CO-STAR Prompt Framework: Write for Any Audience

Study guide

Write for Any Audience with CO-STAR

CO-STAR is a prompt framework that turns a vague request into a useful draft. It stands for Context, Objective, Style, Tone, Audience and Response. You fill six short slots before your actual request, and the AI stops guessing. This course teaches each slot in plain English, with examples from real jobs: a reminder email to staff, a LinkedIn post for clients, a short report for a manager who has never used AI.

It is for busy professionals who write for other people. If your work involves emails, announcements, posts or pages where the reader decides whether it works, this framework is for you. No technical background needed. You will learn where CO-STAR came from, how to fill each part with specific detail, how to fix a draft that feels off, and how to save your best prompts as templates you can reuse in thirty seconds.

What is a prompt framework?

A prompt is the request you type into an AI tool such as ChatGPT, Claude or Gemini. A prompt framework is a reusable structure for that request: a short checklist of the parts a good prompt contains, usually named with an acronym so it is easy to remember. Why it helps: the AI fills in whatever you leave out with generic guesses. A framework makes sure you include the parts that change the answer, such as who it is for, what you want to achieve and the format you need.

Frameworks make prompts repeatable: you fill the same parts each time, save the prompt as a template and share it with colleagues. They are not magic words: the value comes from the detail you put in each part. They help most when a request is complex or open to interpretation; a quick, simple question can stay short. There are well over sixty named frameworks, and most of them remix the same ingredients: role, task, context, audience, format, constraints and examples.

In this course you learn CO-STAR, which is a fill-in-the-slots framework: an acronym where each letter stands for one part of your prompt. It is best for: emails, posts, pages and announcements where audience and tone decide whether it works.

Where CO-STAR comes from and why it works

CO-STAR was published by GovTech Singapore in its Prompt Engineering Playbook in August 2023, written by its Data Science and AI team for practical, repeatable prompting. Data scientist Sheila Teo won Singapore's first GPT-4 prompt engineering competition and wrote it up in December 2023, which is what made the framework widely known. It is now the most cited acronym framework for prompting, and researchers continue to build on it rather than replace it. COSTAR-A, published in 2025, extends the framework by adding an answer anchor, an instruction that defines what a good answer should look like.

The six letters are Context, Objective, Style, Tone, Audience and Response. Each one gives the AI a piece of information it cannot know on its own. Context tells it what happened and what already exists. Objective tells it what the writing must achieve. Style tells it who the writing sounds like. Tone tells it how the writing should feel. Audience tells it who is reading. Response tells it the shape of the deliverable: length, sections, number of versions and the channel.

The labels alone do nothing. The value comes from filling each part with specific, honest detail about your task, your reader and your desired output. A prompt that says 'write an update' gets a generic update. A prompt that says 'write a 120 word update for finance directors who have never used AI, in a warm and direct tone, as an email with a subject line' gets something you can send. The difference is not the acronym. It is the detail you put inside it.

Context and Objective: the situation and the target

Context is everything the model cannot know on its own: what happened, what already exists, and the constraints you are working inside. Without it, the model fills the gap with generic assumptions. Write context as facts, not feelings. Dates, prior messages, limits and who is involved give the model something concrete to reason with, while vague background gives it nothing to use. For example: 'We are moving the office to a new building on 14 March. Staff have asked about parking and desk space. The lease on the old building ends on 28 March.' That is context.

Objective is the outcome the text must produce, phrased as an action the reader will take, such as book a slot this week. A topic describes what to write about. An objective describes what the writing must do. Most weak prompts contain a task but no objective. The model then has nothing to aim at, so it optimises for sounding good instead of working, which is why polished drafts often get no response. Name the action verb in your objective: book, reply, approve, sign. If you cannot name the action you want from the reader, the model cannot aim at it either.

Context and Objective sit at the front of CO-STAR because the later slots depend on them. Style, tone, audience and response format only make sense once the situation and the target are clear. If you change the objective from 'inform staff about the move' to 'get staff to book a desk by Friday', the audience line, the tone and the response format all change with it. Get the first two slots right and the rest of the prompt almost writes itself.

Style and Tone: the voice and the feeling

Style is the writing model the AI should imitate. You can point at a publication, a past piece of your own work, or a known voice such as a friendly support agent. Tone is how the writing should feel to the reader. Common tone words include warm, direct, reassuring, urgent and formal. Style and tone blur together, so keep the split clean. Ask who it sounds like for style, and what emotion it carries for tone. 'Style: like our internal newsletter. Tone: warm and reassuring.' That is a clean split.

Pasting one sample of your style beats any adjective. Two real sentences from your last newsletter teach the AI more than the word professional ever will. For example: 'Style: match this sample: Hi team, quick update on the move. We know change is unsettling, so here is what we know so far.' The model copies the pattern it can see. It cannot copy a description it cannot see.

The S and T lines are short but they carry the voice of the whole prompt. Keep them specific and separate from context and objective. If the draft still feels off, the fix is usually a better sample, not more adjectives. Swap in a closer example of your real writing. If you wrote 'friendly' and got something too casual, paste a sample that shows the level of friendliness you actually want. One good sample is worth ten tone words.

Audience: write for the reader, not the writer

Audience is the A in CO-STAR. It tells the model who reads the text and what they already know, which sets vocabulary and depth. Finance directors who have never used AI need plain words and full explanations, while data scientists need technical terms and no basics. A strong audience line has three parts: who reads it, their knowledge level, and their worry or goal. Together these three parts let the model choose words, examples and depth without you rewriting the draft.

Name the reader's worry or goal. The best text answers the reader's question, not the writer's. If staff fear being locked out of the building during the move, the email should open with that fear before it lists any rules. The audience line changes the first sentence the AI writes. With a named audience and worry, the model starts where the reader is, not where the source document starts. That is the practical payoff of the A slot.

Use one audience per prompt. If two groups need the same message, ask for two versions and check each one fits its reader. A single prompt with two audiences pulls the writing in two directions. For example, one version for staff and one for contractors. Each version can share the same context and objective, but the audience line, tone and examples change. That is faster than trying to write one message that half fits everyone.

Response: the shape you get back

Response is the fifth element of CO-STAR and controls the physical shape of the output: how long it is, how it is structured, which sections it contains, and how many versions you receive. Asking for three versions of 120 words each gives you options, not just one draft. You can choose the strongest, or mix the opening of one with the closing of another. That is useful when you are not sure which angle will land.

Always name the channel. An email needs a subject line plus a body, a LinkedIn post has a character limit, and a slide needs a headline and a small set of bullets. Without a named channel the AI guesses. Format problems are usually missing instructions, not model failures. If you wanted five bullets and got a paragraph, add one sentence describing the structure you want. 'Response: email with subject line, 150 words, three short paragraphs, no bullet points.' That is a complete response instruction.

Response works with the other elements. Context and Objective supply the meaning, Style and Tone supply the voice, Audience supplies the reader, and Response supplies the container that holds it all. If the container is wrong, the content looks wrong even when it is right. A good update in the wrong format still needs reformatting, and reformatting costs you the time the framework was meant to save.

CO-STAR in your job: from one prompt to a weekly habit

The fastest way to learn CO-STAR is to use it on real work you already owe someone, not on practice prompts. Pick a reminder, update or short report you send this week and fill all six slots. Leave no slot empty. If Tone feels hard, write one word such as warm or neutral. A single word still steers the answer and takes two seconds to type.

CO-STAR has no slot for examples, so when voice matters, add one line: here is an example of the voice I want. The AI copies the pattern it can see far better than a description it cannot. CO-STAR also has no slot for steps. For multi-step work, such as research then draft then check, switch to RISEN, which keeps the order of the work in the prompt itself.

Save your best CO-STAR prompts as templates. Next time, change only Context and Objective. Style, Tone, Audience and Response usually stay the same because the readers and the format have not changed. A template turns a five minute prompt into a thirty second edit. That is the real payoff of the framework: reuse, not one perfect prompt. Start with one template for your most common message, and add another when a new recurring task appears.

Frequently asked questions