The CRAFT Framework
Context, Role, Action, Format, Task — five pieces that turn a vague request into a precise one. CRAFT is the framework to reach for when the quality and structure of the answer really matter, whether you are studying for an exam or building a lesson from scratch.
What is the CRAFT framework?
The CRAFT framework is a simple checklist for writing better prompts. Instead of typing the first thing that comes to mind and hoping for a useful answer, you give the AI five things it almost always needs: the Context of your situation, the Role you want it to play, the Action you want it to take, the Format the answer should follow, and the specific Task or goal you are trying to reach.
Large language models like ChatGPT, Claude, and Gemini are not mind readers. They generate a response by predicting what should come next based on what you gave them — so when your prompt is vague, the model fills the gaps with its own average-case assumptions, and you get a generic answer. CRAFT works because it removes that guesswork. Every element you specify is one fewer assumption the model has to make on your behalf. (For the bigger picture on why this matters, see why better prompts matter.)
CRAFT is particularly well suited to students and educators, because schoolwork is exactly the kind of task where context and format make or break the result. A student who tells the model their grade level and asks for practice questions gets a study tool; a teacher who states their class and the standard they are covering gets a usable lesson instead of filler. The rest of this guide breaks down each element with examples drawn from real studying, teaching, and research situations.
The five elements of CRAFT
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 a first-year biology student preparing for a midterm.
- •I teach 9th-grade English and my class is reading To Kill a Mockingbird.
- •I am writing the literature review for my psychology thesis.
Role (R)
Who should the AI act as?
- •Act as a patient tutor who explains one step at a time.
- •Act as an experienced high-school chemistry teacher.
- •Act as a peer-review partner for academic writing.
Action (A)
What should the AI actually do?
- •Create a study guide from the topics I list below.
- •Write five discussion questions for this chapter.
- •Summarize this research paper in plain language.
Format (F)
How should the answer be structured?
- •Use a bulleted list with one key idea per line.
- •Give me a two-column table of terms and definitions.
- •Keep it under 200 words at a high-school reading level.
Task (T)
What is the specific goal behind the request?
- •Quiz me with 10 questions, then explain each answer.
- •Help me check whether I truly understand the concept.
- •Turn these lecture notes into a set of flashcards.
Action vs. Task?
These two are easy to confuse. The Action is the concrete thing you want produced ("write five questions"); the Task is the underlying purpose ("so I can run a class discussion"). Stating the purpose lets the model make better judgment calls when the Action alone is ambiguous.
Why CRAFT produces better answers
Each element of CRAFT fixes a specific failure mode. Context stops the model from answering the wrong version of your question — it is the difference between a study guide pitched at a graduate seminar and one pitched at a high-school sophomore. Role sets the voice and the depth of explanation: "act as a patient tutor" yields slower, step-by-step reasoning, while "act as a strict examiner" yields tougher questions and less hand-holding.
Action and Task together separate the what from the why, which matters more than it sounds: when the model knows your goal is "to check my understanding," it will challenge you rather than simply hand over answers. Format is the most underused element and often the highest-leverage one — asking for a table, a numbered list, or a strict word count turns a wall of text into something you can actually study from or drop straight into a worksheet.
You do not need to label the parts with the literal words "Context" and "Role" — the model does not require them — but writing them out keeps you honest and makes sure nothing important is missing. That is the real value of a framework: it is a thinking aid as much as a formatting one. If you would rather not assemble the pieces by hand, the CRAFT prompt generator gives you a field for each element and writes the finished prompt for you.
CRAFT in action: before and after
The same goal, written two ways. Notice how much the model has to guess in the first version.
Help me study for my physics exam on momentum.
[Role] Act as a patient physics tutor who explains one idea at a time and checks my understanding before moving on.
[Action] Quiz me with 10 progressively harder problems on one-dimensional collisions, mixing elastic, inelastic, and perfectly inelastic cases.
[Format] Give me one problem at a time, wait for my answer, then tell me whether I am right and explain the reasoning in two or three sentences — including whether momentum, kinetic energy, or both are conserved.
[Task] By the end, I want to reliably decide which conservation laws apply to a given collision and set up the equations correctly.
The vague version produces a generic summary you have probably already read in your textbook. The CRAFT version produces an interactive study session pitched at your exact level, drilling the specific thing you keep getting wrong — telling elastic from inelastic collisions — which is how you actually learn the material.
Make me a quiz about the water cycle.
[Role] Act as an experienced middle-school science teacher who writes clear, fair assessments.
[Action] Write a 10-question formative quiz covering evaporation, condensation, precipitation, and collection.
[Format] Include 7 multiple-choice and 3 short-answer questions, each with a plausible distractor set, followed by a separate answer key.
[Task] I want to quickly see which of the four stages students still misunderstand before I move on.
Because the prompt states the grade level, the standard being assessed, and the purpose (a formative check), the AI produces a quiz you can use almost as-is — with an answer key and distractors that actually probe common misconceptions, rather than a random list of facts.
Common mistakes to avoid
- Skipping Context to save time. Context is what makes the answer yours instead of generic. One sentence about your level, your subject, or your audience changes the entire response.
- Leaving out Format. If you do not say how you want the answer shaped, you get a paragraph. Asking for a table, a checklist, or a word limit is often the single biggest improvement you can make.
- Cramming three requests into one prompt. CRAFT works best on one clear job at a time. If you need a summary and a quiz and flashcards, ask in three turns — the model keeps the context between them.
- A vague Role. "Act as a helpful assistant" tells the model nothing it was not already doing. "Act as a strict thesis advisor" or "act as a tutor for a struggling student" actually shifts the tone and depth.
- Forgetting to give it your material. CRAFT structures your request; the AI still needs the raw content. Paste your notes, the chapter, or the rubric alongside the prompt.
When to use CRAFT (and when not to)
CRAFT is the right tool when the output is substantial and structure matters — study guides, essays, lesson plans, summaries, and rubrics. For other situations, one of the sister frameworks may fit better:
- →TAG (Task, Action, Goal) — for quick, everyday, one-off requests where CRAFT would be overkill.
- →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 are the most important part, such as emails or announcements.
- →CARE — when you need the output in a strict, repeatable format every time.
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 CRAFT stand for in prompt engineering?
CRAFT stands for Context, Role, Action, Format, and Task. It is a structured prompt framework that asks you to spell out the background the AI needs, the persona it should adopt, what you want it to do, how the answer should be shaped, and the specific end goal — so the model has everything it needs to respond well on the first try.
Is the CRAFT framework good for students?
Yes. CRAFT is especially helpful for studying because it forces you to be specific about your level and your goal. Telling the AI you are a first-year student, asking it to act as a patient tutor, and requesting practice questions in a particular format produces far more useful study material than a vague request like 'help me study for my exam.'
Do I have to use all five CRAFT elements every time?
No. CRAFT is a checklist, not a rulebook. For a quick question you might only need the Action and Task. The more important or complex the output, the more the Context, Role, and Format elements pay off. If a single sentence already gets you what you need, the simpler TAG framework is a better fit.
Which AI models does the CRAFT framework work with?
All of them. CRAFT is model-agnostic — the same structured prompt improves results in ChatGPT, Claude, Gemini, Microsoft Copilot, and open-source models, because every large language model responds better to clear context, an explicit role, and a defined output format.
How is CRAFT different from the CO-STAR framework?
Both are structured frameworks, but CRAFT emphasizes Role and Format, which makes it strong for detailed, well-organized documents like study guides, essays, and lesson plans. CO-STAR instead adds an explicit Audience and Tone, which makes it better suited to communication where the reader and the voice matter most.
Ready to try it out?
Open the builder, fill in the five fields, and copy your finished prompt into ChatGPT, Claude, or Gemini.