Better Brainstorming with AI: Techniques That Work

Published: July 8, 2025 · 6 min read

Most people brainstorm with AI the way they would ask a magic 8-ball: "give me good ideas for X." The results are generic because the prompt is generic. But AI is far better as a sparring partner — something to argue with, react against, and combine ideas with — than as an oracle that hands you brilliance on demand.

This guide gives you concrete techniques that consistently produce better raw material: assigning the AI a role, asking for bad ideas first, applying structured frameworks like SCAMPER, remixing and arguing, and capturing everything so nothing good slips away. Each one works with any mainstream chatbot.

Why "give me ideas" fails

A vague prompt gets you the statistical average of everything ever written about your topic — which is, by definition, unoriginal. "Ideas for a birthday party" returns balloons and cake because balloons and cake are the most common answers. The AI is not being lazy; it is answering exactly the question you asked.

Specificity in, specificity out. The more context you give — who the idea is for, what constraints apply, what you have already tried, what tone you want — the further the output moves from the generic center. Before prompting, spend two minutes writing down the constraints yourself: budget, audience, timeline, things to avoid. Those constraints are the raw material the AI needs.

It also helps to think alone for a few minutes before opening the chatbot. Even five minutes of solo thinking gives you a baseline to react against, which makes you a sharper judge of the AI's suggestions instead of a passive recipient.

Give the AI a role

Asking the AI to answer from a specific perspective changes which ideas surface. "Suggest marketing ideas" gives you the average; "suggest marketing ideas as a skeptical CFO who hates spending money" gives you scrappy, low-cost angles; "as a first-time customer who has never heard of us" gives you clarity-focused ones. The role acts as a filter on the idea space.

Useful roles to try:

• The expert: "As a veteran event planner, what would you do?" — pulls in domain conventions.
• The skeptic: "Poke holes in these three ideas" — stress-tests before you commit.
• The customer: "As a busy parent, which of these would you actually choose?" — grounds ideas in real preferences.
• The contrarian: "What would everyone in this industry say is a terrible idea, and why might they be wrong?" — surfaces unfashionable angles.

Run the same question through two or three different roles and compare. The differences between the answers are often more interesting than any single answer.

Start with ten bad ideas

This is the single highest-leverage trick in the list. Ask for ten deliberately bad, impractical, or absurd ideas for your problem. It works for two reasons: it removes the pressure to be brilliant, and bad ideas are often good ideas wearing a disguise — strip away the absurd part and a workable core remains.

Then ask the AI to salvage them: "Which of these bad ideas contains a usable insight? Rewrite the three most promising as serious proposals." You will be surprised how often the exercise produces something the straightforward approach missed. The bad-ideas round also trains you to stop self-censoring, which is half of brainstorming.

A variation: ask for ideas that would definitely fail, then invert them. "Ten ways to guarantee this product launch flops" is easier to answer than "ten ways to succeed," and flipping each failure gives you a checklist of things to get right.

Use structured frameworks like SCAMPER

Unstructured brainstorming drifts; frameworks force coverage. SCAMPER is a classic checklist for generating variations on an existing idea — run your starting concept through each lens:

• Substitute: what could replace a component, material, or step?
• Combine: what could you merge this with?
• Adapt: what else is like this, and what could you borrow?
• Modify: what if it were bigger, smaller, faster, slower?
• Put to another use: who else could use this, and for what?
• Eliminate: what could you remove entirely?
• Reverse: what if you did the opposite?

Ask the AI to generate two or three options per lens. You will end up with a grid of variations instead of a single thread, which makes the next step — picking — much easier. Other useful constraints: a tight budget ("under $100"), a format limit ("explain it in a tweet"), or a time box ("ideas we could test this weekend"). Constraints breed creativity because they eliminate the obvious.

Remix, argue, and iterate

The first batch of ideas is raw material, not the answer. The real value comes in the second and third rounds: ask the AI to combine idea 3 and idea 7, to argue against your favorite, or to steelman the one you almost dismissed. "What am I missing?" and "what would make this fail?" are the two most productive follow-ups in brainstorming.

Iterate in rounds with a clear purpose per round: round one for quantity, round two for combinations, round three for criticism. Tell the AI which round you are in — "now be brutal, find the flaws" gets you a very different response than "give me more options." Most people stop after round one and conclude AI brainstorming is shallow; the depth is in the follow-ups.

Capture everything, judge later

Generating ideas and evaluating them are different mental jobs, and doing both at once ruins both. Keep one running document — a note, a doc, a whiteboard — where every idea lands without judgment during the session. Tag the promising ones, star the wild ones, but do not delete anything yet.

Then step away. Come back later — an hour, a day — and evaluate with fresh eyes: which ideas survive contact with your actual constraints? Force-rank the survivors and pick no more than three to develop. A brainstorming session that ends with thirty unranked ideas ends with nothing; one that ends with three ranked contenders ends with a plan.