How do you brainstorm with ChatGPT?

Brainstorming is a method for producing many ideas quickly, then sorting them later. The word was made popular by Alex Osborn, an advertising executive who started working on group idea sessions in 1939 and described the method in his 1953 book Applied Imagination. His approach is usually summed up in four rules: aim for lots of ideas, hold criticism until later, welcome odd ideas, and build on each other's ideas by combining them.

ChatGPT fits those rules surprisingly well. It never runs out of energy, it does not mind being asked for "twenty more," and it has no ego to bruise when you throw half its list away. The trick is to keep the rounds separate, exactly as Osborn advised for human groups: generate first, judge later.

Before you start: one clear question

Osborn insisted that a session should tackle one narrow, clearly stated problem, and he found that sessions juggling several questions at once worked poorly. The same holds with a chatbot. "Ideas for our town" is too wide. "A free Saturday event in October that gets families and seniors in the same room" gives the model, and you, something to aim at.

Give ChatGPT context too: who you are, who the audience is, and what you have to work with. A detailed request gets a far more tailored answer than a bare one.

A worked session: a community event

Imagine a volunteer committee at a small-town public library planning an autumn event.

Round 1: quantity, no judging.

You are an experienced community event organizer. Our small-town library wants one free Saturday event in October that brings families with young children and seniors together. Give me 25 ideas, one line each. Include at least five unusual ones. Don't evaluate them yet.

Expect a mix of familiar ideas (a story hour, a craft fair) and a few surprises. Resist the urge to comment. If you are using the API or the OpenAI Playground, this is the round to raise the temperature setting, which makes the model pick less likely words and so vary its output more; in the regular ChatGPT app, asking for "unusual" or "unexpected" ideas does a similar job.

Round 2: add the constraints.

Now our limits: a budget of $400, six volunteers, the library's community room (40 people) and the lawn outside, and a 3-hour window. Mark which of your 25 ideas fit these limits, and give me 10 new ideas designed for them.

Constraints are where the model starts to be practical, and where your local knowledge begins to count.

Round 3: combine and improve.

Take ideas 3, 11 and 19 from your lists and combine them into three different event concepts. For each, describe the first 15 minutes from the point of view of a seven-year-old and of an 80-year-old.

Asking for two viewpoints exposes weak spots quickly: an activity that delights children may bore or exhaust older guests.

Round 4: critique and rank.

Act as a skeptical committee member. For each of the three concepts, list the two biggest risks and one way to reduce each. Then rank the concepts for our goal of mixing generations, and explain the ranking in one sentence each.

An illustrative outcome: a "story swap" afternoon where seniors tell what school was like in their childhood while children draw the scenes, ending with a small exhibit on the library wall. Cheap, fits the room, and it puts the two groups at the same table.

How do you keep your ideas original?

Start with your own list, before you open ChatGPT. Research on group brainstorming suggests why this matters: in a review of 22 studies, psychologists Michael Diehl and Wolfgang Stroebe found that people working on their own, with their ideas pooled afterward, usually came up with more ideas than the same people brainstorming together. A later review argued that many of those studies ran the method poorly, but the practical lesson holds. Hearing other suggestions first can crowd out your own, and a long, confident list from a chatbot can have the same effect, so put your ideas on paper first.

Next, feed in what only you know. A language model produces likely text based on patterns in what it has read, so its first suggestions often sound like many other events you have seen. Your town's history, the retired teacher who loves to tell stories, the empty lot behind the arena: details like these turn a generic idea into one that belongs to your community. Mention them, and ask the model to build on them.

Finally, treat ChatGPT's list as raw material, not a menu. Use it to push past your first obvious answers, borrow a piece of one idea and a piece of another, and keep asking "what would make this more like us?" The final choice, and the reasons behind it, should be yours.