The 5-part prompt structure

Role: who the model should act as.
Goal: what success looks like.
Context: inputs, audience, constraints, examples.
Format: exactly how the answer should be shaped.
Checks: rubric / edge cases / “ask clarifying questions”.

If you only do one thing: be explicit about format and constraints.

Common mistakes (and fixes)

Vague goal: define what “good” means (success criteria).
No format: specify bullets/table/JSON and required fields.
Too many constraints: keep to 3–6 key constraints.
No examples: provide 1–2 examples when style matters.
No verification: add a rubric or a quick self-check.

Use the Checklist → when you want a fast score + fix list.

Reducing hallucinations

Ask for explicit assumptions.
Require citations/links when summarizing sources.
Use “I don’t know” allowance + clarifying questions.
Use checklists/rubrics instead of “make it good”.

Tip: for important work, run “Draft → Critic → Revise”. It’s cheap quality.

Prompt debugging (2-minute loop)

Restate: ask the model to restate the task + constraints in 3 bullets.
Pin format: force headings/JSON and required fields.
Test one edge case: “If input is missing X, what do you do?”
Revise once: apply fixes, then re-run with the same format.

Before answering, restate:
- Your understanding of the goal
- The key constraints
- The exact output format

If anything is ambiguous, ask up to 3 clarifying questions first.

Copy/paste: prompt starter

Role: You are <role>.
Goal: <goal>.
Context: <inputs, audience, constraints>.
Output format: <exact format>.
Quality check: Before finalizing, list 3 potential mistakes and fix them.

Open Prompt Lab Browse templates

Creator workflow: generate a post

Pick one idea: one technique, one insight, or one news item.
Write the hook (1 line). Then 3–5 bullets. Then 1 CTA.
Adapt to platform (X short, LinkedIn more context, IG caption story style).

Write 6 post variants about: <topic>
Platform: <X | LinkedIn | IG>
Audience: <who>
Tone: <direct, helpful>
Structure: Hook → 3–5 bullets → CTA
Constraints: no hype, concrete examples, under <length>

Creator workflow: carousel outline

Carousels win when each slide contains one idea.

Create a 7-slide carousel outline.
Topic: <topic>
Slide rules: 1 idea per slide, short text.
Slides: 1 Hook, 2–6 Value, 7 CTA.
For each slide: title + 2 bullets + suggested visual.

Image prompting (for creators)

Good image prompts specify subject, scene, style, composition, and constraints. If you want consistent branding, define a reusable style block.

Generate an image for a social post.
Subject: <subject>
Scene: <setting>
Style: minimal editorial, clean gradients, high contrast
Composition: centered subject, lots of negative space, 4:5
Text: none
Brand colors: violet + cyan accents
Deliver: 3 variations

For multi-turn image work, use an “edit” loop: generate → pick best → request variations/edit.

For software engineers: high-leverage prompts

1) PRD to tickets
Input: PRD text
Output: 10–20 tickets with title, acceptance criteria, risks, estimate

2) Code review
Give: issues (severity), minimal diff patch, tests to add

3) Incident postmortem
Output: timeline, root cause, contributing factors, action items, owners

Tip: ask for “minimal patches” and “acceptance criteria” to keep output actionable.

For creators: repurpose one idea into 5 formats

Repurpose this idea into:
- X thread (5 tweets)
- LinkedIn post (hook + 3 bullets)
- IG carousel (7 slides)
- Short script (45 seconds)
- Newsletter snippet (150 words)

Constraints: concrete, no hype, include 1 example.

Newsletter workflow (simple)

Collect sources → summarize into your 4-bullet news format.
Pick 10–15 techniques → each gets a name, use-case, example prompt.
Run a rubric check: clarity, specificity, usefulness.
Ship.

What to publish (high value)

Patterns that generalize (not one-off tricks).
Small examples that readers can re-use.
“When NOT to use this” notes.
Short, consistent format (builds habit).

How to Create Better AI Content (Fast)