ChatGPT for Brainstorming: Explore Ideas and Choose a Test

Use ChatGPT for brainstorming by describing a specific problem, requesting several different approaches, and comparing those approaches against criteria you set. Then choose a small test yourself. The useful result is a set of options you can explain and investigate, rather than a long list that sounds creative but changes nothing.

A tool like fenbs keeps your chosen direction, rejected options, and next experiment outside any single chat. That gives a later assistant the reasons behind the choice. The prompts below also work as instructions for other assistants; you do not need a special brainstorming feature.

How do you start a useful brainstorming session?

1. Write the problem before requesting ideas. Identify who experiences it, what happens today, and what a better outcome would look like. Keep your preferred solution out of this first description so alternatives have room to emerge.
2. State the constraints that change the answer. Include the people available, the deadline, the resources you can use, and anything you cannot change. Tell ChatGPT which constraints are firm and which you might reconsider.
3. Ask for distinct approaches rather than variations in wording. Request one simple option, one that changes the process, and one that questions your starting assumption. Require a sentence explaining the difference between each approach.
4. Choose the comparison criteria before ranking the results. Check the assumptions behind each option, select a manageable shortlist, and decide what evidence would justify trying one.

OpenAI’s guide to using ChatGPT↗ (opens in a new tab) includes brainstorming and comparing options among the jobs for a back-and-forth chat. You can steer that conversation as you learn. If the first answer misses a constraint, correct the constraint before requesting more ideas.

What does a strong brainstorming prompt look like?

Say a neighborhood repair workshop wants more people to bring small appliances to an event. “Give me marketing ideas” leaves the important questions unanswered. A better instruction states who the event serves, what the organizers have already tried, and what they can realistically run.

Problem: [who needs what, and what happens now].
Desired outcome: [an observable change].
Firm constraints: [time, people, resources].
What we tried: [attempts and what we observed].
Suggest six genuinely different approaches.
For each, explain the idea, its biggest assumption, and a small way to test it.
Do not invent audience research or treat your suggestions as approved.

Add an example only when it helps explain the quality you want. An example can also pull every answer toward the same style, so ask for options that deliberately depart from it. OpenAI’s prompting documentation↗ (opens in a new tab) supports giving the goal, relevant context, output, and boundaries rather than relying on a rigid formula.

If you want to compare assistants for the work, use the same problem and criteria in each. The broader differences belong in Claude vs ChatGPT; this exercise works by making your instructions and judgment explicit.

How do you get beyond obvious ideas?

Change one dimension at a time. Ask what the workshop could try without advertising, what would help a first-time attendee, or what an organizer could do with an existing community partner. A new constraint can reveal another direction more clearly than asking for increasingly “creative” ideas.

Separate generation from evaluation. Collect possible approaches before arguing about which deserves a place on the shortlist. When you do evaluate, ask for the strongest objection to each idea, including the one you already like. This gives you a way to challenge the list without asking ChatGPT to flatter your preference.

• Reverse the question: what would make the experience harder, and what does reversing that suggest?
• Change the audience: what would a newcomer need that a returning participant would not?
• Reduce the scale: what could one organizer try at the next event without changing the whole program?
• Combine selectively: which two ideas solve different parts of the problem without adding conflicting requirements?

These are ways to structure your own thinking, not evidence that a suggested approach will work. Bring observations from real conversations, prior events, or a small trial into the next round.

How should you compare a shortlist fairly?

Use criteria you can explain to another person. Avoid treating a generated score as a measurement. Ask what supports the score and what remains unknown; if the answer is only an assumption, label it that way. A simple comparison often exposes a missing fact sooner than another round of ranking.

For the hypothetical workshop, an attractive poster and a partner invitation might depend on different assumptions. The first depends on people seeing and understanding the message; the second depends on a partner reaching the right audience. Compare those assumptions before choosing which one to test first.

In fenbs Decisions and rules, a person can record the selected approach and its reason while leaving unresolved choices open. Assistants can record what a person decided; they cannot decide or sign off.

How do you turn an idea into a small experiment?

Write the experiment as an action with an observation attached. “Try a partner invitation” is incomplete. “Ask one existing partner to share an invitation, then ask attendees how they heard about the event” tells someone what to do and what to look for. Choose a result you can actually observe.

Set the review point before you start. Record what would make you continue, change the approach, or stop. A quiet trial can still teach you something if the question was clear; a busy one can remain inconclusive if nobody collected the relevant observation.

Keep rejected ideas with a short reason when that reason will matter later. “Needs equipment we do not have” may change next season, while “does not solve the stated problem” helps prevent another assistant from proposing the same detour. See other ChatGPT alternatives if you want to try the same exercise in another app.

How to do this with fenbs

1. Create a free account and add a project for the problem you are exploring. Describe the audience, desired outcome, and firm constraints.
2. Save the useful options and assumptions in an AI context note. Record a person’s choice in Decisions and rules, including the reason and conditions for revisiting it.
3. Add tasks for the first experiment and its review. Include the action, owner, observation to collect, and check that will determine whether the task is complete.
4. Connect an assistant using Integrations. MCP, a standard way for an AI app to use an outside tool, allows it to read and update the work you authorize.
5. Save Where we left off when you pause. After the experiment, record what happened and what to do differently as a lesson, so the next session starts with evidence.

fenbs is free for one person, with no credit card. It is not the right fit for a disposable idea list or as a substitute for audience research. Use it when the reasoning, decision, and follow-up need to survive the chat. Find more workflows in the AI at work library.