AI for Brainstorming: How to Use AI as a Creative Partner

Using AI for brainstorming has become a default for a lot of people working on creative and strategic problems. It's fast, tireless, and produces output on demand. The problem is that most people use it in exactly the way that makes it least useful: asking it to generate ideas for them.

When you ask an AI to brainstorm on your behalf, you get a list of plausible-sounding ideas generated from training data that includes a lot of brainstorming outputs. The list looks complete. The ideas feel reasonable. Almost nothing on it will surprise you, and very little will be genuinely useful, because the AI is pattern-matching to what brainstorming outputs typically look like — not thinking through your specific problem.

This doesn't mean AI is useless for ideation. It means you have to use it differently.

The Problem with Generating Ideas Directly

The research on AI and creativity is still developing, but studies from 2023 and 2024 have started to quantify what practitioners have noticed anecdotally. A study by Doshi and Hauser (2024) found that large language model assistance improved average idea quality in a product innovation task — but reduced the variance of ideas across participants. People using AI converged on similar ideas. The average got better; the distribution got narrower.

That's a significant tradeoff. Most creative breakthroughs come from the tail of the distribution — the unusual ideas that most people wouldn't generate. If AI use consistently compresses that distribution, it may raise the floor while lowering the ceiling.

Divergent thinking research has long established that creative quality correlates with originality, not just fluency. Producing more ideas only helps if those ideas span a wide range of conceptual territory. AI-generated idea lists tend to be fluent but not diverse — they cover the obvious angles thoroughly and the non-obvious angles poorly.

The productive uses of AI in brainstorming are the ones that preserve or enhance the divergence, not the ones that replace human generation with AI generation.

5 Uses That Actually Work

1. Expand Your Constraints

Instead of asking AI to generate ideas, ask it to add constraints to your problem. "Give me 10 unusual constraints I could apply to this problem." Then brainstorm under each constraint yourself.

This works because constraints are a proven mechanism for forcing non-obvious thinking. Constraint-based ideation produces more original ideas than unconstrained ideation — the research from Patterson and colleagues on this is consistent — but most people don't naturally impose interesting constraints on themselves. They go with the constraints that are already obvious.

AI is good at generating diverse constraint variations because constraints are mostly formal and structural, not substantive. "What if it had to be free?", "What if it had to work without electricity?", "What if you had one hour instead of six months?" — these don't require deep domain understanding to generate. They require category knowledge about types of constraints, which AI handles well.

The ideas you generate under those constraints will be yours. The AI just expanded your ideation space.

2. Devil's Advocate Pressure-Testing

After you've generated a list of ideas through conventional brainstorming, ask the AI to argue against each one as forcefully as possible. "Act as a skeptic who thinks this idea will fail. Give me the strongest case against it."

This isn't asking AI to evaluate your ideas — it's using AI to make the evaluation more rigorous than you'd make it yourself. You'll dismiss weak attacks instinctively and notice strong ones. The ideas that survive serious AI devil's advocate pressure are more likely to be genuinely robust.

This use works because AI has read a lot of failure postmortems, critical analyses, and adversarial arguments. It can generate objections across more domains than you'll naturally consider. You're not outsourcing judgment; you're expanding the coverage of your self-criticism.

3. Cross-Domain Analogy Generation

Ask AI to describe how your problem is solved in a completely different domain, then translate the solution back.

"How do hospitals solve the problem of managing a large number of simultaneous requests with variable priority?" Then map that answer onto your actual problem. "How do airports manage the problem of routing people with different destinations through shared infrastructure?" Same thing.

Analogical reasoning is one of the most reliably effective mechanisms for creative insight. The difficulty is that humans naturally draw analogies from domains they already know, which limits the range. AI has domain coverage that far exceeds any individual's. Asking it to generate cross-domain parallels — and then doing the translation work yourself — leverages AI's breadth while keeping the generative work human.

Be specific about what type of solution you want. "How does supply chain management solve the problem of synchronizing multiple dependencies?" produces more useful output than "What are some analogies for my problem?"

4. Question Generation, Not Answer Generation

Ask AI to generate questions about your problem rather than solutions. "Give me 20 questions someone should ask before attempting to solve this problem." "What's the most important thing I don't know about this?"

This exploits a different AI capability. AI is good at recognizing what information is typically relevant to a problem type because it has seen many such problems. You may be missing obvious but important information. Having that information surfaced as a question, rather than as an answer, keeps you in the driver's seat of the ideation.

The Starbursting technique from traditional brainstorming does this manually — generating questions across Who, What, When, Where, Why, and How before attempting answers. AI can do this across a much wider range of question types.

5. Perspective Multiplication

Ask AI to generate perspectives you haven't considered. "How would a logistics manager approach this problem? How would a child? How would someone from 1950? How would someone who hated your current solution?"

Then use those perspectives as brainstorming prompts for yourself. You generate the ideas; the AI generated the vantage points.

This is an AI-augmented version of Rolestorming, a technique where you ideate from an assigned character's perspective to reduce evaluation apprehension and force perspective shifts. The AI expands the range of available perspectives beyond the ones you'd naturally choose.

3 Uses to Avoid

Asking for a complete list of ideas. The output will be comprehensive and forgettable. You'll feel like you've done ideation when you've actually just read a summary of what brainstorming on your topic typically looks like.

Using AI output as a starting list. Starting from AI's list anchors your thinking to it. You'll extend and modify the AI's ideas rather than generating your own. Anchoring effects are strong enough that this materially reduces the originality of what you produce. Generate first; use AI afterward.

Asking AI to evaluate which ideas are best. AI will give you an answer, but it's selecting from a training distribution of "ideas people thought were good," not reasoning about your specific context, constraints, and capabilities. You know things about your situation that the AI doesn't. Use AI for inputs and pressure-testing; keep the judgment.

How to Prompt for Better Creative Thinking

The difference between AI as a useful creative partner and AI as a shortcut that produces mediocre output is largely in the prompting.

Be specific about your constraint. "Brainstorm ideas for a new product" produces generic output. "I'm building a tool for tax accountants who hate using computers. Generate ten constraints I could apply to make the product simpler than anything else in the market" produces something you can work with.

Assign a role with genuine domain knowledge. "As an emergency room triage nurse" or "As a mechanical engineer who designs for reliability" activates more domain-specific output than generic prompting.

Ask for things AI is good at: coverage (have I missed a domain?), structure (what framework applies here?), constraint generation, perspective multiplication. Don't ask for things AI is bad at: your specific context, genuine originality, judgment about what's actually good.

Iterate on the interesting parts. When an AI output contains something unexpected, go deeper on that specific thread rather than treating the full list as the answer. The surprises are usually where the value is.

The underlying principle is that AI should expand the range of inputs to your creative thinking, not substitute for the thinking itself. The creative process involves divergent generation, convergent evaluation, and the specific judgment about what matters in your context. AI can accelerate the first; it's poorly suited for the last two.

Ready to train your creativity? Try science-backed exercises that measure and improve your creative thinking. Start a Free Exercise