Language models are good brainstorming partners, but default outputs tend towards the obvious. A few techniques produce better ideas.
Give Rich Context
Describe the problem, audience, constraints, what's been tried and what "good" looks like. Generic prompts yield generic ideas.
Ask for Quantity and Variety
"Give 20 ideas across different approaches: low-cost, high-effort, unconventional, digital, in-person." Explicitly ask for ideas that differ from each other.
Push Past the Obvious
- "Exclude the five most common ideas and give me others."
- "What would a competitor in a different industry try?"
- "What would we do with ten times the budget? With none?"
Separate Generation From Evaluation
First generate broadly, then in a separate step evaluate ideas against criteria: cost, impact, feasibility, risk.
Combine and Develop
Ask the model to merge promising ideas or develop one into a concrete plan with steps and risks.
Use Higher Temperature
Where you can adjust sampling, a higher temperature increases variety for brainstorming.
Keep Human Judgement
Models don't know your organisation's politics, capabilities or history. Treat output as raw material for discussion, not decisions.