How to brainstorm product ideas with AI
Good AI brainstorming starts with a sharp problem statement, not a solution. Write one sentence on who has the problem, how often it hurts and what they do today instead. Then ask the panel for several different approaches before you evaluate any of them, so the first idea does not anchor the discussion.
Once you have options, switch from generating to critiquing. A debate works best with personas that have conflicting incentives: a customer who cares about value, an engineer who cares about cost and complexity, and a skeptic whose only job is to find the flaw.
Questions every product idea should survive
Who exactly is the first customer, and would they pay this month? What are they using now, and why is switching worth the effort? What is the smallest version that proves demand? What happens when a bigger competitor copies the feature? Which single assumption, if wrong, makes everything else irrelevant?
Ask the panel to answer these explicitly. The verdict on Debatly highlights the strongest argument for and against, which usually maps directly to your riskiest assumption and the experiment you should run first.
Mistakes to avoid when using AI for product decisions
Do not treat the panel as market research. AI personas are excellent at surfacing risks and blind spots, but only real users can confirm demand. Use the debate to decide what to test, then go test it. Also avoid vague prompts: pasting your actual landing page copy, pricing table or feature spec produces far more useful critique than a one-line idea. For recurring work, build custom personas such as your ideal customer or your most demanding power user and reuse them across debates.
A simple AI brainstorming workflow
Step one: generate five to ten directions in a quick chat or group room. Step two: pick the two most promising and send each to a debate with a customer, an engineer and a skeptic. Step three: compare the verdict scores for upside, risk, difficulty and confidence, then ask follow-up questions about the winner. Step four: turn the next actions into a one-week validation plan with interviews, a landing page test or a prototype.