The Four-Part Prompt Structure

Reliable prompts tend to share four ingredients: role, task, context, and format. Once you can spot which of these is missing from a weak prompt, fixing it becomes almost mechanical.

The four ingredients

Role — who should the AI act as? This sets the lens it responds through. "You're an editor reviewing this for clarity" produces a different, more useful response than no role at all.

Task — what specifically do you want done? Not "help with this email" but "shorten this email to 3 sentences without losing the ask." Specific verbs beat vague ones: summarize, compare, rewrite, list, critique.

Context — what does the AI need to know that it can't guess? Your audience, a constraint, what you already tried, relevant background. If a fact matters to the answer and the AI has no way to know it, this is where it goes.

Format — how should the output be structured? A table, a numbered list, plain prose, a specific word count. Naming this upfront saves a follow-up round-trip almost every time.

Why this order matters

Role first, because it frames everything after it. Task next, because it's the actual instruction. Context follows to fill in what the AI can't infer. Format last, because it constrains the shape of the final output without affecting what gets said.

A worked example

Before (missing all four): "Write about our new pricing."

After:

Role: You're a product marketer writing for existing customers. Task: Announce that we're introducing a new pricing tier. Context: The new tier is aimed at larger teams; existing customers on the current plan are unaffected and don't need to do anything. Format: A 100-word email, plain and direct, no exaggerated claims.

Same underlying request, dramatically more usable output — because nothing is left for the AI to guess.

Checklist to reuse

Before sending a prompt, check: is there a role that would help? Is the task specific, not vague? Have I given context the AI can't infer on its own? Have I said what format I want back?

Key takeaway: most disappointing AI responses trace back to one of these four being silently missing, not the AI being incapable of the task.