Prompt Engineering Frameworks Compared: CRAFT, RTF, RACE, CO-STAR and PARTS

Search for prompt engineering and you hit a wall of acronyms: CRAFT, RTF, RACE, CO-STAR, PARTS, TAG. They all try to solve the same problem - a vague prompt produces a vague answer - and they mostly overlap. This guide compares them properly, using one identical task, so you can pick one and stop collecting frameworks.

The task we will use

Write an email to a client explaining a two-week delay on a project deliverable.

Every framework below gets the same job.

RTF - Role, Task, Format

The simplest one, and a good starting point.

Role: You are an account manager. Task: Write an email explaining a two-week delay. Format: Under 150 words, three short paragraphs.

Strengths: memorable, fast. Weakness: no place for context or constraints, so the model invents details.

CRAFT - Context, Role, Action, Format, Tone

CRAFT adds the two things RTF misses: the situation and the voice.

Context: The delay is caused by a late third-party integration; the client is already frustrated. Role: Account manager. Action: Explain the delay and propose a new date. Format: Email, under 150 words. Tone: Direct, accountable, no excuses.

This is the framework we teach, because context and tone are where most business writing goes wrong. Full breakdown: The CRAFT Prompt Framework.

RACE - Role, Action, Context, Expectation

Nearly identical to CRAFT with tone folded into "expectation". Useful if you think in terms of a success criterion rather than a style.

Expectation: The client should reply agreeing to the new date without escalating.

Framing the outcome instead of the format is genuinely useful for persuasive writing.

CO-STAR - Context, Objective, Style, Tone, Audience, Response

The most detailed of the common acronyms, and the best fit for marketing and content work, because it separates style (how it is written), tone (how it feels) and audience (who reads it).

Its cost is length. For a one-line question, CO-STAR is overkill.

PARTS - Purpose, Audience, Role, Task, Structure

A teaching-friendly variant popular in education and internal enablement. It front-loads purpose and audience, which makes it strong for documents that will be read by many people.

Side by side

What they all have in common

Strip the letters away and every framework asks for five things:

1. Who the model should be
2. What the situation is
3. What exactly you want done
4. Who will read the result
5. What the output should look like

If your prompt answers those five, the acronym you used does not matter.

Which one should you learn?

Learn one properly. Rotating frameworks is the main reason people feel stuck: they keep changing the wrapper instead of improving the inputs. We use CRAFT throughout our course because it is short enough to type from memory and complete enough for real work.

If you want the beginner's grounding first, start with What Is Prompt Engineering?, then apply the framework safely using the SAFE checks for AI at work.

Frequently asked questions

What are the main prompt engineering acronyms? The most common are RTF, CRAFT, RACE, CO-STAR and PARTS. They all structure a prompt around role, context, task and output format.

Which prompt framework is best? CRAFT is the best default for everyday professional work. CO-STAR suits marketing content, RTF suits quick questions.

What does the PARTS framework stand for? Purpose, Audience, Role, Task, Structure. It is used mainly in training and internal enablement material.

Do prompt frameworks still matter with newer models? Yes. Newer models forgive sloppy grammar but cannot guess your context, audience or constraints. Those still have to be supplied.

How long should a prompt be? Long enough to include context and constraints, usually four to ten lines for business tasks. Length for its own sake does not help.

Next steps

• Practise with the free prompt library
• Take the free Module 1 of the AI POWERED course
• Read How to Use AI at Work