Prompt Frameworks: The Complete Guide

Study guide

Prompt Frameworks: The Complete Guide

There are more than sixty named prompt frameworks in circulation, and new acronyms appear every month. This course gives you the map first. You will learn why frameworks exist, what all the acronyms have in common, and which few you actually need for the work you do. Instead of memorising names, you will learn the ten ingredients behind every framework, then see how each named framework is simply a selection and order of those ingredients.

The course is built for busy, non-technical professionals who paste prompts into ChatGPT, Claude or Gemini. You will work through quick frameworks like RTF, APE and TAG, writing frameworks like CO-STAR, CRAFT and BAB, and structured frameworks like RISEN, RACE, CARE and TIDD-EC. Every lesson explains the why, the how, a worked example in a real job, and a common mistake. By the end you will be able to look at any new acronym and see the same parts underneath, then choose the right level of structure for the task in front of you.

Why prompt frameworks exist

A prompt framework is a checklist of what to include in a single request, so you do not forget the parts that change the answer, such as audience, format, tone and length. There are well over sixty named frameworks in circulation. Most of them reshuffle the same ingredients under different acronyms, so learning the ingredients matters more than learning the names.

The 5W3H study from 2026 found that structure helps most when a request is ambiguous. On fuzzy business tasks it cut follow-up rounds from about 3.3 to 1.1. The same study found structure can hurt on simple requests. If the answer is obvious, adding slots just adds noise, so match the structure to the task.

The real goal is not memorising acronyms. It is building the habit of never forgetting the ingredient that matters for the task in front of you. A framework is a reminder, not a rule. When you understand why the slots exist, you can decide which ones to use and which to leave out.

The ten ingredients behind every acronym

Every prompt framework is built from the same ten ingredients. The core three are role, task and goal. Role tells the model who to be, task tells it what to do, and goal tells it what a good answer looks like. Most weak prompts are missing the goal. The material and shape ingredients are background, audience, style and format. Background supplies the situation, audience names the reader, style sets the tone and format decides whether the answer is a table, an email or a list.

The guardrail ingredients are fences and sample. Fences are the do and do not rules that narrow the answer. A sample shows the model a past example so it copies the structure you already like. The loop is the habit of iterating, evolving and sense checking. You read the first draft, ask for changes and verify facts yourself. A prompt is a conversation, not a single throw.

Once you can name the ten ingredients, a new acronym is just a selection and order of them. CO-STAR picks role, task, style, tone, audience and response. RISEN picks role, instructions, steps, end goal and narrowing. Same parts, different order. Not every prompt needs all ten. A quick question may need only task and format. A high stakes document may need all ten. Choosing which ingredients to include is the real skill.

Quick frameworks: RTF, APE and TAG

RTF stands for Role, Task, Format. It is the default for quick everyday asks because it covers who, what and how in three short slots you can write in under thirty seconds. APE stands for Action, Purpose, Expectation. The Purpose slot is the powerful one, because saying why you need something changes what the model optimises for. TAG stands for Task, Action, Goal and suits decisions. A second TAG exists, Task, Audience, Guardrails, so always spell out the letters when you write or share a prompt.

Use a quick framework when the task is small and the risk of a wrong answer is low. For bigger or riskier work, step up to a heavier framework. Quick frameworks are not weaker versions of the big ones. They are the right size for small jobs, and using a heavy framework on a tiny task just wastes your time.

A common mistake is reaching for a heavy framework when a quick one would do. If you need a one line summary of a meeting, RTF is enough. If you need a policy that will be read by a regulator, you need more. The skill is matching the size of the framework to the size of the job.

Writing frameworks: CO-STAR, CRAFT and BAB

CO-STAR stands for Context, Objective, Style, Tone, Audience and Response. It came from GovTech Singapore's prompt playbook in 2023 and is the most cited writing framework, which makes it a safe default for public writing. CRAFT stands for Context, Role, Action, Format and Tone. It is a lighter five-slot alternative for everyday business writing, useful when you know the reader and want a fast, clean draft. BAB stands for Before, After, Bridge. It is a copywriting shape, not a full prompt framework, so it can sit inside a CO-STAR or CRAFT prompt and guide the structure of the writing itself.

Reach for these frameworks whenever someone else will read the result. Emails, posts, pages and announcements all benefit because the AI is told who the reader is and how the text should sound. The audience slot does the heaviest lifting. Studio owners who already use our invoicing product gives the AI far more to work with than a general audience, and the wording changes accordingly.

Vague slots such as friendly tone or professional style waste the framework. If a slot could describe any company anywhere, it is not specific enough to change the output. The fix is to replace adjectives with details. Instead of friendly, say short sentences, no jargon, one clear next step. Instead of professional, say formal but plain, no contractions, no exclamation marks.

Structured work: RISEN, RACE, CARE and TIDD-EC

RISEN stands for Role, Instructions, Steps, End goal and Narrowing. It was created by Kyle Balmer and suits multi-step deliverables like plans and policies. The Steps slot sets the order of work, and Narrowing sets the limits. RACE stands for Role, Action, Context and Expectation. The Expectation slot is its strength because it makes explicit how you want the answer delivered, such as showing working, citing sources or marking uncertainty.

CARE stands for Context, Action, Result and Example. It is built around one example of a case handled well, which the AI then uses as the pattern for its answer. This suits repeatable case notes and responses. TIDD-EC stands for Task type, Instructions, Do, Don't, Examples and Context. The Do and Don't lists give explicit rules for compliance-heavy work, covering required wording and forbidden claims.

A structured framework only works when every slot is filled. Leaving Steps, End goal or Do and Don't empty produces a generic answer with no order and no rules, which defeats the purpose of choosing the framework. If you find yourself skipping slots, step down to a lighter framework until you have the information you need to fill them.

Choosing the right framework for your job

Match the task to the framework. Quick and low-risk work suits RTF or APE, writing for other people suits CO-STAR or CRAFT, and multi-step work suits RISEN. Analysis fits RACE, compliance fits TIDD-EC, and persuasion fits BAB or AIDA. Reasoning techniques layer on top. Chain of thought and few-shot examples are not separate frameworks. They sit on top of whichever framework you choose, asking for steps or showing sample outputs.

The model makers agree. Anthropic, OpenAI and Google all advise the same basics. Define the outcome, give context and examples, then let the model work. Two frameworks beat fourteen. Keeping two frameworks for your own job and using them until they are habit builds real skill. The other courses in this series go deep on each one.

Frameworks are scaffolds, not scripts. The acronym reminds you what to include. The words you write still come from your own job knowledge and your own judgement about what good looks like. A framework will not tell you what to say. It will make sure you tell the model everything it needs to say it well.

Frequently asked questions