CARE Prompt Framework: Context, Action, Result, Example

Study guide

CARE Prompt Framework: Show One Good Example

CARE stands for Context, Action, Result, Example. The first three letters set the scene, the job and the finish line. The fourth shows the AI what good looks like by pasting in one real case that was handled well. This course teaches you why that fourth slot matters so much, and how to use it in everyday work with ChatGPT, Claude or Gemini.

It is built for busy, non technical professionals who handle cases all day: support tickets, candidate reviews, patient notes, client requests. You will learn how to write a context block from the real ticket, name one clear action, describe the result from the reader's side, and paste one strong example wrapped in example tags. You will also build a small case library and a reusable template for your most common case type.

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 CARE, which is a fill-in-the-slots framework: an acronym where each letter stands for one part of your prompt. It is best for: case-based work: tickets, candidates, patients, clients, where one good example sets the standard.

Why examples beat instructions

CARE gives Example its own slot because examples are among the most reliable ways to steer format, tone and structure. A single sample carries all three at once. The AI can copy a pattern it can see rather than interpret a description. One good example teaches more than a paragraph of rules, and demonstrations survive longer prompts with less drift.

Spell out your letters, because three other CARE variants exist. One uses Context, Ask, Rules, Examples for compliance work, and an empathy version is used in support. If you just write CARE, nobody knows which one you mean. Write Context, Action, Result, Example in full the first time.

Place the example last in your prompt. The AI reads it closest to the point of writing, so the pattern is freshest when the answer is generated. Keep it short. Two or three lines that show the opening, the tone and the closing are enough to lock the shape of the reply.

Context: the case

Context is the first C in CARE and it does the heaviest lifting. It tells the AI which customer, which history and which policy apply, so the answer fits one case instead of an average of all cases. Context has three parts: the customer, the history and the policy. Write them in that order in your head, but put the policy nearest the task so the rule shapes the reply before the AI starts writing.

Retelling a case in your own words loses the details that change the answer. Paste the actual ticket or notes instead, then trim anything the AI does not need. Strip personal data you do not need before pasting. Names, addresses, card numbers and dates of birth rarely change the answer. Plan type, tier or status often does, so keep those.

A simple four slot context block keeps you consistent: customer, history, policy, keep private. The fourth slot tells the AI what to leave out of its reply, which protects the person in the case. Context is not background colour, it is the case file. If a colleague could not act on your context block alone, the AI cannot either.

Action and Result: the job and the finish line

Action is the single job you want done, such as draft the reply, assess the candidate, or summarise the case. Keep it to one clear task so the AI does not split its effort across several jobs. Result describes what a good outcome looks like for the person on the other side. For a customer, that often means they understand the next step and feel heard.

Writing the Result from the reader's side keeps the output useful rather than polite. Polite text can still leave the reader confused, while a reader side Result forces clarity. Action and Result work as a pair. The Action says what to produce. The Result says what it must achieve. If either is vague, the AI fills the gap with guesswork.

A quick test for your Result line is to ask who benefits. If the answer is you, rewrite it. If the answer is the customer, candidate, or colleague reading it, you have it right.

Example: one case handled well

An example does three jobs at once. It sets the structure of the answer, the tone of the writing and the level of detail. Saying what makes your example good, in one short line, helps the model copy the right things. Anthropic recommends example tags so the model can tell your sample apart from your instructions. Everything inside the tags is the sample, everything outside is the rule.

For best results Anthropic recommends three to five diverse examples. Diverse means the cases differ in a useful way, such as a simple case, a case with a missing document and a case with a limit. Bad examples teach bad habits. If you paste a rushed or vague case, the model copies that shape. Pick your best case, not your most recent.

The example sits after Context, Action and Result. It shows what good looks like rather than describing it, which is why it often changes the output more than any other part of CARE.

Building a case library

A case library is a small collection of model answers, one per case type you handle often. It saves you from writing every prompt from scratch and keeps your results consistent. When you paste an example into CARE, the model uses it as a pattern for tone, structure and detail level. It does not copy the example word for word, so you still get a fresh answer for each new case.

Vary the details in your examples, such as names, amounts and addresses. If every example uses the same surface details, the model may copy them into new answers by mistake. After each use, compare the new answer to your example. If the new answer is better, update the library by replacing the old example. This keeps your baseline improving over time.

Keep the library small. One strong example per case type is enough. Too many examples can confuse the model and make it harder to pick the right pattern. Review your library every few weeks. Remove examples that no longer match your current tone or process, and add new case types as they appear in your work.

CARE in your job

Build your template around your most common case type, not your hardest task. Frequency is what makes a template worth the effort, because you reuse it again and again. The Example slot carries the standard. Without a good example, CARE loses its main strength, so keep one strong past version of the task ready to paste in.

If you have no example, use RACE and describe the expectation in words. Say what good looks like instead of showing it, and the model still has a target. Add a critique step after the draft. Ask the model to compare its draft with your example and fix the differences, so the model reviews itself against your standard.

Keep your example short and clean. A long example invites the model to copy the wrong parts, such as old dates or names that no longer apply. Review your template every few months. A template that is never updated drifts away from what actually works in your job.

Frequently asked questions