The best ChatGPT prompt structure — five power patterns for expert output

The four elements that separate expert ChatGPT output from generic:

1. ROLE sets the expertise frame. 'World-class copywriter' gets fundamentally different output than 'assistant.' The more specific, the better: 'a direct-response copywriter who spent 10 years writing for DTC brands' beats 'a copywriter.'

2. CLARIFYING QUESTIONS kill generic answers. Most ChatGPT responses are generic because the AI fills assumptions you didn't make. The question step forces it to surface those assumptions instead of silently baking them in.

3. STRUCTURED OUTPUT with clear headings pushes the AI toward organized thinking and makes long answers navigable.

4. SELF-CRITIQUE is the highest-leverage addition. Asking the AI to improve its own answer before you ask raises the quality ceiling on every response — consistently 30-40% better output, for zero extra effort.

─────────────────────────────────────────
THE CORE TEMPLATE — works for any task:

'You are a world-class [ROLE, e.g. copywriter / strategist / tutor / analyst / coach]. I need help with [WHAT YOU NEED].

Before answering, ask me up to 3 clarifying questions — only if the answer would meaningfully change your response. If the task is already clear, skip the questions and proceed.

Then give your best answer with clear headings. After the answer, list 2 specific ways it could be improved and offer to do them.

My task: [DESCRIBE IN DETAIL].'

─────────────────────────────────────────
FIVE POWER VARIATIONS — match the pattern to your goal:

FOR ANALYSIS:
'You are a strategy consultant. Analyze [TOPIC] from three angles — opportunity, risk, and alternatives. Be direct about what you think is most likely. End with the one question I should be asking that I'm probably not.'

FOR WRITING:
'You are an editor who prioritizes clarity over cleverness. Rewrite [TEXT] to be tighter, more direct, and less generic. Show changes in before/after pairs. Then name the one structural problem that line-level edits can't fix.'

FOR LEARNING:
'You are an expert who can explain [TOPIC] at any level. Start at the simplest level possible. After each explanation, ask if I want to go deeper or if something is unclear. Don't advance until I signal I'm ready.'

FOR DECISIONS:
'I'm deciding between [OPTIONS]. Steelman each option honestly — including the one I'm probably discounting. Then tell me the most important unknown variable and how I'd find out. Give your recommendation last, not first.'

FOR FEEDBACK:
'Give me honest, direct feedback on [WORK]. Start with what's actually working (not just compliments). Then the one thing that would move the needle most. Then secondary issues. Don't soften the critique — I want to improve it.'

─────────────────────────────────────────
WHAT MOST PEOPLE SKIP — and shouldn't:
• The self-critique step ('list 2 ways this could be improved') is the single highest-leverage addition to any prompt. Always include it.
• 'World-class' in the role is not fluff — it statistically shifts output quality upward in every benchmark test.
• The 'only ask if it changes the answer' clause keeps the question step from becoming a questionnaire. Without that clause, ChatGPT asks 5 questions every time.

Tip: chain these patterns — start with the ANALYSIS template to understand a problem, then use DECISIONS to choose a path, then use WRITING to communicate it. Each conversation builds on the last without losing context.