-gemini quillbot\" \u003e}}\n\n## Exercises: from vague to complete\n\nPractice with three typical requests. For each one, decide which blocks are missing. Then compare your version with ours.\n\n| Vague prompt | Missing | A complete version |\n|---|---|---|\n| \"Write an agenda.\" | Context, format | \"Write an agenda for a 45-minute kickoff meeting for our website relaunch. Attendees: marketing, IT and an outside agency. Goals: agree on the timeline and name one contact person per team. A numbered list with times.\" |\n| \"Plan our team day.\" | Context, format | \"Suggest a schedule for a half-day team event for 12 people. Two team members use wheelchairs, so every activity must be accessible. A table with three columns: time, activity, what to book.\" |\n| \"Check my slides.\" | Role, context, format | \"Act as a customer who sees our product for the first time. Read the slide text below and list the three points that are hardest to understand, each with a simpler wording. Slides: ### [text] ###\" |\n\nThen try it with your own work:\n\n1. Pick a real task from this week.\n2. Write it as one sentence with a clear verb.\n3. Add two or three lines of context.\n4. Add the format: length, structure and tone.\n5. Add a role if you want feedback or a special point of view.\n6. Run the one-sentence version and the full version, and compare the two answers.\n\n{{\u003c callout type=\"try\" title=\"A template to copy\" \u003e}}\n*\"You are [role]. [Task with a clear verb]. Context: [audience, purpose, facts, limits]. Format: [length, structure, tone]. Before you start, ask me up to three questions if anything is unclear.\"*\n{{\u003c /callout \u003e}}\n\nWhen you compare the answers, ask yourself two questions. Which version needed fewer edits? And which block made the biggest difference for this kind of task? The answers tell you where to put your effort the next time.","text":"Most weak AI answers have a simple cause: the prompt left out something the assistant needed. A prompt is the request you type into an AI tool. Strong prompts follow a pattern that you can learn in an afternoon. It has four building blocks: role, task, context and format.\nThis is part 1 of Level 1 in our AI Productivity learning path. In Level 0, you wrote your first prompt. Now you will learn what each block does, when you need it and how to combine the blocks. The exercises at the end help you practice with tasks from your own work.\nFour questions every good prompt answers Think of a prompt as a short briefing for a new colleague. A good briefing answers four questions. Who should the colleague be for this job? What exactly should they do? What do they need to know? And what should the finished work look like?\nThe four building blocks of a prompt and the question each one answers. Each block solves a different problem. The table shows what typically goes wrong when one of them is missing:\nPrompting Basics: Role, Task, Context and Format Block Question it answers Without it, you often get Role Who should the AI be? A neutral, one-size-fits-all voice Task What should it do? An answer to a question you did not ask Context What does it need to know? Generic text that fits nobody Format What should the result look like? The wrong length, layout or tone You do not need all four blocks every time. The task is the only block you always need. The other three are tools you add when the result matters.\nTask: one clear verb and a goal The task tells the assistant what to do. Start with a clear verb: write, rewrite, summarize, list, compare, explain, translate or check. Then name the object and, if you can, the goal.\nVague: \u0026ldquo;Something about our new vacation rule.\u0026rdquo; Clear: \u0026ldquo;Write an email that explains our new vacation booking rule to the team.\u0026rdquo; Give one task per prompt while you are learning. If a job is big, split it. Ask for an outline first, check it, and then ask for the full text. Small steps are easier to check, and a mistake in step one does not spread through the whole result.\nContext: what the assistant cannot know The assistant knows a lot about the world, but nothing about your situation. Context is the information it needs for this specific job: who will read the result, why you need it, which facts it must use and what limits apply.\nAudience: \u0026ldquo;for new team members without a technical background\u0026rdquo; Purpose: \u0026ldquo;they should book their vacation in the new HR tool from next month on\u0026rdquo; Facts: \u0026ldquo;requests now need two weeks\u0026rsquo; notice; the paper form is no longer used\u0026rdquo; Limits: \u0026ldquo;do not mention overtime rules, they are still under discussion\u0026rdquo; When you paste a longer text, separate it from your instructions. Use a delimiter: a simple marker such as triple quotes (\u0026quot;\u0026quot;\u0026quot;) or a line with ### that shows where the pasted text starts and ends. This helps the assistant tell your instructions apart from the material it should work on.\nContext is also where private data slips in. Replace names, customer details and internal figures with placeholders such as \u0026ldquo;Customer A\u0026rdquo; before you paste anything into a tool your company has not approved.\n✦Let the assistant ask\nNot sure which context to add? In Level 0, you used this trick as a follow-up. It works just as well at the end of the prompt itself: \u0026ldquo;Before you start, ask me up to three questions about anything that is unclear.\u0026rdquo; The assistant will name the information it is missing, and you can answer in one short message.\nFormat: describe the finished result The format tells the assistant what the result should look like. If you skip it, the assistant guesses, and its guess is often too long. Be specific about:\nLength: \u0026ldquo;under 150 words\u0026rdquo;, \u0026ldquo;three sentences\u0026rdquo;, \u0026ldquo;one page\u0026rdquo; Structure: \u0026ldquo;bullet points\u0026rdquo;, \u0026ldquo;a table with three columns\u0026rdquo;, \u0026ldquo;a subject line and a short body\u0026rdquo; Tone: \u0026ldquo;friendly but not casual\u0026rdquo;, \u0026ldquo;neutral and factual\u0026rdquo; Language: \u0026ldquo;US English\u0026rdquo;, \u0026ldquo;plain words, no jargon\u0026rdquo; What to leave out: \u0026ldquo;no introduction\u0026rdquo;, \u0026ldquo;no emojis\u0026rdquo; The strongest format instruction is often an example. Paste a short sample of the style you want and add: \u0026ldquo;Follow the structure and tone of this example, but not its content.\u0026rdquo; One good example can replace many words of description. Level 2 shows how to work with several examples and other prompt patterns that work.\nRole: set the point of view A role tells the assistant from which point of view to work: \u0026ldquo;You are an experienced editor\u0026rdquo;, \u0026ldquo;Act as a patient teacher for beginners\u0026rdquo; or \u0026ldquo;Act as a skeptical customer who reads this offer.\u0026rdquo; The role changes the choice of words, the tone and what the assistant pays attention to.\nA role does not add knowledge, though. Calling the assistant a lawyer does not make its legal statements correct. Use roles to set a perspective, not to borrow authority. They help most when you want feedback: \u0026ldquo;Act as a busy manager and tell me what is unclear in this report.\u0026rdquo;\nPutting the blocks together Each block narrows down what the assistant can give you. The task alone produces a usable but generic text. Every block you add makes the answer fit your situation better.\nFrom vague to precise: each block you add removes one kind of guesswork. Here is a complete prompt with all four blocks and an example. The order is a habit, not a rule. What matters is that nothing important is missing.\nAnatomy of a complete prompt for an everyday email. Do all prompts have to be this long? No. A short prompt is fine for quick, general questions, such as \u0026ldquo;Explain what a pivot table is in two sentences.\u0026rdquo; Longer prompts pay off when the result goes to other people, when details matter or when the task comes back every week. Save your best prompts for those repeating tasks. Level 3 shows how to turn them into reusable prompt templates.\nShort or long? The more the result matters, the more blocks you add. The four blocks are not tied to one product. You can use them in ChatGPT, Gemini or any other chat assistant, whichever model runs behind it. Our LLM database lists 343 language models if you want to see what is inside. Specialized tools often turn some blocks into buttons: in a writing tool, you may pick the tone from a menu instead of typing it.\nChatGPT (GPT Image)Text-to-Image · AI at the coreOpenAI\u0026#39;s assistant; creates and edits images from a text prompt in the chat. Google Gemini (Nano Banana)Text-to-Image · AI at the coreGoogle\u0026#39;s assistant; generates and edits images with the Nano Banana models. QuillBotWriting \u0026amp; Research · AI at the coreWriting assistant for paraphrasing, proofreading and summarizing text. Short descriptions from our tool comparison, based on the vendors' information — not tested by ABXK.\nExercises: from vague to complete Practice with three typical requests. For each one, decide which blocks are missing. Then compare your version with ours.\nPrompting Basics: Role, Task, Context and Format (2) Vague prompt Missing A complete version \u0026ldquo;Write an agenda.\u0026rdquo; Context, format \u0026ldquo;Write an agenda for a 45-minute kickoff meeting for our website relaunch. Attendees: marketing, IT and an outside agency. Goals: agree on the timeline and name one contact person per team. A numbered list with times.\u0026rdquo; \u0026ldquo;Plan our team day.\u0026rdquo; Context, format \u0026ldquo;Suggest a schedule for a half-day team event for 12 people. Two team members use wheelchairs, so every activity must be accessible. A table with three columns: time, activity, what to book.\u0026rdquo; \u0026ldquo;Check my slides.\u0026rdquo; Role, context, format \u0026ldquo;Act as a customer who sees our product for the first time. Read the slide text below and list the three points that are hardest to understand, each with a simpler wording. Slides: ### [text] ###\u0026rdquo; Then try it with your own work:\nPick a real task from this week. Write it as one sentence with a clear verb. Add two or three lines of context. Add the format: length, structure and tone. Add a role if you want feedback or a special point of view. Run the one-sentence version and the full version, and compare the two answers. ▸A template to copy\n\u0026ldquo;You are [role]. [Task with a clear verb]. Context: [audience, purpose, facts, limits]. Format: [length, structure, tone]. Before you start, ask me up to three questions if anything is unclear.\u0026rdquo;\nWhen you compare the answers, ask yourself two questions. Which version needed fewer edits? And which block made the biggest difference for this kind of task? The answers tell you where to put your effort the next time.\n","words":1467},"date":"2026-10-14","description":"The four building blocks of a good prompt — role, task, context and format — with examples, a template and exercises for your daily work.","image":"https://abxk.ai/images/cover/prompting-basics-role-task-context-format.webp","json":"https://abxk.ai/en/ai-productivity/learn/prompting-basics-role-task-context-format/index.json","kind":"page","language":"en","learning_path":{"category":"ai-productivity","level":1,"level_name":"Beginner"},"modified":"2026-10-14","related_tools":["ChatGPT (GPT Image)","Google Gemini (Nano Banana)","QuillBot"],"section":"ai-productivity","tags":["prompting","ai-assistants","beginner","productivity"],"title":"Prompting Basics: Role, Task, Context and Format","translations":[{"json":"https://abxk.ai/de/ai-productivity/learn/prompting-basics-role-task-context-format/index.json","language":"de","url":"https://abxk.ai/de/ai-productivity/learn/prompting-basics-role-task-context-format/"}],"url":"https://abxk.ai/en/ai-productivity/learn/prompting-basics-role-task-context-format/"}