The CARE Framework
Context, Action, Result, Example — four pieces that lock the AI into a precise, repeatable format. CARE is the framework to reach for when the shape of the answer matters as much as the content: a grading rubric that looks the same every time, worked solutions in a fixed template, feedback you can hand back without editing.
What is the CARE framework?
The CARE framework is a structured way to write prompts when you need the answer in a specific, consistent format. You give the AI four things: the Context of your situation, the Action you want it to take, the Result rules the output must obey, and an Example that shows exactly what a good answer looks like. The Example is what makes CARE special — instead of only describing the format you want, you hand the model a sample to copy.
Large language models like ChatGPT, Claude, and Gemini are excellent at pattern-matching, but they are not mind readers. Ask vaguely and the model fills the gaps with its own average-case assumptions, so the format drifts from one answer to the next. CARE removes that drift. The Result element states the hard rules, and the Example shows the model the pattern directly — and showing is far more reliable than telling. (For the bigger picture on why this matters, see why better prompts matter.)
CARE is a natural fit for students and educators, because so much academic work depends on a fixed structure. A teacher who grades thirty essays wants every rubric filled in the same way; a physics student wants every worked problem laid out in the same Given / Find / Equations / Solution / Check order so nothing gets skipped under exam pressure. When the format has to be identical every time, an example beats a paragraph of instructions. The rest of this guide breaks down each element with examples drawn from real studying, teaching, and research situations.
The four elements of CARE
Each card shows the question the element answers, with examples for studying, teaching, and research.
Context (C)
What background should the AI know about your situation?
- •I am grading 30 lab reports for an intro physics course.
- •I am a student writing up every homework problem the same way.
- •I am logging sources for a literature review and need uniform entries.
Action (A)
What should the AI actually do?
- •Score this essay against the four-criterion rubric below.
- •Lay out the solution to this kinematics problem.
- •Turn each source into a uniform annotated-bibliography entry.
Result (R)
What rules must the output obey?
- •Use exactly these five headings, in this order, every time.
- •Keep each comment under 40 words; no praise without a reason.
- •Show units on every quantity and round to three significant figures.
Example (E)
What does a perfect answer look like?
- •One filled-in sample rubric for a model essay.
- •A single worked problem in the exact template to copy.
- •One finished bibliography entry showing punctuation and order.
Why the Example does the heavy lifting
AI models learn formats by imitation far better than by description. Handing the model one finished sample is called one-shot prompting; giving it two or three is few-shot prompting. Either way, the model studies the pattern you provided — the headings, the order, the punctuation, even the tone — and reproduces it. A sentence like "use a clear, consistent layout" is open to interpretation; a sample is not. When you genuinely need the same structure every single time, the Example is the element that makes it happen.
Why CARE produces better answers
Each element of CARE removes a different kind of guesswork. Context tells the model what world the task lives in, so it does not answer the wrong version of your question. Action states the single job to be done, keeping the response focused instead of wandering. The real power, though, is in the last two elements working together.
Result sets the explicit rules — the headings to use, the length to hit, the things to leave out — and Example shows those rules in action. Rules alone can be misread; an example alone can be copied too loosely. Side by side they reinforce each other, which is why CARE is so dependable for anything you have to produce repeatedly. Grade twenty essays with the same filled-in sample rubric attached, and all twenty come back in the same shape, ready to compare at a glance.
You do not need to label the parts with the literal words "Context" and "Result" — the model does not require them — but writing them out keeps you honest and makes sure the example you provide actually matches the rules you stated. That is the discipline CARE enforces. If you would rather not assemble the pieces by hand, the CARE prompt generator gives you a field for each element and writes the finished prompt for you.
CARE in action: before and after
The same goal, written two ways. Notice how the Example locks the AI into one repeatable format.
Give me some feedback on this student lab report.
[Action] Read the lab report I paste below and write feedback against four criteria: Hypothesis, Method, Data & Analysis, and Conclusion.
[Result] For each criterion, give a score out of 5 and one or two sentences of specific, actionable comment. End with a single "Next step" line. Do not add any other sections or commentary.
[Example] Match this format exactly:
Hypothesis — 4/5. Clearly stated and testable, but you did not predict the direction of the effect. State which way the period should change before collecting data.
Method — 3/5. Steps are reproducible, but the controlled variables are not listed. Add a sentence naming what you held constant.
...(continue for Data & Analysis and Conclusion)...
Next step: Rework the uncertainty estimate before resubmitting.
The vague version produces a loose paragraph that looks different for every student, which makes grades hard to defend. The CARE version pins the output to a fixed four-criterion shape with a model entry the AI copies — so all thirty reports come back scored the same way, in the same order, ready to skim side by side.
Solve this projectile motion problem for me.
[Action] Solve the problem I paste below and write it up in a strict five-part template: Given / Find / Equations / Solution / Check.
[Result] List each known quantity with its symbol and units under Given. State the target variable under Find. Write the equations symbolically before plugging in numbers. Carry units through every line and round the final answer to three significant figures. Under Check, confirm the units of the answer and whether its size is physically reasonable.
[Example] Use this exact layout (sample: a ball dropped from rest):
Given: h = 20.0 m, v₀ = 0, g = 9.81 m/s²
Find: time to reach the ground, t
Equations: h = ½ g t²
Solution: t = √(2h / g) = √(2 · 20.0 / 9.81) = 2.02 s
Check: Units √(m / (m/s²)) = s ✓ — and ~2 s for a 20 m drop is reasonable.
The vague version hands back a one-off answer with no structure, which teaches you nothing about how to organize your own work. The CARE version forces the symbolic-first, units-checked layout every time — the same discipline that earns method marks on an exam — and the Example removes any doubt about what each section should contain.
Common mistakes to avoid
- Skipping the Example. It is the element that makes CARE worth using. Describing a format in words leaves room for drift; pasting one finished sample removes it. If you only have time for one element, make it the Example.
- An example that contradicts your rules. If your Result says "under 40 words" but your sample runs to 60, the model has to guess which one you mean — and it usually copies the sample. Make the example obey the rules you just wrote.
- A sample that is too complicated. The example is a pattern, not the answer. Keep it short and clean so the structure stands out; a long, busy sample buries the format you actually want copied.
- Forgetting to forbid extra commentary. Models love to add a friendly preamble. If you need clean, drop-in output, say so in the Result — "return only the rubric, no introduction" — or it will wrap your format in chatter.
- Forgetting to give it your material. CARE structures your request; the AI still needs the raw content. Paste the essay, the problem, or the source list alongside the prompt, clearly separated from the example.
When to use CARE (and when not to)
CARE is the right tool when the format is the whole point — anything you produce repeatedly and need to look identical: rubrics, feedback templates, worked solutions, flashcard sets, uniform bibliography entries. For other situations, one of the sister frameworks may fit better:
- →CRAFT (Context, Role, Action, Format, Task) — the all-rounder for substantial documents like study guides, essays, and lesson plans.
- →RISE (Role, Input, Steps, Expectation) — for multi-step assignments where you want the AI to follow a procedure.
- →CO-STAR — when the audience and tone of the message matter most, such as emails or announcements.
- →TAG (Task, Action, Goal) — for quick, everyday, one-off requests with no strict format to enforce.
Not sure which to pick? How it works walks through the whole methodology, and the use cases page shows the frameworks applied to real situations.
Frequently asked questions
What does CARE stand for in prompt engineering?
CARE stands for Context, Action, Result, and Example. It is a structured prompt framework built for precise, repeatable output: you give the AI the background it needs, the task you want done, the exact rules the answer must follow, and a worked example of the format you expect it to copy. The Example is what sets CARE apart — it shows the model the pattern instead of just describing it.
When should I use CARE instead of CRAFT?
Use CARE when the shape of the answer matters as much as the content — when you need the same strict format every time, such as a grading rubric, a feedback template, or worked solutions laid out identically. CRAFT is the better all-rounder for substantial documents like study guides and essays. The deciding question is simple: if you could hand the AI a sample and say 'make it look exactly like this,' CARE is the tool.
Why is the Example element so powerful?
Showing beats telling. A description of a format leaves room for interpretation, but a concrete sample removes the guesswork — the model copies the structure, labels, and tone it can see. This is called one-shot prompting (one example) or few-shot prompting (several), and it is one of the most reliable ways to get consistent output from any AI model.
Do I have to use all four CARE elements every time?
The Example is the heart of CARE, so leaving it out usually defeats the purpose. Context and Action you will almost always want. The Result element — your explicit rules — pays off most when there are hard constraints to enforce, like a word limit, a fixed set of sections, or 'no extra commentary.' For a quick one-off question with no format requirements, a simpler framework like TAG is a better fit.
Which AI models does the CARE framework work with?
All of them. CARE is model-agnostic — the same example-driven prompt improves results in ChatGPT, Claude, Gemini, Microsoft Copilot, and open-source models, because every large language model is good at pattern-matching. Give any of them a clear sample to imitate and the output snaps into the format you want.
Ready to try it out?
Open the builder, fill in the four fields — including your example — and copy your finished prompt into ChatGPT, Claude, or Gemini.