CO-STAR is a prompting framework that structures prompts into six components: Context, Objective, Style, Tone, Audience, and Response format. Developed by practitioners at Singapore’s Government Technology Agency, it addresses the most common cause of inconsistent AI output — under-specified prompts that leave the model making too many assumptions. By explicitly defining each component, CO-STAR dramatically reduces output variability and produces results that are calibrated to your specific requirements from the first attempt.
The Six Components
C — Context. The background information the model needs to perform the task. What situation are we in? What has happened before? What is the business context? Context answers the question: “What does the model need to know to approach this task correctly?”
O — Objective. The specific task you want the model to complete. The more precise, the better. Not “write something about our product” but “write a 200-word product description for our new project management tool, highlighting its AI automation features.”
S — Style. The writing style you want: formal, conversational, journalistic, academic, technical, accessible. If there is a specific author, publication, or brand voice to emulate, specify it here.
T — Tone. The emotional register: professional, warm, urgent, reassuring, enthusiastic, authoritative. Style is about how you structure language; tone is about the emotional quality of the communication.
A — Audience. Who will read this output? Their expertise level, their relationship to you, their likely questions and objections, their context. The same information written for a technical expert and for a first-time user should read very differently.
R — Response. The format and structure of the output: length, format (prose, bullet points, numbered list, table), any sections or headings to include, any elements to exclude.
CO-STAR in Practice: Email Example
| Component | Example Content |
|---|---|
| Context | We are a B2B SaaS company. This customer has been with us 2 years and recently downgraded their plan. |
| Objective | Write a re-engagement email offering a personalised incentive to upgrade. |
| Style | Conversational, direct, not salesy. |
| Tone | Warm, genuine — like a customer success manager who cares. |
| Audience | A busy operations manager who values efficiency and has limited budget authority. |
| Response | 150–200 words, no bullet points, one clear CTA at the end. |
When CO-STAR Is Most Valuable
CO-STAR delivers the most value for communication tasks — emails, reports, presentations, proposals, marketing copy — where the relationship between the writer and reader matters and where style and tone significantly affect how the output is received. For technical tasks like code generation, data extraction, and classification, simpler prompt structures are often more efficient. Use CO-STAR when the full six-component specification meaningfully changes the output; use simpler prompts when the task is mechanical enough that only objective and response format matter.
Building CO-STAR Templates for Your Team
The most productive use of CO-STAR is as a template library. Build a CO-STAR template for each major communication task your team performs: client email responses, internal reports, proposal sections, social media posts, job descriptions. Store the templates in your shared prompt library with the CO-STAR components pre-filled for your specific context and just the objective and any task-specific details left blank for the user to fill in. Teams that use CO-STAR templates consistently produce more consistent output quality across different team members, regardless of individual prompting skill.
CO-STAR in Action: A Before and After
The difference a full CO-STAR specification makes is most visible in comparison. Without CO-STAR: “Write an email to a customer about their late payment.” Result: a generic, impersonal reminder that could have been generated for any business and any customer. With CO-STAR: Context: SaaS company, long-term customer of 3 years, first time they have ever missed a payment, relationship managed by their account manager. Objective: Send a friendly reminder that assumes good faith and preserves the relationship. Style: Conversational, warm, direct. Tone: Understanding but clear. Audience: Finance manager who is probably just busy, not in financial difficulty. Response: 100–150 words, no bullet points, end with an easy action (link to pay or reply to discuss). Result: a personalised, relationship-preserving message that reflects the actual situation rather than a generic dunning notice.
The difference in output quality justifies the additional thirty seconds to complete all six components. For high-stakes communications — client relationships, sensitive internal announcements, customer-facing content for large audiences — that thirty seconds is well spent every time.
Adapting CO-STAR for Different Content Types
CO-STAR was designed for communication tasks but adapts to other content types with minor modifications. For marketing content, the Audience component is the most critical — it defines the psychographic and demographic profile of the reader the content is targeting. For technical documentation, Style should specify the technical level and the documentation conventions (imperative or descriptive voice, code example requirements). For internal reports, the Audience component captures the decision-making authority of the reader and whether they want recommendations or just analysis. The framework is flexible enough to handle these variations while maintaining its core value: forcing explicit specification of the six dimensions most likely to affect output quality.
Some teams create CO-STAR template libraries for their most common content types — a template for client proposal sections, one for product update announcements, one for sales outreach emails. The template pre-fills the components that are consistent across instances (company context, brand style, tone guidelines) and leaves the Objective and Audience fields blank for the user to complete for each specific use. This reduces the time to complete a full CO-STAR specification to under a minute while ensuring all six components are always considered.
When to Use a Simpler Prompt
CO-STAR is most valuable for communication tasks where multiple dimensions of the output — not just the content but the style, tone, and reader relationship — matter significantly. For purely functional tasks like data classification, structured extraction, or code generation, a simpler prompt that specifies the task and output format clearly is more efficient. The overhead of six components is not justified when the task is mechanical enough that only two or three of those dimensions are actually relevant. Develop the judgment to apply CO-STAR when the full specification adds value, and use simpler prompts when it does not — the goal is always the best output for the least prompt engineering effort.
Apply CO-STAR to your next important client communication. The output quality difference will confirm whether the framework is worth incorporating into your regular prompting practice.
CO-STAR as an Onboarding Tool
New team members often produce inconsistent AI outputs when they start using AI tools because they have not yet developed prompting intuition. CO-STAR gives them a framework that produces good results from day one, without requiring the months of trial-and-error that develop prompting intuition organically. Include CO-STAR in your AI onboarding material alongside your acceptable use policy and your prompt library. A new team member who understands the six components and has access to your template library can produce on-brand, appropriate-quality AI output in their first week — rather than spending their first month developing prompting skills independently.
CO-STAR templates also serve as implicit brand standards documentation. The Style, Tone, and Audience components of your production templates encode your brand voice requirements in terms that directly translate to AI output. When someone asks “what is our brand voice?”, pointing to a well-configured CO-STAR template gives them a concrete, actionable answer rather than a vague description.
Adapting CO-STAR for Different Team Contexts
Different teams within the same organisation have different communication contexts that call for different CO-STAR configurations. A sales team’s client communications have different style and tone requirements than an engineering team’s technical documentation, which has different requirements than an HR team’s internal communications. Rather than one universal CO-STAR template, maintain team-specific variants with the Style, Tone, and Audience components pre-configured for each team’s communication context. The Objective and Response components remain variable across tasks; the other four components encode the team’s specific communication standards. When a new team member joins, they inherit the team’s CO-STAR configuration and produce on-standard output immediately — without needing to internalise the communication standards that the configuration encodes.
When to Deviate From the Framework
CO-STAR is a tool, not a rule. There are contexts where filling out all six components is unnecessary overhead — a simple formatting request, a quick calculation, a factual lookup. Develop the judgment to apply CO-STAR when the full specification adds value and to write simpler prompts when it does not. The signal that CO-STAR is warranted: when you find yourself re-prompting because the output did not reflect the right style, tone, or audience calibration. The signal that it is not warranted: when a direct, specific instruction produces the output you need on the first attempt. Build CO-STAR fluency so you can use it quickly when it helps, not as a ritualistic process that adds friction to tasks that do not require it.