Ask for structure, not just content
Return the following as markdown: ## Problem (2 sentences), ## Options (numbered list, one line each), ## Recommendation (one paragraph), ## Risks (bullet list).
Why it works: Explicit structure makes outputs scannable and comparable across runs.
Specify negative constraints
Write a blog post opener. No clichés. Do not use: 'in today's fast-paced world', 'unlocking the power of', 'the key to success'. Start with a specific concrete scene.
Why it works: Telling the model what NOT to do is often more effective than telling it what to do.
Force a length constraint
Summarize in exactly 5 bullets. Each bullet must be 12 words or fewer. No bullet may start with 'The' or 'It'.
Why it works: Hard length limits force prioritization. Banning common openers forces variety.
Specify a tone + a banned tone
Tone: direct and warm (like a brilliant colleague who respects your time). Banned: corporate, hype-marketing, 'I hope this message finds you well', emojis, exclamation marks.
Why it works: Positive + negative tone constraints produce more consistent voice than either alone.
Output as JSON for downstream use
Return ONLY valid JSON matching this schema: { 'title': string, 'bullets': string[], 'confidence': 'low' | 'med' | 'high' }. No prose before or after.
Why it works: Structured output lets you pipe the response directly into code without regex parsing.
Ask for N options, not one answer
Give me 5 meaningfully different variants of this subject line. Each must use a different hook strategy (contrarian, numeric, story, question, curiosity gap). Subject: [PASTE]
Why it works: Asking for one 'best' answer gets a safe middle answer. Asking for diverse variants surfaces the range.