The 4-Part Prompt Framework: Role, Context, Constraints, Format

Most people blame the model when AI gives them a mediocre answer. The model is rarely the problem. The same ChatGPT, Claude, or Gemini that writes a generic listicle for one person writes a board-ready strategy for another — because the second person structured the request. This is that structure.

Why vague prompts fail

When you type "give me marketing ideas," the model has no idea who you are, what you sell, what you've tried, or what a useful answer looks like. So it does the safe thing: it averages every marketing article it has ever seen and hands you the blandest possible list. It isn't being dumb — it's being under-instructed.

The fix is not a longer prompt. It's a structured prompt. Here are the four parts, in the order you should write them.

1. Role — who the AI should be

The first line should tell the model what expert to become. "Act as a senior direct-response copywriter" primes a completely different vocabulary, set of references, and standard of quality than answering as a generic assistant.

Weak

write me some ad copy

With a role

Act as a senior direct-response copywriter who has written ads that scaled past seven figures.

Be specific and senior. "A marketer" is weaker than "a growth marketer who has scaled DTC brands from $0 to $1M." The role sets the ceiling for the answer.

2. Context — your real situation

This is the part people skip, and it's the one that matters most. The model can only tailor an answer to the details you give it: your audience, your offer, your budget, your timeline, what you've already tried.

My product is a $49/mo meal-planning app for busy parents. My budget is $500 and I have 5 hours a week. I've tried Instagram with no traction.

Use bracketed placeholders for the specifics so you can reuse the prompt: [audience], [budget], [what you've tried]. Real context is the difference between advice and your plan.

3. Constraints — what to prioritize and rule out

Left unconstrained, the model wanders. Constraints point it at what actually matters to you and fence off what doesn't.

Prioritize channels with the fastest time-to-first-sale. Rule out anything that needs paid ads or a large existing audience. Keep it realistic for a 5-hour week.

Good constraints include priorities ("optimize for speed, not scale"), exclusions ("no paid ads"), and scope ("under $500, this month").

4. Format — exactly how to deliver it

Finally, tell the model the shape of the answer. "A numbered 5-step plan, each step with one action, one tool, and a done-check" produces something you can execute. Without a format instruction you get a wall of prose you then have to reorganize yourself.

Format as a numbered list of 5 channels. For each: the first action to take this week, the realistic 30-day result, and one free tool.

Putting it together

Here's a vague prompt and the same request rebuilt with all four parts.

Before

give me marketing ideas

After

Act as a senior growth marketer who has scaled DTC brands from $0 to $1M. My product is a $49/mo meal-planning app for busy parents; my budget is $500 and I have 5 hours a week; I've tried Instagram with no traction. Recommend 5 marketing channels ranked by speed-to-first-sale. Rule out anything needing paid ads or a big existing audience. Format as a numbered list — for each channel give the first action to take this week, the realistic 30-day result, and one free tool.

Same model. Completely different answer. The first returns a listicle; the second returns a plan you can start tonight.

Don't guess — score it

You don't have to eyeball whether a prompt has all four parts. Our free prompt grader scores any prompt 0-100 on exactly these four dimensions and tells you which one is weak. If you already have a rough prompt, the prompt improver rewrites it into this structure for you, and the prompt builder assembles one from a few quick questions.