How to brainstorm with AI without losing your own ideas

Ask an AI to "brainstorm" about your idea and you usually get a long, confident list. Some of it is useful. Most of it is generic, and after a few rounds it's hard to remember which parts were yours. The idea gets bigger and blurrier instead of clearer.

That's not a reason to avoid AI for brainstorming. It's a reason to use it differently: as someone who reads your notes closely and pushes back, not as someone who writes the ideas for you.

Start with your own dump, not a prompt

Before you ask anything, write down what's in your head. Ten minutes, no structure, one line per thought: facts, worries, half-ideas, names, numbers, things you've decided and things you haven't.

This matters for two reasons. First, the AI can only work with what you give it, and a one-line prompt ("ideas for a community garden") gets you a one-size-fits-all answer. Second, the dump is your record of what you actually thought before anyone else weighed in. Everything later can be checked against it.

Ask it to reflect before it adds

The most useful thing an AI can do with a brain dump is not to add ideas. It's to show you what you already said, more clearly than you said it. Ask for these, in this order:

1. The idea in one line. If it can't be said in one line, that's worth knowing.
2. What it rests on. The assumptions your notes take for granted. "The city will renew the lease if we maintain it" is an assumption, not a fact.
3. How the parts connect. Which pieces depend on each other. Often one unresolved question (who gets plots?) is blocking three other things.
4. Tensions, risks and gaps. Where two of your plans pull against each other, what could go wrong, and what the idea needs that you haven't mentioned.

Only after that is it worth asking for new ideas.

Make it mark what it added

This is the step most people skip, and it's the one that keeps the ideas yours. Ask the AI to keep two things visibly apart:

• what you said, in your words
• what it's suggesting, clearly labelled as a suggestion

Without that line, a suggested risk reads like something you already knew, an inferred deadline reads like a decision, and a guess about your budget reads like a fact. A week later you can't tell the difference, and neither can anyone you share it with.

A simple rule to give it: never present an addition as something I said, and never invent facts about my situation. If you need a fact, ask me.

Ask for adjacent ideas, not more ideas

"Give me more ideas" produces volume. "What's adjacent to this that I haven't considered?" produces angles: a different way to fund it, a smaller first version, a group of people who'd care that you didn't mention, an existing thing you could join instead of starting from scratch.

Ask for a few, not twenty. You're looking for the one that changes how you see the idea, and that's easier to spot in a short list.

End with questions, not conclusions

A good brainstorm ends with better questions than it started with. Ask the AI for a short list of questions you should answer next, and hold it to two standards:

• Specific. "Who is your audience?" fits any idea. "Does the school want two beds for the whole season, or just the spring term?" fits yours.
• Decision-changing. If the answer wouldn't change what you do, it's not worth asking now.

Then answer them yourself. Your answers are new material in your own words, and they're often where the real idea shows up. Feed them back in and go round again.

Know when to stop exploring

Brainstorming is for when you don't know what you think yet. Once you do, switch modes: the job changes from exploring to explaining. That usually means a short page for someone else, like a brief, a plan or a recap, with the open questions narrowed down to the ones only they can answer. Our guide on turning meeting notes into action items covers one version of that.

A quick checklist

• Dump first, prompt second.
• Ask for reflection (one line, assumptions, connections, tensions) before additions.
• Keep your words and its suggestions visibly apart.
• Ask for a few adjacent angles, not a long list.
• End with specific questions, and answer them yourself.
• Switch from exploring to explaining once you know what you think.