BROKE AI Prompt Framework: Background, Role, Objective, Key Result, Evolve
The BROKE AI Prompt Framework shapes a model request as Background, Role, Objective, Key Result, and Evolve. Key Result is a test of done. Evolve is a planned second instruction. The pattern does not make a weak brief into expertise.
Framework Card
- Name:
- BROKE AI Prompt Framework
- Goal:
- Shape a model prompt as Background, Role, Objective, Key Result, and Evolve so the first draft is testable and the next turn has a planned refinement.
- Flow:
- Background → Role → Objective → Key Result → Evolve
- Best For:
- Generic model drafts of professional documents; Tasks that need a role, a success test, and a second pass; Prompts that name a persona but never define done
Why it matters
A one-line ask such as “write a press release” produces generic prose because the model was not given the situation, the stance, the job, or a way to know the draft succeeded. People then add a persona (“act as an expert”) and still get costume without a test.
BROKE is a way to put those five parts in order, then spend a second turn on a named improvement. It is a prompt pattern, not a communication strategy for humans.
What it is
BROKE stands for Background (what the model must know), Role (the stance to take), Objective (the task), Key Result (format, constraints, and a success test), Evolve (how to refine after the first draft). Production teaching often shows a press-release example that fills B, R, O, and K and underplays Evolve. Evolve is still native: ask for a tighter lead, a missing proof, or a different audience in a second pass.
CRISPE stresses capacity, insight, personality, and experiment. CO-STAR stresses context, objective, style, tone, audience, and response format. BROKE’s distinctive pair is a checkable Key Result plus a planned Evolve turn. Pyramid and PREP structure arguments for people. They are not prompt patterns.
BROKE does not guarantee model quality. It reduces vagueness in the ask.
How it works
A useful BROKE pass fills five slots from the user’s task, then writes a first prompt and an evolve instruction.
1. Background
Facts the model cannot guess: product, audience, constraints, what already exists.
2. Role
A stance that changes the output (analyst, editor, counsel). A celebrity name is usually costume.
3. Objective
The deliverable in one job (draft, critique, outline). Mixed jobs belong in Evolve or in a split.
4. Key Result
What done looks like: length, sections, claims that must appear, claims that must not, a check the user can run. “Make it good” fails.
5. Evolve
A second instruction after reading the first draft: shorten, add proof, change tone, list gaps. Evolve is planned, not endless.
If Background is missing, say so. Do not invent a company story to complete the acronym.
How it compares
When another lens fits better, or when you need a complementary view, these frameworks do different jobs. They are not interchangeable labels for the same question.
| Framework | What it helps you see | How it differs from BROKE |
|---|---|---|
| CRISPE AI Prompt Framework | Capacity, insight, personality, experiment | Different parts. Personality and experiment are first-class. |
| CO-STAR AI Prompt Framework | Context, objective, style, tone, audience, response | Style and audience are first-class. Key Result and Evolve are BROKE’s pair. |
| Pyramid Principle | Governing thought then grouped support | Human argument order, not a prompt pattern |
| PREP Framework | Point, Reason, Example, Point | Spoken opinion shape, not a model ask |
BROKE is the lens for a testable prompt plus a planned refinement. Other methods help when the question is a different prompt pattern or a human argument structure.
When to Use This Framework
- Generic first drafts. The model repeats public tropes because the prompt was a slogan.
- Role without a test. “Act as a lawyer” with no format or success check.
- Work you expect to iterate. The second turn should be specified, not random chatting.
Example
A concrete example makes the structure easier to reuse when you are under uncertainty.
Example: Press release that sounds like every other launch
Ask: “Write a press release for our new product.” Output is interchangeable.
Implication: BROKE fills Background (what launched, for whom, what is actually new), Role (trade-press writer, not hype intern), Objective (draft the release), Key Result (headline plus three proof points, no superlatives without a source), Evolve (cut 20 percent and add one customer quote placeholder). The pattern improves the ask. It does not prove the product is newsworthy.
Example: Email that needs a second pass
A manager wants a note to a delayed vendor. First drafts are either rude or vague.
Implication: Key Result can require one ask, one date, and no blame labels. Evolve can be “rewrite for a CC to legal: shorter, no adjectives.” That is BROKE. It is not a negotiation strategy.
Takeaway
What BROKE can help with
- Turning a slogan ask into five filled slots
- Making done checkable via Key Result
- Planning a second turn via Evolve
- Catching role-play without a success test
What BROKE cannot replace
- CRISPE. Different prompt parts (personality, experiment).
- CO-STAR. Style, tone, audience as first-class controls.
- Pyramid or PREP. Human argument structure, not a model prompt pattern.
- Domain skill. A complete BROKE prompt cannot invent missing expertise.
- A one-shot lookup. Five parts are unused weight for a simple fact.
Honest scope: BROKE structures a prompt. It does not replace the methods above, and it should not be sold as guaranteed model quality.
Frequently asked questions
If you will only ever send one shot, Evolve can be thin. If you know you will iterate, specify the second turn. Leaving it blank is how people fall back into random chatting.
Objective is the job. Key Result is how you will know the draft succeeded. “Write a memo” is Objective. “One page, three options, no recommendation until the last line” is Key Result.
Five filled slots from the user’s facts, a first prompt, and an Evolve instruction. If the output only says “give more context,” you do not have BROKE yet.
They solve overlapping prompt problems with different parts. Choose by which controls you actually need.
No. It can only ask the model with more precision. Empty Background stays empty.