AI Brainstorming Workshop Agenda for a 75-Minute Session

AI in facilitationMethods

A 75-minute AI brainstorming workshop agenda: brainwriting first, an AI pass to widen the pool, a criteria-based shortlist, then clustering and a vote.

•••
11 min read

Quick answer

An AI brainstorming workshop for 8 people fits into 75 minutes: 30 minutes of silent brainwriting, 10 for each person to widen the pool with AI, 5 to have the AI cut its output to a shortlist against the group's criteria, 5 of buffer, 15 for an affinity map and 10 for dot voting. Humans first is a design choice; studies on AI and idea diversity point both ways, and none observed a facilitated workshop.

Key takeaways

  • A 2018 peer-reviewed review in Frontiers in Psychology reports that groups of four brainstorming aloud tend to generate only about half as many ideas as the same number of individuals brainstorming alone.
  • Findings on AI and idea diversity point both ways: a 2024 Science Advances experiment on short-story writing found AI-enabled stories more similar to each other, while an experiment with more than 800 participants in more than 40 countries found that high exposure to AI ideas increased collective idea diversity.
  • Humans first is not shown to be the best order: Atlassian's Teamwork Lab had 15 HR business partners brainstorm human-first and AI-first, both groups scored roughly the same for creativity after 20 minutes, and 12 of 12 participants felt more ownership over human-first ideas.
  • A 75-minute block for 8 people adds up as 30 minutes of brainwriting, 10 to widen the pool with AI, 5 to narrow it, 5 of buffer, 15 for an affinity map and 10 for dot voting.
  • The method library lists brainwriting at 30 to 45 minutes for 5 to 15 people, the affinity map at 10 to 90 minutes and dot voting at 5 to 20 minutes.

An AI brainstorming workshop for 8 people fits into 75 minutes. People write their own ideas in silence and build on each other's cards. Each person then asks an assistant for options the group lacks and has it cut its output to a shortlist against criteria the group agreed on. The group clusters what is left and votes.

Putting humans first is a design choice. Research on group creativity and on AI-assisted ideation gives reasons for it and sets limits, and none of the cited studies tested this agenda. Basics of an unassisted session are in how to run a brainstorming session.

1. Open with a silent written round

Start with silent writing on cards. A peer-reviewed review of collaborative ideation by Paulus, Baruah and Kenworthy (Frontiers in Psychology, 2018) reports, from studies by Diehl and Stroebe (1987) and Mullen and colleagues (1991), that groups of four brainstorming aloud "tend to generate only about half as many ideas as the same number of individuals brainstorming alone". That is a count of ideas and says nothing about their quality.

The same review names the cause: in face-to-face groups only one person can share ideas at a time, which it calls production blocking. Writing ideas down limits that blocking and reduces evaluation apprehension, because people are less aware of whose idea is whose (review).

Brainwriting is the method for this round. In the Gamestorming write-up that the method library credits, players silently write ideas on index cards, then pass the cards on and build on them.

AI built for workshop planning
Describe your goal and get an agenda draft built from facilitation methods. You stay in charge of every block.
7 days free. No credit card.
Try it free

2. Keep the chatbot closed until the cards are written

No assistant opens during the written round. Three lines of evidence bear on that choice.

Fixation on what is shown first. In Kohn's dissertation at Texas A&M University (2009), exchanging ideas in a group led members to explore fewer domains of ideas, and people conformed their ideas to those suggested by others. A systematic review of 53 experimental studies on fixation (Wang, Okada and Takagi, Frontiers in Education, 2023) describes a 1991 experiment by Jansson and Smith in which participants copied elements of example solutions, even flawed ones, and the copying impeded creativity. Neither involved AI output; that an assistant's list anchors a room the same way is this article's inference.

Similarity of AI-assisted ideas. Findings on idea diversity point both ways.

  • Doshi and Hauser (Science Advances, 2024) ran an online experiment on short-story writing: stories by writers who obtained ideas from an LLM were rated more creative and were more similar to each other than stories by humans alone.
  • Anderson, Shah and Kreminski (2024), in a 36-participant user study accepted to the ACM Creativity & Cognition conference, found that different users tended to produce less semantically distinct ideas with ChatGPT than with an alternative creativity support tool.
  • Ashkinaze and colleagues found the reverse in an experiment with more than 800 participants in more than 40 countries, accepted at ACM Collective Intelligence 2025: high exposure to AI ideas before brainstorming did not affect the creativity of individual ideas and increased collective idea diversity.

None of the three observed a facilitated group session.

The order itself. No source cited here shows that humans first is the best order.

  • Atlassian's Teamwork Lab reports a study of its own, company research reported on its own blog, with 15 of the company's HR business partners in two 20-minute sessions: some worked without AI for the first half, the rest used AI from the start. By minute 20, after both groups polished with AI, creativity scores were roughly equal, and 12 of 12 participants felt more ownership over the human-first ideas (same page).
  • Paulus and colleagues (Human Factors, 2015) studied brainwriting without AI in a high-technology company, and in their first study participants who generated ideas first as a group and then as individuals performed best.
  • Kumar and colleagues, in two pre-registered experiments with 1,100 participants (CHI 2025 per the arXiv record), found that LLM assistance can boost creativity during assisted tasks and may hinder independent performance afterwards; AI came first in their design, so a human round followed by AI was not tested.

Atlassian's advice is to let the team work on the problem first, share seed ideas across teammates and then bring in AI to expand and stress-test them. On a 2024 Stanford Graduate School of Business podcast, Jeremy Utley, appearing with Kian Gohar, advised alternating "between individual personal thinking and thinking assisted and amplified by AI". The Paulus, Baruah and Kenworthy review is more cautious and says only that a mix of alone and group idea generation "may be ideal".

3. Use AI to widen the pool with a constrained prompt

Once the cards are written, each person takes their own ideas to an assistant and asks for options the group does not have yet. What the evidence says to expect:

  • In a larger Teamwork Lab study of 797 knowledge workers, also company research, participants using AI generated up to 43% more ideas in their first round than the control group without AI.
  • Lee and Chung (Nature Human Behaviour, 2024) ran five experiments with ChatGPT (GPT-3.5): ideas were more creative than with no technology or a web search, and the tool "was most effective at generating incrementally (versus radically) new ideas".
  • In a pre-registered field experiment with 776 professionals at Procter & Gamble, an NBER working paper from 2025, individuals with AI matched the performance of teams without AI; those teams were pairs.
  • Shaer and colleagues (2024) evaluated a group brainwriting framework with an LLM added, and their findings "suggest that integrating LLM in Brainwriting could enhance both the ideation process and its outcome".

How people prompt shapes the spread:

  • In the Meincke, Mollick and Terwiesch preprint (2024), idea pools generated by GPT-4 were less diverse than ideas from groups of human subjects, prompt engineering substantially improved the diversity, and Chain-of-Thought prompting came close to the human groups.
  • Ethan Mollick, in Automating creativity (2023), advises adding constraints because "AI works best to generate ideas when it is most constrained", and notes that the researchers he describes repeated their prompt several times "because generating a lot of ideas is useful".

A sample prompt:

Our team is working on this question: [question]. These are the ideas I already have: [paste your cards]. Work step by step. First, list 20 short idea titles that differ from my ideas and from each other. Second, check the list for near-duplicates and replace them. Third, describe each remaining idea in one sentence of at most 30 words.

The prompt is an illustration written for this article. Its count of 20 is this article's choice, since no study cited in this article tested a number per prompt. A group new to prompting can warm up by co-writing prompts in small groups.

4. Narrow before anything goes on the wall

Have each person's assistant cut its own output to a shortlist before the group sees it. Volume is the reason: in the 36-participant user study by Anderson, Shah and Kreminski, ChatGPT users generated a greater number of more detailed ideas. Say 8 people each ask for 20 options: that is 8 × 20 = 160 ideas after one prompt each.

The structure of diverging with AI, letting the AI narrow against criteria and converging as a group comes from Tanner Kohler's article for Nielsen Norman Group, and the method library describes it as Diverge-Narrow-Converge with AI for 3 to 30 people and 30 to 45 minutes.

Agree on the criteria as a group before the AI pass, so every shortlist is cut by the same rules; criteria and a scorecard for narrowing AI ideas covers how to choose them. Each person then reads the shortlist and may swap in an idea the AI dropped.

One caveat on timing: the minutes for the AI steps are a planning assumption, because no source cited here measured how long prompting and reading take. If the block runs late, cut the buffer first.

5. Converge with methods written for your group size

Cluster first, then prioritise, and pick both methods by the bounds in their write-ups.

Clustering. The library's Affinity Map write-up says it fits once a brainstorm has produced 20 or more notes, and advises against lists under roughly 15 items and runs under 10 minutes. The Gamestorming write-up it credits attributes the affinity diagram to Jiro Kawakita in the 1960s.

Prioritising. The library recommends Dot Voting when an ideation round has produced more options than the group can discuss in depth. Its write-up warns against binding or high-stakes decisions, groups smaller than four and public votes under strong hierarchy, and treats the dot count as a signal that still needs a feasibility check. The library's rule of thumb for votes is one third of the number of options, rounded down, so 12 options means 4 dots per person.

A 75-minute agenda that adds up

Picture an 8-person product team with 75 minutes and one question. The team is invented for this example; the durations and group sizes come from the 750+ method library on this site. The group has already agreed on three criteria.

Minutes Block What happens Library bounds
0-30 Brainwriting (30 min) Silent writing on cards, then building on them. No chatbots. 30-45 min, 5-15 people
30-40 AI widen (10 min) Each person prompts for 20 options that differ from their cards. n/a
40-45 AI narrow (5 min) The AI picks five per person against the three criteria; each person checks. n/a
45-50 Buffer (5 min) Notes go on the wall; people stretch. n/a
50-65 Affinity Map (15 min) The group sorts cards and shortlists into clusters and labels them. 10-90 min, 1-20 people
65-75 Dot Voting (10 min) Everyone votes on the clusters at the same time. 5-20 min, 2-100 people
Total: 75 min

The arithmetic: 30 + 10 + 5 + 5 + 15 + 10 = 75 minutes, and eight people sit inside every group-size range in the table.

On the wall are 8 × 5 = 40 AI notes plus the brainwriting cards. Fifteen minutes is short against the library's full scripted affinity map run of 50, so expect rough cluster labels.

To end on discussion, swap the vote for 1-2-4-All in two groups of four. Its original write-up on the Liberating Structures site, which credits Henri Lipmanowicz and Keith McCandless, starts as this agenda does: "Individuals silently reflect and write down ideas." The method library's record lists it from 4 people and from 12 minutes: 1 alone, 2 in pairs, 4 in fours and 5 with the whole group (1 + 2 + 4 + 5 = 12). Take 2 minutes from the affinity map: 30 + 10 + 5 + 5 + 13 + 12 = 75, with clustering still above its 10-minute minimum.

Conclusion

The evidence under this agenda is mixed and indirect. Humans first stays a design choice with three reasons: writing limits production blocking, the cards hold ideas no prompt has shaped, and ownership was higher in the one small company study that compared orders with AI.

For your next ideation block, write the question and three criteria on one page, copy the table into your plan and replace the group size and clock times with your own. Then check each duration against its library bounds.

References

Share:

Ready to plan your next workshop?

7 days free. No credit card. No setup fees. Cancel anytime.