How to brainstorm with AI effectively
Give the AI a goal, an audience and constraints. "Ideas to get 50 new B2B trial sign-ups a month with no paid ads" produces far better output than "marketing ideas." State what you have already tried so the panel does not repeat it.
Separate the two phases of brainstorming: diverge first to get many options, then converge by testing them against your constraints. A panel does both, because some personas generate while others critique.
Who to put on your brainstorm panel
A reliable mix is a visionary for bold ideas, a customer voice for reality checks, a skeptic for risks and an operator for feasibility. For specialized topics, add a custom character with the right expertise, such as a growth marketer, a UX researcher or a CFO.
Keep the panel small. Three or four voices produce focused, lively exchanges; more voices tend to repeat each other. If you want a second angle, run another session with a different mix instead.
Brainstorming techniques that work well with AI
Try "worst possible idea" first, then invert the bad ideas into good ones. Use SCAMPER prompts (substitute, combine, adapt, modify, put to another use, eliminate, reverse) as follow-up questions. Ask for constraints that force creativity, like "only ideas that cost under 500 dollars."
After the verdict, ask the panel to develop the winning idea further or to argue why the runner-up might be better. Another useful follow-up is "What would a competitor do if we launched this?" because it exposes weaknesses the first round missed.
Mistakes to avoid
Do not accept the first list. Do not judge ideas before you have enough of them. And do not skip the critique step: an idea that survives a skeptic is worth far more than ten that were never challenged. Finally, pick one idea and test it quickly in the real world; a brainstorm only pays off when something ships.