Prompting basics: the anatomy of a great prompt

You don't need to be a 'prompt engineer' — that's mostly hype. You need a repeatable structure so you stop relying on luck. Almost every weak answer is missing one of four ingredients, and once you can name them you can fix any bad answer in seconds. Here's the structure, plus a pack of reusable prompt skeletons.

Role — who you want it to be (a blunt editor, a patient tutor, a careful analyst); this sets the lens. Task — the one concrete thing you want, stated plainly. Context — the background it can't guess: audience, goal, constraints, an example. Format — how you want the answer shaped: length, bullets or prose, tone. Role primes the thinking, context does the heavy lifting, and format saves you a round of cleanup.

You don't have to label all four every time — that's exhausting. Use them as a checklist when an answer disappoints. Bland and generic? You skipped context. Right idea, wrong shape? You skipped format. Off-tone? You skipped role. This turns 'the AI is bad at this' into 'I under-specified part three', which is always fixable. Add just the missing part and re-send in the same chat.

The single biggest upgrade above the four parts is examples. Describing a style with adjectives ('punchy and warm') is slow and unreliable, because those words mean different things to everyone. Pasting two or three examples of exactly what you want lets the model copy the real pattern — voice, length, structure — instead of guessing. Whenever quality matters, show a sample, don't describe a vibe.

1. 1Role — 'Act as a blunt editor / a patient tutor / a careful analyst.'
2. 2Task — say the one thing you want, plainly.
3. 3Context — give the background it can't guess, plus an example.
4. 4Format — say how you want it back: length, bullets or prose, tone.

Reusable prompt skeletons