A junior analyst writes this prompt: "Write a marketing email about our new product." The output is generic — bland subject line, off-brand voice, doesn't fit the audience. They're considering revising the prompt using the CRAFT framework (Context, Role, Action, Format, Tone) they learned in class.
Which revision applies CRAFT correctly?
Why did you pick that answer? Two or three sentences. The act of articulating it is what builds the judgment — not the click that follows.
CRAFT works because it forces you to specify what most prompts leave implicit. Context tells the model the situation (who you are, who the audience is); Role gives it a coherent persona that integrates style and judgment; Action names the specific deliverable; Format constrains structure; Tone sets voice. Each element is doing real work — the prompt isn't longer for the sake of being longer; it's longer because each addition removes a degree of freedom that produces generic output. The other options either add no structure (A, D) or add structure without the conceptual scaffolding that makes CRAFT effective (C).
This is the genuinely tempting wrong answer because adding more adjectives feels like adding more guidance. But adjectives compete with each other — engaging pulls toward attention-grabbing, professional pulls toward restrained, conversion-focused pulls toward sales-y, brand-appropriate depends on a brand the prompt doesn't specify. Stacking adjectives without structural dimensions gives the model conflicting pulls, and the output reflects whatever the model resolves them as. CRAFT works because it separates the dimensions (Context, Role, Action, Format, Tone) instead of stacking adjectives in one bucket.
Listing constraints in one paragraph is better than nothing, but it conflates structural instructions with stylistic ones and provides no role or context. CRAFT separates the dimensions because each plays a different role in shaping output.
This is the genuinely tempting wrong answer because examples are powerful — few-shot prompting works, and good marketing teams have actual past emails they can use. The problem: examples without surrounding framework produce variations on the examples, not appropriate output for a different situation. Three past emails don't tell the model that the new product is a B2B SaaS HR dashboard, or that the audience is existing customers, or that the tone should be conversational like a teammate sharing news. Examples are powerful combined with CRAFT, not as substitutes for it.