Prompt frameworks such as CO-STAR, CARE, RTF and RISEN are acronyms that remind you what to put in a prompt: usually the background, the task, who it is for and the shape of the answer. They differ mainly in length and in which parts they name. None is official at OpenAI or Anthropic; Google publishes its own four-part version. Use whichever one you can remember, and drop the letters a task does not need.
Prompt frameworks at a glance
The plain point: five common frameworks, what the letters mean, and where each one comes from, as far as a source can show.
| Framework | What the letters stand for | Where it comes from |
|---|---|---|
| CO-STAR | Context, Objective, Style, Tone, Audience, Response | GovTech Singapore’s Prompt Engineering Playbook (2023) |
| CARE | Context, Ask, Rules, Examples | Nielsen Norman Group article (May 2024) |
| RTF | Role, Task, Format | Origin unclear |
| RISEN | Role, Instructions, Steps, End goal, Narrowing | Credited to Kyle Balmer in a 2023 newsletter; his own first write-up not found |
| Persona, Task, Context, Format | Who the AI is, the job, the background, the shape of the answer | Google Workspace’s own prompting guide |
Each row is sourced in its section below.
What is a prompt framework?
The plain point: a checklist with a catchy name. It does not change how the AI works; it stops you forgetting a part.
A prompt framework is like the packing list on your fridge before a trip. Nobody needs the list to pack, but it stops you leaving without your charger. The letters are the list. The AI reads only the words you write, not the acronym.
Most frameworks share the same core: the situation, the task, and the form of the answer. Some add the reader, the tone, an example or a step list. That is why they look alike: they are different ways of remembering what makes a good prompt.
CO-STAR: what does it stand for, and where is it from?
The plain point: six letters, written up by Singapore’s government technology agency, GovTech, for its staff.
GovTech’s Prompt Engineering Playbook, produced by its Data Science and AI Division, spells it out: “CO-STAR stands for Context, Objective, Style, Tone, Audience and Response.”1 In 2025 GovTech said its officers still use it.2 The playbook pages this site read do not name who first came up with it, so this page credits the playbook, not a person.
The playbook itself says you need not use every letter: “You do not have to use every single one of them, every single time.”3 One 2025 study, a preprint from Kennesaw State University researchers, found that CO-STAR “improves prompt clarity and aligns outputs for larger LLMs” but works less consistently with small models run on your own computer.4 A preprint has not yet been peer reviewed.
Fits: writing where tone and reader matter, such as emails, posts and reports.
CARE: which CARE do people mean?
The plain point: the best-known CARE is from Nielsen Norman Group (NN/g), a user-experience research firm: Context, Ask, Rules, Examples.
NN/g published it in May 2024: “Use the acronym CARE (context, ask, rules, and examples)”.5 This site also teaches a different CARE of its own (Context, Anchor, Restrict, Exemplify), aimed at image and video prompts. Both are explained on the CARE prompt framework.
Fits: everyday chat requests. The “E” is its strength: an example often says more than a description.
RTF and RISEN: the shortest and a longer one
The plain point: RTF is three letters for quick jobs; RISEN adds steps, a goal and limits for bigger ones.
RTF stands for Role, Task, Format, as listed in a 2023 newsletter round-up of frameworks.6 That round-up gives no origin, and this site found none it could check. Origin unclear.
RISEN stands for Role, Instructions, Steps, End goal, Narrowing. The same round-up credits it to prompt educator Kyle Balmer: “Thanks to Kyle Balmer for creating and sharing this framework.”6 This site could not find Balmer’s own first write-up, so treat that as a credit, not a confirmed origin. “Narrowing” means limits such as length, tone or format.
Fits: RTF for quick, everyday asks; RISEN for multi-step jobs where you want the AI to follow an order.
Persona, Task, Context, Format: Google’s own four
The plain point: of the five, this is the one an AI company publishes for its own product.
Google’s prompting guide for Gemini in Workspace says: “The four main areas to consider when writing an effective prompt are: Persona, Task, Context, Format”. It adds that you do not need all four every time.7 It is close to RTF with context added.
Fits: work tasks in Gmail, Docs and similar tools, or anywhere you want the vendor’s own shape.
One task, five frameworks: a worked example
The plain point: the same request, written five ways. Notice how much they overlap. The task: tell a team the office is closed on Friday for a move.
All five carry the same facts. In this site’s reading, the facts are what make the answer good: the date, the reason, what staff should do. The labels only helped decide what to include.
What do the AI companies’ own guides recommend?
The plain point: clear, specific instructions in plain language, with a role and examples where they help. Not a particular acronym.
- Anthropic says “Claude responds well to clear, explicit instructions”, and offers a test: show your prompt to a colleague with little context; if they would be confused, so will the model.8 It also suggests a role: “Even a single sentence makes a difference”.9
- OpenAI splits instructions into Identity, Instructions, Examples and Context, “usually in this order (though the exact optimal content and order may vary by which model you are using)”.10
- Google gives Persona, Task, Context, Format, and its first tip is “Use natural language.”11
Neither Anthropic’s nor OpenAI’s guide mentions CO-STAR, CARE, RTF or RISEN by name.First-hand: both guides searched for these four names on 19 Oct 2026, with no match. The techniques behind the letters, such as roles, examples and format, are explained on prompting techniques explained, with the vendor quotes for each.
Which prompt framework should you use?
The plain point: this is this site’s reading, not a ranking. Pick the shortest one that covers what your task needs.
- Quick question or small job: RTF, or just one clear sentence.
- Writing for a reader: CO-STAR, because it names tone and audience.
- You have an example to copy: NN/g’s CARE.
- A job with an order to follow: RISEN.
- Work tools such as Gmail or Docs: Google’s Persona, Task, Context, Format.
Frameworks are memory aids, not magic words. The AI reads your facts, not your acronym, so the details you add matter more than which letters you follow.
Questions people ask
What is the best prompt framework?
None of the guides and write-ups on this page names one framework as best for every task.Reasoning, 19 Oct 2026 — based on the GovTech, NN/g, Google, Anthropic and OpenAI pages and the newsletter cited on this page; none of them names one framework as best for all tasks; the COSTAR-A preprint compares only CO-STAR with its own variant. The vendors ask for clear, specific instructions; the right framework is the one that reminds you to give them.
What does CO-STAR stand for?
Context, Objective, Style, Tone, Audience and Response, as spelled out in GovTech Singapore’s Prompt Engineering Playbook (source above).
Do I need a prompt framework at all?
No. Google’s guide says you do not need all four of its parts in every prompt, and GovTech’s playbook says the same of CO-STAR (sources above). A clear request with the right details works without any acronym.
Does a framework work for image and video prompts?
These five were written with chat prompts in mind. Image and video tools each publish their own order; see what the vendors actually say and this site’s own image-focused CARE on the CARE prompt framework.
Where to go next
- “Show me the parts of a good prompt.” What makes a good prompt and the printable six-part prompt checklist.
- “I want ready-made prompts.” The AI prompt library and starter prompts.
- “Build one for me.” The prompt generator.
- “What do the vendors say for chat models?” The LLM prompting cheat sheet.
- “What do these words mean?” Prompt engineering, system prompt, role prompting, few-shot prompting.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. What does the page say a prompt framework is?
- A memory aid that reminds you what to put in a prompt
- A setting that changes how the AI model works
- An official format all AI companies require
Show the answer
Answer: A memory aid that reminds you what to put in a prompt. The page calls frameworks checklists with catchy names: memory aids, not magic words.
2. Which framework on the page is published by an AI company for its own product?
- RISEN
- Persona, Task, Context, Format, from Google
- RTF
Show the answer
Answer: Persona, Task, Context, Format, from Google. The page says Google’s Workspace prompting guide gives Persona, Task, Context and Format.
3. Where does the page say the RTF framework comes from?
- Nielsen Norman Group
- GovTech Singapore’s playbook
- Its origin is unclear
Show the answer
Answer: Its origin is unclear. The page says the round-up that lists RTF gives no origin, and the site found none it could check.
- 1GovTech Data Science & AI Division, Prompt Engineering Playbook (Beta v3), last updated 30 Aug 2023, read at source 19 Oct 2026, page 26“In short, CO-STAR stands for Context, Objective, Style, Tone, Audience and Response.” ↑
- 2GovTech Singapore, Mastering the art of prompt engineering with Empower, 3 Apr 2025, read at source 19 Oct 2026“At GovTech, our officers follow the CO-STAR Framework, featured in our Prompt Engineering Playbook, to craft better prompts.” ↑
- 3GovTech Data Science & AI Division, Prompt Engineering Playbook (Beta v3), last updated 30 Aug 2023, read at source 19 Oct 2026, page 40“You do not have to use every single one of them, every single time. Mix and match depending on what your objectives are”. ↑
- 4Kennesaw State University researchers, COSTAR-A: A prompting framework for enhancing Large Language Model performance on Point-of-View questions, arXiv preprint, submitted 14 Oct 2025, read at source 19 Oct 2026“the original COSTAR framework improves prompt clarity and aligns outputs for larger LLMs”; “its performance is less consistent with smaller, locally optimized models.” ↑
- 5Nielsen Norman Group, CARE: Structure for Crafting AI Prompts, Kate Moran, 24 May 2024, read at source 12 Oct 2026 (as cited on this site’s CARE framework page): “Use the acronym CARE (context, ask, rules, and examples)”.↑
- 6The Prompt Warrior, 5 prompt frameworks to level up your prompts, 13 Oct 2023, read at source 19 Oct 2026“R - Role”, “T - Task”, “F - Format”. ↑a“R - Role”, “I - Instructions”, “S - Steps”, “E - End goal”, “N - Narrowing”; “Thanks to Kyle Balmer for creating and sharing this framework.” ↑b
- 7Google Workspace, Writing effective AI prompts (Prompting guide 101), undated, read at source 19 Oct 2026“The four main areas to consider when writing an effective prompt are: Persona, Task, Context, Format”; “You don’t need to use all four in every prompt, but using a few will help!” ↑
- 8Anthropic, Prompting best practices, Claude developer docs, undated, read at source 19 Oct 2026“Claude responds well to clear, explicit instructions.”; “Golden rule: Show your prompt to a colleague with minimal context on the task and ask them to follow it. If they’d be confused, Claude will be too.” ↑
- 9Anthropic, Prompting best practices, read at source 19 Oct 2026“Setting a role in the system prompt focuses Claude’s behavior and tone for your use case. Even a single sentence makes a difference”. ↑
- 10OpenAI, Prompt engineering, API docs, undated, read at source 19 Oct 2026“usually in this order (though the exact optimal content and order may vary by which model you are using)”; Identity: “Describe the purpose, communication style, and high-level goals of the assistant.” ↑
- 11Google Workspace, Writing effective AI prompts, read at source 19 Oct 2026“Use natural language.”; “Write as if you’re speaking to another person.” ↑
Read at source 19 Oct 2026: GovTech Data Science & AI Division, Prompt Engineering Playbook (Beta v3); GovTech Singapore, Mastering the art of prompt engineering with Empower; COSTAR-A (arXiv preprint); The Prompt Warrior, 5 prompt frameworks to level up your prompts; Google Workspace, Writing effective AI prompts; Anthropic, Prompting best practices; OpenAI, Prompt engineering.
Reused from this site’s CARE framework page, read there 12 Oct 2026: Nielsen Norman Group, CARE: Structure for Crafting AI Prompts.
The worked-example prompts, the packing-list comparison, the “Fits” lines and the advice on which framework to use are this site’s own, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK