In this guide
Start with the decision behind the brainstorm
Define what the ideas will help you decide: improve an onboarding process, reduce a recurring delay, create campaign directions or design a team session. Name the audience, timeframe and success measure. ‘Give me ideas’ produces volume; a decision produces relevance.
Confirm that brainstorming is the right activity. If the issue is missing evidence, unclear authority or an unresolved policy question, more ideas may distract from the real work.
Prepare permitted context and constraints
Provide the current process, what has already been tried, non-negotiable requirements, available capacity and known risks. Remove confidential, personal or regulated information unless your organisation has approved the system and use.
Constraints can improve usefulness. A team of three, a four-week window and no new software create a more realistic search space than an unlimited brief. Ask the system to state assumptions and mark missing information.
Generate widely before judging
Ask for distinct approaches rather than minor variations. Request ideas from different perspectives: customer, frontline employee, manager, process owner or sceptic. Delay scoring until the option set is broad enough to reveal alternatives.
OpenAI Academy recommends a wide-to-narrow flow that separates generation from evaluation. This reduces the risk of selecting the first fluent answer before exploring the problem.
Group ideas and expose duplicates
Ask the system to cluster ideas by underlying mechanism, not just wording. Label which options are quick experiments, longer-term changes or dependent on missing evidence. Merge duplicates and identify genuine differences.
Review the clusters yourself. AI may group ideas that sound similar while hiding an important operational distinction. Rename categories in the language your team uses.
Fictional prompt
We need ways to reduce late regional reports. Constraints: existing software only, a two-week test and no change to approval authority. Generate 15 distinct options from process, capability and communication perspectives. Do not evaluate yet. Then group them by mechanism and identify assumptions requiring validation.
Evaluate with explicit criteria
Choose criteria before ranking: expected value, effort, time, risk, reversibility, evidence and fit with strategy. Treat scores as prompts for discussion rather than objective truth. Ask what evidence would change the ranking.
Include the people who will do or experience the work. AI cannot see local incentives, relationships and constraints unless they are supplied—and some information should not be supplied. Human discussion remains essential.
Pressure-test the leading options
For each finalist, examine failure modes, affected stakeholders, dependencies and early warning signals. Ask for a constructive counterargument and a smaller test. Separate plausible concerns from claims that require evidence.
Do not let the system invent research, customer feedback or internal facts. Verify external claims with authoritative sources and discuss consequential risks with qualified people.
Turn one idea into a learning experiment
Choose an owner, boundary, measure and review date. Record why the idea was selected and what would cause the team to stop, adapt or expand it. A brainstorm creates value only when an option becomes a responsible action.
After the test, compare the result with the original assumptions. Save useful prompts and lessons, but revisit the context each time. A reusable method is more valuable than a permanent list of ideas.
Before you move on
- The brainstorm supports a defined decision.
- Context and constraints are useful, permitted and specific.
- Idea generation is separated from evaluation.
- Clusters represent genuinely different mechanisms.
- Criteria and missing evidence are visible before ranking.
- Relevant people examine risks and local constraints.
- One option becomes a bounded experiment with an owner and review.
Sources and further reading
These official resources informed the guardrails and practical method in this guide.
This preview uses fictional examples where indicated. Read our editorial approach.