AI Prompt Frameworks Compared: RCTF vs RTF, CO-STAR, and CRISPE

Search "prompt framework" and you get an alphabet soup: RTF, RCTF, CO-STAR, CRISPE, TAG, APE, each with its own acronym diagram and its own crowd insisting the others are amateur hour. An owner with fifteen minutes between jobs does not need six frameworks. This is the comparison we wish had existed before we picked one: what each framework asks for, what that buys you, and which one earns a place in a small business.

The short version up front: they are the same idea wearing different name tags. Every framework on this list is a checklist for the same four gaps (who the AI should be, what it cannot know, what to produce, what shape to produce it in). The differences are how many boxes they add on top, and whether those boxes pay rent on a Tuesday afternoon.

The contenders, in one table

| Framework | Asks for | Extra beyond the core four | Weight |

|---|---|---|---|

| RTF | Role, Task, Format | none; drops context | lightest |

| RCTF | Role, Context, Task, Format | none; this is the core four | light |

| CO-STAR | Context, Objective, Style, Tone, Audience, Response | style, tone, audience as separate boxes | medium |

| CRISPE | Capacity, Insight, Statement, Personality, Experiment | personality notes, asking for variants | medium |

| TAG / APE and friends | Task, Action/Purpose, Goal/Expectation | renames, mostly | lightest |

RTF: fast, and missing the part that matters

Role-Task-Format is the framework people learn in an afternoon webinar, and it works for generic jobs: "You are an editor. Shorten this to 100 words. Keep the bullet list." The hole is context, and for a business that hole is the whole game. An editor prompt needs no context; a quote follow-up, a complaint reply, or a job ad without your situation in it produces the beige boilerplate that makes owners quit AI. RTF is fine for tasks where your business is irrelevant. Most of your tasks are not those.

RCTF: the core four, nothing else

Role-context-task-format adds the missing box and stops there. Context is everything the AI cannot know (your business, the customer, the history, the constraint), and it is routinely the longest part of a working prompt. We teach RCTF first in every client engagement and use it for every master prompt we publish, for one reason: four boxes is the most an owner will fill in at 9pm, and the four are sufficient. Ten finished briefs built this way are in our examples collection if you want to see it on real jobs.

CO-STAR: the marketer's version

CO-STAR splits what RCTF calls "format" into style, tone, and audience, and renames task to objective. For marketing copy, the split earns its keep: an Instagram caption and an investor update differ mostly in style, tone, and audience, and forcing three separate answers sharpens all three. The cost is six boxes every time, including for jobs where style and tone are obvious. If most of your AI use is public-facing copy, CO-STAR is a reasonable house standard. If your AI use is spread across quotes, hiring, operations, and support, you will resent the ceremony by Thursday.

CRISPE: variants and personality

CRISPE's contribution is the E, experiment: asking the model for multiple takes by default. That is a genuinely good habit for creative work, and a waste of reading time for an invoice reminder. Its "personality" box mostly duplicates role. We borrow the variants idea as a line inside format ("give me 3 versions") on creative tasks and skip the rest.

The one that wins is the one you fill in

Here is the finding from teaching this for a year: framework choice matters far less than context quality. A mediocre framework with two honest paragraphs of context beats the fanciest acronym with a thin one. All six frameworks fail identically when the context box gets a polite half-truth instead of the real situation. And when tone keeps landing wrong no matter which framework you use, the fix is not another box; it is showing instead of describing, which is few-shot prompting and works with any of them.

So our recommendation holds no suspense: RCTF for everything, CO-STAR's three-way split borrowed inside the format box when you write public-facing copy, CRISPE's variants line for creative work. Not because RCTF is cleverer, but because the framework you consistently fill in beats the framework you abandon.

Where every framework breaks

No framework fixes a wrong decision, checks a fact, or knows your margins. A perfectly briefed prompt for a bad idea returns a well-formatted bad idea. Frameworks structure the asking; the judgment stays yours, and anything touching money, employment, or legal wording still gets human review before it ships.

Putting it to work

If you have a prompt that keeps disappointing, the Prompt Rewriter restructures it into the core four and interrogates you for what is missing. If your prompts work but live scattered across note apps, the Prompt Library Builder is the maintained-library version. And the prompt engineering courses teach the craft from zero, with your business as the example.