90-Minute AI Divergent Thinking Agenda (Facilitator Guide)

As an Amazon Associate I earn from qualifying purchases. Product links on this page are affiliate links — they cost you nothing extra.

⏱ 22 min read

The 90-Minute AI Divergent Thinking Workshop Framework

The 90-minute AI divergent thinking workshop framework pairs unassisted human brainstorming with artificial intelligence expansion across four distinct phases: Framing (15 minutes), Human Divergence (20 minutes), AI-Augmented Expansion (35 minutes), and Synthesis (20 minutes). This sequence forces teams to establish their own baseline perspectives before using large language models to shatter domain assumptions and introduce non-obvious alternatives. By restricting algorithmic input to the middle of the session, the framework prevents premature consensus around familiar, average ideas.

Why separate human ideation from AI tools instead of opening ChatGPT or Anthropic’s Claude in minute one?

Divergent thinking is a cognitive process where participants generate a high volume of distinct, exploratory solutions to an unstructured challenge before any assessment or sorting begins. Traditional group brainstorming usually fails this objective. In research conducted at the Kellogg School of Management, Professor Leigh Thompson documented that in traditional six-person brainstorming sessions, two people account for more than 60% of all talk time. The remaining participants succumb to evaluation apprehension or social loafing. The group anchors on the first three plausible suggestions, creating a consensus trap that yields conventional outcomes.

[Phase 1: Framing]
15 Minutes: Problem Setup
       |
       v
[Phase 2: Human Divergence]
20 Minutes: Silent Solo Work
       |
       v
[Phase 3: AI Expansion]
35 Minutes: Prompt Provocation
       |
       v
[Phase 4: Synthesis]
20 Minutes: Clustering

When you introduce AI without guardrails, it amplifies this conformity. Standard prompts produce the statistical midpoint of the training data. If three product managers ask an AI for "innovative onboarding ideas," the system provides identical lists featuring progress bars, gamified badges, and interactive tooltips.

To overcome this, the facilitator must treat the model as an intentional provocateur rather than an oracle. In a 2023 study published by Christian Terwiesch and Karl Ulrich at the Wharton School of the University of Pennsylvania, researchers evaluated ideation performance and found that generative AI produced raw ideas 40 times faster than human cohorts. However, the absolute value lay in pushing prompt constraints into extreme territories, such as demanding concepts for zero-budget scenarios or cross-pollinating ideas from unrelated industries. This approach connects directly to proven divergent thinking methods that systematically disrupt status-quo thinking.

To execute this dynamic, the facilitator enforces one non-negotiable rule: zero evaluation during divergence. The human brain cannot generate novel connections while assessing execution risk. You must actively protect the boundary between generation and selection, a principle explored thoroughly in balancing divergent and convergent thinking in design. If a team member says "legal won’t approve that" during the 35-minute expansion window, stop the conversation immediately.

Run the schedule using a physical display clock placed where every participant can track it.

Understanding the mechanics of the role of divergent thinking in creative breakthroughs prepares you to manage the room’s energy through these transitions. The four-phase structure delivers between 80 and 120 raw concept fragments from an eight-person group before the final bell rings.

Now review the exact minute-by-minute facilitator script and prompt sequences required to run Phase 1.

Table of Contents

• The 90-Minute AI Divergent Thinking Workshop Framework
• Minute-by-Minute Facilitator Timer Schedule and Stage Objectives
• Prompt Archetypes That Force AI into Lateral Territory
• Troubleshooting Participant Friction and Common Workshop Pitfalls
• Complete Facilitator Script and Printable Timer Cheat Sheet
• Sources & Further Reading

Minute-by-Minute Facilitator Timer Schedule and Stage Objectives

A strict 90-minute AI divergent thinking workshop splits into four discrete, non-negotiable blocks that systematically decouple idea generation from evaluation. When facilitators mix concept creation with critical judgment, group output drops rapidly. By enforcing rigid time boxes, you prevent teams from prematurely settling on obvious solutions and force them to build cognitive volume before attempting any synthesis.

Divergent thinking is a cognitive process that explores multiple spontaneous, non-linear solutions to an open-ended problem. In this workshop framework, generative artificial intelligence functions not as an editor, but as a provocation engine to stretch team exploration far beyond familiar institutional patterns.

[00:00 - 00:15]
Perimeter & Orthodoxies
       |
       v
[00:15 - 00:35]
Analog Silent Run
       |
       v
[00:35 - 00:70]
AI Expansion Loops
       |
       v
[00:70 - 00:90]
Clustering Without Voting

Minutes 00–15: Setting the Problem Perimeter and Psychological Safety

The first 15 minutes establish operational boundaries, identify sacred cows, and remove the fear of peer evaluation. Spend the opening 4 minutes delivering a crisp problem statement centered on a single target metric or user friction point. If the team attempts to solve three problems at once, divergence collapses into confusion.

Dedicate the next 6 minutes to cataloging "forbidden orthodoxies." An orthodoxy is an unexamined operational assumption that teams treat as an immutable law, such as "customers will never pay upfront" or "deployments require manual QA approval." Have participants call these out and write them on a physical board under the header "Off-Limits Assumptions." Banning these baseline assumptions early unlocks genuine divergent thinking skills because it cuts off conventional fallback positions.

Use the final 5 minutes to set baseline ground rules for psychological safety. Citing Amy Edmondson’s foundational research on team psychological safety from Harvard Business School, explain that high-performing teams actively reward candid risk-taking while treating intermediate failures as information. State explicitly that rank does not grant veto power during ideation. To set expectations, compare this setup against structured tools like the 60-Minute Ideation Workshop Agenda (With Script), where strict operational safety protects early concept generation.

Minutes 15–35: Analog Silent Generation

Participants spend these 20 minutes capturing raw concepts individually on paper cards without touching a laptop, opening a browser, or speaking aloud. Digital interfaces introduce immediate distraction through notifications and clean formatting instincts that slow raw conceptual flow. Placing a visual countdown clock on the central screen keeps urgency visible without requiring verbal interruptions from you.

Each participant must write at least 8 distinct concepts on physical index cards, targeting an output rate of one idea every 2.5 minutes. Require participants to describe every concept with a clear verb-noun pair and a single operational mechanism. Enforcing complete silence prevents dominant voices from steering the room toward safe consensus.

Minutes 35–70: Active AI Expansion Loops

This 35-minute block pairs human concepts with large language models using structured five-minute cycles. Participants open their AI interfaces (such as Claude, ChatGPT, or local deployment clients) and execute rapid prompt sprints designed to force concepts into extreme boundaries. For structured prompt architectures designed for these specific sprints, review the 90-Minute GenAI Divergence Workshop (With Prompts) guide.

Run exactly six 5-minute sprints back-to-back, leaving 5 minutes for buffer transitions:

• Sprint 1 (Resource Inversion): Force the model to solve the problem with an operational budget of zero dollars, then rerun it assuming a $50 million line item.
• Sprint 2 (Domain Collision): Paste analog concepts and prompt the model to recombine them using mechanisms drawn from biological ecosystems, commercial aviation, or industrial logistics.
• Sprint 3 (Extreme Regulatory Constraints): Instruct the model to adapt concepts to survive complete data privacy bans or zero internet connectivity.
• Sprint 4 (Speed Stress-Testing): Demand operational architectures that deliver user outcomes in under 3 seconds instead of 3 days.
• Sprint 5 (Antagonist Modeling): Prompt the system to design aggressive counter-measures assuming a well-funded rival wants to make the concept obsolete within 30 days.
• Sprint 6 (Edge-Case Magnification): Force the AI to adapt the product entirely for the bottom 1% of non-standard users.

During these loops, participants do not refine language or fix grammar. They mine the output strictly for unexpected mechanisms and write them on colored cards to differentiate them from analog ideas. This exercise mirrors structured divergent thinking methods, using generative models to surface non-obvious permutations that internal team habits routinely filter out.

Minutes 70–90: Concept Clustering Without Down-Voting

The final 20 minutes organize raw outputs into an operational map while prohibiting elimination or voting. Most facilitators kill momentum at this stage by introducing red dot stickers to select "winners." Doing so forces premature convergence and rewards familiar ideas that feel safe to the middle of the room.

Instead, direct the team to group ideas based on structural contradictions and operational friction. A team of 8 people routinely generates between 120 and 160 distinct index cards across the analog and AI phases. Have the room mount these cards onto a large workspace wall, grouping them under tensions such as "Zero Latency vs. High Verification" or "Complete Automation vs. Human Oversight." To explore how tension mapping feeds downstream development, study balancing divergent and convergent thinking in design.

Instruct the room to tag outliers—ideas that do not fit into any cluster—with bright blue border tape. These isolated edge cases often pinpoint the most valuable exploratory territory.

Keep the momentum moving, because the next protocol details how to turn these friction clusters into testable hypothesis statements without losing their original radical scope.

Prompt Archetypes That Force AI into Lateral Territory

Default generative AI prompts produce regression to the mean because large language models calculate the statistically most probable next word across billions of internet training texts.

Regression to the mean in artificial intelligence refers to an algorithm’s tendency to output safe, common, and highly predictable answers because it is optimized to generate the mathematically average response found across its training dataset.

If you prompt an LLM with "Give me 10 marketing ideas for a financial app," you receive the same 10 ideas every fintech team discussed three years ago. In an active workshop setting, your role as facilitator is to break this consensus engine. Three prompt archetypes pull tools like OpenAI’s ChatGPT and Anthropic’s Claude out of predictable patterns and force them into lateral territory.

1. Inversion Prompting

Inversion flips the problem objective upside down to reveal structural blind spots that teams overlook during standard brainstorms. This method builds on the classical inversion principle popularized by Charlie Munger in Poor Charlie’s Almanack, where solving the opposite of a goal highlights critical operational risks.

Instead of asking AI how to optimize a customer experience, instruct it to engineer complete failure. For example: "Design a customer onboarding sequence for our enterprise software that guarantees 90% user churn within 14 days without alerting management."

When the model lists the specific mechanisms of failure—such as burying setup permissions inside nested menus or requiring mandatory phone verifications—the team maps each failure point to its product backlog. The inverse prompt surfaces friction points that polite corporate consensus usually ignores. Teams running divergent thinking methods use these inverted outputs as direct criteria for redesigning their actual workflows.

2. Cross-Domain Collision Prompts

Cross-domain collision forces the AI to solve your problem using the operating principles of an unrelated discipline, biological system, or historical era. These collisions strip away standard industry jargon and expose the underlying mechanics of the challenge.

Structure collision prompts by pairing your specific business bottleneck with an alien operational environment:

• Biological mechanics: "Explain how an ant colony manages load balancing during flash floods, and translate those exact three coordination principles into our microservices data architecture."
• Unrelated industries: "Redesign our hospital emergency room intake triage using the pit-stop operational model of a Formula 1 racing team."
• Historical constraints: "How would naval dockyards in 18th-century Venice solve our supply chain inventory tracking without digital computing?"

As documented by Christian Terwiesch and Karl Ulrich in their research for the Wharton School at the University of Pennsylvania, generative AI produces its highest-variance concepts when constrained by strict analogical boundaries. The resulting concepts give teams unexpected operational metaphors that spark distinct, testable product features.

3. The ‘Discard the First Five’ Rule

When participants run initial divergent prompts during a 90-Minute GenAI Divergence Workshop (With Prompts), the model outputs the most common concepts in its vector database first. In a 2023 study by researchers at the Wharton School analyzing 200 raw business ideas generated by LLMs, the first ideas scored highest on immediate feasibility but lowest on originality.

To bypass this generic baseline, establish a strict workshop intervention: participants must discard the first 5 ideas generated by any prompt.

Instruct the room to use explicit exclusion scripting directly in their prompts:

Problem: Reduce onboarding drop-off for our SaaS product.
Task: List 15 distinct interventions.
Constraint 1: Identify the 5 most standard, obvious industry 
solutions to this problem.
Constraint 2: Delete those 5 solutions completely.
Constraint 3: Output only ideas 6 through 15, ensuring each 
relies on a non-digital or counter-intuitive mechanism.

This filter eliminates standard ideas—such as adding a progress bar, sending reminder emails, or offering a signup discount—before team discussion starts. By forcing the algorithm past its statistical median, participants spend their limited workshop minutes evaluating unconventional concepts that balance raw divergence with actual execution requirements, a cornerstone when balancing divergent and convergent thinking in design.

Once your participants understand how to construct these prompts, the next step is managing the clock so teams do not get trapped in prompt loops—starting with the exact 15-minute prompt sprint schedule detailed below.

Troubleshooting Participant Friction and Common Workshop Pitfalls

Participants in generative ideation workshops lose up to 40% of their allotted concept-generation time to prompt tweaking rather than idea evaluation. When teams interact with large language models, they gravitate toward debugging syntax instead of testing raw concepts.

The prompt rabbit hole happens when a user endlessly edits input phrasing to make a single output look polished, mistaking refined wording for conceptual novelty.

To eliminate this friction, enforce a strict two-turn rule across all divergent ideation brainstorming methods. If a prompt does not surface an unexpected direction in two iterations, the participant must abandon that branch and post the raw result to the shared workspace. A study by researchers at Harvard Business School examining consultant performance with Boston Consulting Group found that professionals using OpenAI’s GPT-4 completed 12.2% more tasks and produced results rated 40% higher in quality, but only when they avoided over-refining single prompt strings. Treat the model as a volume generator, not a copy editor.

Uneven software literacy creates a split room within 15 minutes. Power users run complex multi-step chains, while non-technical contributors freeze at an empty chat box. Solve this disparity by providing pre-built prompt scaffolding cards directly inside your digital workspace, such as Miro or Mural. Pair participants into two-person cells: one domain expert and one tool driver. This peer pairing keeps idea velocity steady across technical skill levels, ensuring that domain knowledge drives the output rather than software fluency. This approach keeps teams aligned with core divergent thinking skills without bogging down the session in tool tutorials.

[Prompt Card]
     |
     v
[Domain Input]
     |
     v
[Raw AI Output]
     |
     v
[Post to Board]

At minute 50 of a 90-minute agenda, participant energy hits an identifiable wall. Teams exhaust their obvious associations and begin recirculating variations of their first three concepts.

Break this cognitive plateau by injecting an absurd external constraint right at the 50-minute mark. Stop all ongoing work and introduce a hard operational limitation: the solution must run on zero electricity, cost under $0.50 per unit, or serve customers entirely in total darkness.

Forcing teams to apply divergent thinking methods against an extreme operational limit disrupts linear thinking patterns. In Lateral Thinking, author Edward de Bono documented how intermediate impossibilities serve as stepping stones to viable ideas that standard logic misses. Give the room exactly 7 minutes to generate 10 radical solutions that obey the new constraint before returning to the original brief.

Maintaining this tight time discipline requires a visible countdown mechanism that the entire room monitors simultaneously.

Once teams clear the minute-50 hurdle, the challenge shifts from unblocking stuck participants to capturing their fastest-moving concepts before convergence begins.

Knowing how to unstick a stalled group sets you up for the critical transition into structured scoring, which requires a firm shift in facilitator posture.

Complete Facilitator Script and Printable Timer Cheat Sheet

A 90-minute AI divergent thinking workshop succeeds or fails based on strict time boundaries and explicit prompt instructions that prevent participants from debating the model’s raw output during generation rounds.

A run-of-show is a minute-by-minute operational schedule that details every event transition, participant cue, technical trigger, and facilitator script line for the duration of a structured meeting or workshop.

When running ideation with generative tools, individuals often get sucked into rabbit holes, editing a single prompt response instead of multiplying ideas. Research published by Ethan Mollick and Lilach Mollick at the Wharton School of the University of Pennsylvania shows that structured human-AI co-intelligence produces broader idea variance only when participants use targeted prompts rather than open-ended queries.

To maintain momentum, facilitators must keep spoken cues crisp and rely on physical cues like a visual countdown timer to enforce the 5-minute sprint limits.

Verbatim Facilitator Transition Scripts

Use these exact wordings to direct the room, manage energy, and establish psychological safety without wasting time on theory.

1. Workshop Kickoff and Ground Rules (Minute 00:00 – 00:10)

"Welcome. Today we have 90 minutes to generate high-variance concepts for our challenge using artificial intelligence. Rule one: quantity beats quality in this room today. Rule two: do not argue with or correct the AI’s first answer—push it sideways instead. Rule three: our timers are hard stops. When the chime sounds, your hands leave the keyboard. Open your chosen AI assistant, paste your problem statement at the top of your document, and look up at me when you are ready."

2. Launching Sprint 1: Extreme Constraint Inversion (Minute 00:10 – 00:25)

"We are starting Sprint 1. You have 5 minutes of heads-down solo work. Copy Prompt 1 from your handout directly into your chat. Do not tweak the phrasing yet; run it as written, read the ten outputs, and pick the two most unusual ideas to drop into your workspace. Timer starts in three, two, one, go."

(At Minute 00:15 / Chime sounds)

"Keyboards down. Take 60 seconds to review what appeared on your screen. You have four minutes with your table partner: read your two wildest concepts aloud, record them on yellow sticky notes, and do not judge them yet. Share first, critique never. Go."

3. Launching Sprint 2: Cross-Industry Analogies (Minute 00:25 – 00:40)

"Sprint 2 shifts context entirely. We are taking our core bottleneck and looking at how other industries solved identical mechanics. Copy Prompt 2 into your chat. Give the AI 30 seconds to respond, then demand three implementations borrowed strictly from hospitality or aviation. Five minutes solo on the clock. Start now."

(At Minute 00:30 / Chime sounds)

"Hands off keyboards. Transfer your best cross-industry idea to a blue sticky note. You have four minutes to share with a new partner across your table. Go."

4. Transitioning to Convergence and Synthesis (Minute 00:55 – 00:70)

"We have exited divergence. We generated roughly 60 distinct angles across this room. Now we apply commercial filters. You have eight minutes to cluster your notes by shared underlying mechanism, not by department. If two ideas solve the problem using speed, group them together. Move to the boards now."

Practitioners who run a 90-Minute GenAI Divergence Workshop (With Prompts) know that clear scripts prevent the paralysis that occurs when teams stare at a blank prompt box. Using these verbatim lines preserves the structure necessary for The Role of Divergent Thinking in Creative Breakthroughs.

5-Minute Sprint AI Prompt Bank

Distribute these three prompts on a physical handout or paste them into your digital workspace before the session begins. They translate classical Divergent Thinking Methods into direct AI instructions.

Sprint 1: Extreme Constraint Removal & Inversion

Act as a disruptive product strategist. We are solving this problem: [INSERT PROBLEM STATEMENT]. Generate 10 distinct solutions under the following extreme condition: Assume our budget is zero dollars, but we have access to 100,000 passionate university volunteers. Eliminate all traditional software development. Focus strictly on behavioural nudges, physical logistics, and manual coordination. Output as a numbered list with a one-sentence rationale for each.

Sprint 2: Cross-Industry Mechanism Transfer

Here is our operational bottleneck: [INSERT BOTTLENECK]. Identify how three unrelated sectors—specifically deep-sea commercial salvage, luxury resort concierge operations, and emergency trauma triage—solve this underlying structural challenge. For each of the three industries, describe their mechanism in 2 sentences, then translate that exact mechanism into an unconventional solution for our bottleneck.

Sprint 3: The "Worst Possible Idea" Reframe

We want to solve: [INSERT PROBLEM STATEMENT]. First, generate 5 intentionally terrible, catastrophic ideas that would make this problem 10 times worse, increase customer churn, and destroy efficiency. Second, for each terrible idea, extract the underlying mechanic and flip it 180 degrees into an unconventional, high-value intervention that we would never normally consider.

Act as an executive facilitator. I need a 90-minute divergent thinking workshop agenda focused on this problem: [INSERT SPECIFIC PROBLEM STATEMENT]. 

The room contains [NUMBER] participants with [BEGINNER/INTERMEDIATE/ADVANCED] AI experience.

Build a complete facilitator schedule formatted in a markdown table with four columns:
1. Elapsed Time (Minutes 00:00 to 90:00)
2. Activity & Solo/Group Mode
3. Spoken Facilitator Script (verbatim instructions)
4. Visual or Timer Cue

Include three distinct 5-minute AI generation sprints using divergent lenses (such as analogical transfer, constraint elimination, and reversal). For each sprint, write the exact prompt the participants must paste into their chat tool.

1-Page Printable Run-of-Show Timer Sheet

Print this schedule and keep it on your clipboard to track every transition. A study by the Nielsen Norman Group on design thinking facilitation emphasizes that visual time trackers keep distributed and in-person teams 28% more aligned on task goals compared to ad-hoc verbal notifications.

Maintaining balance across this run-of-show ensures teams avoid generating dozens of abstract concepts without selecting viable paths forward, a core tenet of Balancing Divergent and Convergent Thinking in Design. If you want a structured debrief format to pair with this schedule, adapt the time blocks from our Disney Creative Strategy: Setup, Timers & Script (Template) or review the mechanics in our 60-Minute Ideation Workshop Agenda (With Script).

Print this run-of-show sheet, paste your specific problem statement into the Sprint 1 prompt template above, and schedule your team’s 90-minute AI divergent thinking session on the calendar today.

Sources & Further Reading

Grounding an AI divergent thinking workshop in established cognitive science and empirical management research prevents the session from devolving into unstructured brainstorming.

Divergent thinking is a cognitive process that generates creative ideas by exploring many possible solutions in a short timeframe rather than seeking a single correct answer.

In a 2023 study published by the Harvard Business School, researchers Karim Lakhani and Fabrizio Dell’Acqua examined 758 consultants working with artificial intelligence. The study found that professionals pairing generative models with structured ideation workflows expanded their creative output by 40% compared to control groups working without synthetic assistance.

Cognitive fixation occurs when early, dominant concepts suppress novel thought pathways during collective brainstorming. An agenda divided into tight, rapid sprint cycles directly breaks this tendency.

In his pioneering text Lateral Thinking: Creativity Step by Step (1970), Edward de Bono proved that human lateral search requires deliberate mechanical interruptions to escape entrenched judgment patterns. Pairing human team prompts with rapid, asynchronous large language model generations gives facilitators precisely that mechanical interruption.

Featured image by Alla Eddine Taleb on Pexels