When using AI for team brainstorming, review the range of approaches as well as the appeal of each suggestion. A shortlist of polished ideas can still leave the team choosing between several versions of the same solution. Try collecting independent ideas first, introducing AI suggestions in a separate round, and preserving the original contributions while comparing the combined set.
That sequence is an editorial proposal, not a proven best method for every team. The practical goal is to make it visible when AI adds a different approach, improves an existing one, or merely gives it a better name. This guide offers a fictional workplace example and a review process; we have not run a brainstorming experiment or independently tested a model for this article.
Research distinguishes a stronger idea from a broader collection
In their 2024 Science Advances study, Anil Doshi and Oliver Hauser studied people writing short stories with or without access to AI-generated story ideas. AI-assisted stories received stronger evaluations, while the collection became more similar. The study involved nonprofessional writers and a constrained writing task. It does not establish how a management workshop or product team will perform with current tools.
A different design produced a different diversity result. Ashkinaze and colleagues’ dynamic experiment, in its July 2025 revision, asked participants to suggest unusual uses for everyday objects after seeing examples. Earlier participants’ ideas became examples for later participants. High exposure to AI examples increased collective diversity without improving individual creativity. The study used ChatGPT-3.5 examples, a nonrepresentative sample and task-specific measures, including an automated creativity classifier.
These findings should not be averaged into a universal claim that AI either expands or narrows thinking. The tasks, exposure and measures differ. They do establish why it is useful to ask two questions: is this idea worth developing, and does the collection offer meaningfully different ways forward?
Neither study proves the proposed independent-first sequence below. Treat it as a process to evaluate on your own recurring decisions, with its costs and failures visible.
Define the decision before asking for ideas
A useful brief names the problem, the people affected, the constraints and the decision the team can actually make. “Improve employee experience” leaves too much room for attractive but unrelated suggestions. “Reduce delays between approved equipment requests and employees receiving usable equipment” gives the group something more concrete to examine.
For this fictional equipment process, assume the team can change internal coordination and stock handling but cannot add headcount or replace its procurement system during the trial. These are invented constraints, not findings about an organization. The session should generate options within them and identify which assumptions need checking.
Avoid embedding a preferred solution in the brief. “How could we build a better request dashboard?” has already selected visibility as the intervention. If the underlying delay comes from unavailable equipment, a dashboard may explain the wait without shortening it.
Agree on evaluation criteria before viewing polished proposals. In this example, the team might consider expected effect on delay, implementation effort, operational burden and evidence needed before committing. Do not treat fluent AI explanations as evidence that an option will meet those criteria.
Preserve independent contributions before the AI round
Ask participants to record an initial approach privately, including what they think causes the delay and how their proposal would change it. Keep the original wording and assumptions. People who process orders, approve requests and receive equipment may notice different parts of the problem.
Then introduce the AI round using the approved brief and appropriate, nonsensitive context. Ask for approaches that address different causes, and require each suggestion to explain its mechanism and a condition under which it would fail. This makes comparison easier than requesting a long list of catchy initiatives.
Ethan Mollick’s 2023 practitioner essay, Automating creativity, argues for using AI in idea generation while acknowledging similarities among its ideas and limits of creativity tests. It offers practical prompting suggestions, but it is historical commentary, not evidence that a particular prompt guarantees diverse business options. We use that perspective alongside the original papers rather than treating it as another independent experiment.
Retain a simple record of where an option first appeared: participant contribution, AI suggestion or later combination. This helps the facilitator see what the AI added. It is not an attribution contest or a reason to favor one origin over another when selecting an operational trial.
Group options by the change they would make
The following invented shortlist illustrates the difference between surface variation and a different approach. No option has been tested, and none is a recommendation for an actual equipment operation.
| Proposed option | How it is meant to reduce delay | Distinction to inspect |
|---|---|---|
| Shared request-status dashboard | Exposes stalled requests so someone can intervene | Visibility and follow-up |
| Automated status digest | Alerts the team to stalled requests | Similar underlying approach to the dashboard |
| Small reserve of standard equipment | Avoids waiting for a new purchase in eligible cases | Availability, with storage and stock costs |
| Defined substitute-equipment policy | Allows an approved alternative when the preferred item is unavailable | Flexibility, with suitability constraints |
The first two could be useful alternatives, but they should not be counted as two unrelated ways to address the delay. The reserve and substitution options target different bottlenecks and introduce different risks. They broaden the choice only if those mechanisms fit the actual process.
Ask participants to explain the grouping in ordinary language. If two options change the same step for the same reason, record their differences as variants. Keep a minority proposal visible when it rests on a distinct, plausible explanation, even if it is less polished than the most popular suggestion.
AI can help organize the list, but let participants challenge its categories. A summary can merge two ideas that share vocabulary while hiding an important operational difference. Preserve the original contributions so the team can recover that difference.
Use AI to expose gaps before ranking the shortlist
Once the group sees the mechanisms represented, ask what is missing. In the equipment example, the list might assume every approved request is correctly specified. A different line of inquiry would examine whether missing compatibility information causes rework before fulfillment.
A useful follow-up request is: “These options address visibility, availability and substitution. Identify another plausible cause of delay, explain what evidence would distinguish it, and propose a reversible response.” The output remains a hypothesis. The model has not observed the workplace and cannot establish the cause by naming it confidently.
Challenge the favored proposal as well. Ask which conditions would make it ineffective, who would maintain it and what extra work it creates. Keep novelty and feasibility separate: an unusual idea may be impractical, while a familiar option may be the correct choice. Diversity is useful for exploring the decision, not a requirement to select the strangest answer.
Judge the session by what survives investigation
Choose a small number of options for evidence gathering or a reversible trial. Name the assumption, the responsible person, the expected observation and the stopping condition. For a reserve-stock proposal, the first investigation could be whether a sufficiently consistent set of equipment requests exists to make a reserve useful. That question should precede buying stock.
Compare the process across similar decisions before standardizing it. Record preparation and review effort, distinct approaches retained, assumptions checked and options that survive feasibility review. Keep the task context: a naming session and an operational redesign need different assessments. A simple category count is a local aid, not an objective creativity score.
Our workplace AI evaluation guide explains why accepted outcomes and review effort matter. For brainstorming, preserve the original pool and examine what AI contributes to it. Continue using the sequence where it produces useful choices at a reasonable cost; adjust it when it creates redundant options or extra facilitation without improving the decision.