Ask for Options, Not Answers

Beginner

Five drafts to choose from beat one draft to fix

01 The Concept

When you ask for the answer, you get the model's single most average take. When you ask for several genuinely different options, you get a spread — and your judgment does what it's best at: picking and combining.

The trick is forcing real variety. "Give me 5 headlines" produces five near-clones. "Give me 5 headlines using 5 different angles: urgency, curiosity, social proof, data, contrarian" produces five actually-different candidates. Name the axes of variation, then pick a winner and iterate on it alone.

02 Weak vs. Strong

EX 01Naming a feature

Suggest 6 names for a feature that continuously saves work in our design tool (like Figma). Vary them across these axes, one name per row, in a table: 1. Literal/descriptive 2. Metaphor from the physical world 3. One invented word 4. Playful 5. Enterprise-serious 6. Verb-first action name For each: the name, why it works, and one risk (confusing? trademark-y? too cute?). Audience: professional designers. Avoid anything with 'smart' or 'AI' in it.

→ Why it works

Six forced-different angles, a table format, per-option rationale and risk, plus a banned-word constraint. Selection beats generation.

EX 02Making a decision with tradeoffs

We're a 5-person B2B startup selling to HR managers; goal is 200 qualified leads this quarter with a $3k budget and one part-time marketer. Give me 3 lead-generation options (may include webinar, ebook, or anything better). For each: effort in person-days, realistic lead range, time-to-first-lead, and — most important — **the scenario where this option fails**. Then recommend one and say what evidence would change your recommendation.

→ Why it works

Options with explicit failure scenarios and a falsifiable recommendation turn the model into a decision aid instead of a cheerleader for whatever you mentioned first.

03 Key Points

01One answer = the statistical average; options expose the range

02Force variety by naming the angles: tone, audience, length, strategy

03Ask the model to label each option with its approach so you can say 'more like #3'

04Pick-then-refine: choose one winner and iterate on it, don't fix all five

05For decisions, ask for options with tradeoffs — 'give 3 approaches and when each one loses'

04 Model-Specific Notes

Claude follows 'vary across these named axes' instructions faithfully and is good at honest per-option risks. If options still feel samey, add 'options 5 and 6 should be ones a cautious person would never suggest.'

05 For Your Role

Planning a birthday message, a trip itinerary, or dinner? Ask for 5 options across different vibes and pick. You'll find the one you didn't know you wanted.

Next: What Not to Paste: Privacy Basics →