90-Minute GenAI Divergence Workshop (With Prompts)

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

⏱ 26 min read

The Complete 90-Minute GenAI Divergence Workshop Architecture

A GenAI Creative Divergence Workshop is a 90-minute structured ideation session that uses AI text and image models to rapidly expand the breadth of potential solutions across 4 distinct phases—Framing (15m), Volume Expansion (30m), Extreme Boundary Testing (30m), and Clustering (15m). Unlike conventional brainstorming, participants do not ask AI for answers; they use parameterized prompt archetypes to generate 50+ divergent conceptual angles before any filtering begins.

A parameterized prompt archetype is a reusable text template containing fixed structural constraints alongside variable slots for user constraints, operational limits, and forced analogies. This structure forces an AI model to explore non-obvious solution spaces instead of returning default statistical averages.

Without these templates, open-ended AI sessions collapse into predictable dead ends. The Role of Divergent Thinking in Creative Breakthroughs requires expanding the problem space before selecting answers, but conversational interfaces lure teams into immediate convergence.

The Ad-Hoc Failure Mode: The 10-Minute Homogenisation Trap

Unguided teams almost always fail in the first 10 minutes of AI brainstorming. A participant types a prompt like, "Give me 10 innovative ways to reduce customer churn in our SaaS platform." Within 4 seconds, the large language model outputs the statistical center of its training data: gamified onboarding, tiered loyalty discounts, and automated check-in emails.

In a 2023 research study titled Navigating the Jagged Technological Frontier, researchers Fabrizio Dell’Acqua and Karim Lakhani from Harvard Business School examined 758 consultants using generative AI on creative problem-solving tasks. The study found that while individual output quality rose, collective diversity across ideas plummeted by 41% when participants relied on standard conversational prompting. The models generate homogenous, safe concepts because reinforcement learning from human feedback penalizes bizarre or high-variance suggestions.

When participants receive these clean, grammatically polished answers, cognitive anchoring takes over. The team spends the remaining 80 minutes debating variations of the same 3 standard recommendations instead of exploring uncharted angles. Applying Unlocking Creative Flow: Bias-Free Ideation Techniques prevents this premature consensus by imposing strict barriers between prompt generation and concept evaluation.

Participant Setup Requirements

A divergence workshop requires strict operational preparation. If participants enter the session with mismatched tooling or unvetted problem statements, the 90-minute timeline disintegrates.

Every participant must meet 3 operational prerequisites before minute zero:

1. Standardized Model Access: Every participant must use the exact same foundation tier—such as OpenAI ChatGPT Plus running GPT-4o or Anthropic Claude 3.5 Sonnet. Free-tier accounts introduce differing context window caps and reasoning limits, which skew team output. Participants generating visual concepts also require identical access to Midjourney v6 or DALL-E 3.
2. Seed Problem Framing Rules: The facilitator must lock the challenge into a single sentence under 30 words that contains zero solution verbs. A statement like "Design a mobile app to improve warehouse safety" fails because it dictates the delivery mechanism. The correct framing is: "Warehouse pickers face a 14% injury rate during peak holiday shifts due to physical fatigue." For more detail on structural session planning, consult the 60-Minute Ideation Workshop Agenda (With Script).
3. Zero-Prior-Prompting Baseline Agreement: Every participant must open a completely fresh browser session with an empty chat history. Pre-existing conversation histories inject hidden context into the model’s memory buffer, creating semantic drift that derails standardized prompt testing.

With the technical baseline secured, the focus shifts to running the minute-by-minute protocol on the workshop floor.

Table of Contents

• The Complete 90-Minute GenAI Divergence Workshop Architecture
• Minute-by-Minute 90-Minute Divergence Workshop Timeline
• Four Facilitator Prompt Archetypes for Creative Volume
• Live Facilitation Rules to Prevent Semantic Groupthink
• The Facilitator Run Sheet and Copy-Paste Prompts
• Sources & Further Reading

Minute-by-Minute 90-Minute Divergence Workshop Timeline

A 90-minute generative AI divergence workshop succeeds only when strict timeboxes prevent participants from editing prompt outputs during the ideation phase. Most corporate teams default to judging concepts as soon as the language model generates them. This habit kills exploration and leaves teams with obvious, incremental ideas.

By enforcing distinct operational phases, you direct the AI model to surface lateral associations before human critique narrows the field. This structure relies directly on the role of divergent thinking in creative breakthroughs, where raw volume and cognitive variation must precede any selection process.

Phase 1 (Minutes 00–15): Problem Unpacking and Constraint Stripping

Phase 1 isolates your core challenge and strips away institutional assumptions that limit original thinking. In standard project meetings, teams automatically assume fixed budgets, standard compliance cycles, and legacy software limits. You must remove these defaults before writing a single prompt.

Constraint stripping is an ideation method where a team deliberately lists and removes standard business rules—such as budget caps, existing tech stacks, or approval gates—so they can explore solutions outside business-as-usual limits.

Spend the first 5 minutes writing the core problem statement on a shared digital board in tools like Miro or Mural. For the remaining 10 minutes, have every participant list three unquestioned assumptions about the problem, such as "this requires enterprise sales reps" or "users will only access this via an iPhone app." Strip them. Direct everyone to paste these negative constraints into their AI chat interfaces (such as OpenAI’s ChatGPT Enterprise or Anthropic’s Claude) alongside the base problem statement.

Instruct the model explicitly: "List 5 underlying systemic causes of this problem, assuming zero legacy software dependencies and zero sales team overhead." This baseline keeps the AI from regurgitating standard corporate playbooks.

Phase 2 (Minutes 15–45): High-Velocity Volume Sprints

Phase 2 generates raw volume under strict time pressure to bypass self-censorship. Research from Harvard Business School by Karim Lakhani shows that working with frontier AI models significantly boosts idea generation across knowledge workers, lifting overall creative diversity when teams iterate rapidly.

Run six individual, 5-minute prompting rounds without group discussion between rounds. Each participant works in private AI chat sessions to produce at least 30 raw solution angles per person. The rule is absolute: no evaluating, no editing, and no sharing until the 30-minute block ends.

If you have run a 60-minute ideation workshop agenda (with script), you know that open verbal brainstorming allows loud voices to dominate. Silent, parallel AI generation ensures total psychological safety and maximizes variation.

Participants use conversational prompt chaining during this phase. Prompt chaining means passing the output of one model prompt into a subsequent prompt as context to build complex lines of thought step-by-step.

Sprint 1 targets extreme simplicity. Sprint 2 targets extreme automation. Sprints 3 through 6 push into operational opposites, such as shifting a digital solution entirely into a tactile, physical service. Use a clear visual timer to keep participants moving.

By minute 45, an eight-person workshop will have accumulated 240 distinct angles.

Phase 3 (Minutes 45–75): Orthogonal Constraint Injection

Phase 3 forces the AI models out of conventional thinking by introducing non-obvious frames of reference. Left to themselves, large language models output the statistical middle—the most predictable answer to any brief. You break this regression to the mean by introducing three orthogonal lenses in three 10-minute bursts.

Orthogonal thinking is the practice of solving a challenge by borrowing principles from completely unrelated industries or systems rather than looking at direct competitors.

Run these three lenses in sequence:

1. Biomimicry (Minutes 45–55): Command the AI to map the problem onto biological systems. For example: "Analyze this client onboarding bottleneck through the mechanics of mycorrhizal fungal networks in old-growth forests. Provide 5 operational mechanisms."
2. Historical Parallels (Minutes 55–65): Ground the dilemma in pre-digital logistics. Frame the brief around postal networks in 18th-century Europe, naval communication during the 19th century, or Roman aqueduct maintenance. You can adapt techniques from generative AI for creative writing prompts to force unexpected metaphors.
3. Absurd Economic Scenarios (Minutes 65–75): Force the model to solve the problem under two extreme conditions: an infinite budget ($100 million with zero timeline limits) and an impossible budget ($0 with a mandatory launch in 48 hours).

These lenses strip away default corporate answers and yield concepts that direct competitors will never consider.

Phase 4 (Minutes 75–90): Semantic Distance Clustering

The final 15 minutes sort ideas by the magnitude of their conceptual departure, rather than typical functional categories like "Marketing," "Product," or "Operations." Grouping by standard business departments immediately pushes teams back into siloed thinking.

Semantic distance is the degree of conceptual difference between two ideas, measured by how few contextual associations or vocabulary terms they share.

According to research published by cognitive psychologist Yoed Kenett in the journal Trends in Cognitive Sciences, high creative achievement correlates directly with navigating distant semantic concepts rather than close associative networks.

Participants select their top 4 most unconventional angles from Phases 2 and 3 and paste them into the group workspace. You then guide the room to sort them into three specific clusters:

• Low Semantic Distance: Incremental improvements to existing workflows.
• Medium Semantic Distance: Structural changes that borrow models from adjacent industries.
• High Semantic Distance: Radical paradigm shifts that alter the business model, unit economics, or user interface entirely.

This sorting shows the team the full range of options. It prevents safe, incremental ideas from drowning out genuine breakthroughs. Teams exploring broader organizational adjustments often pair this final sorting with a 90-minute systems thinking workshop agenda (with script) to map downstream effects before committing engineering resources.

WORKSHOP SETUP
Total Duration: 90 Minutes
Tooling: Individual AI accounts (ChatGPT, Claude, or internal LLM), shared digital board (Miro/Mural)
Objective: Generate 200+ divergent angles and isolate high-semantic-distance concepts

PHASE 1: PROBLEM UNPACKING & CONSTRAINT STRIPPING (00:00 - 00:15)
[00:00 - 00:05] Facilitator Announcement:
"Welcome. Today is about divergence, not decisions. We are generating raw concept volume using AI. We will not evaluate, judge, or edit anything during the next 75 minutes. The core problem statement is: [INSERT CORE PROBLEM STATEMENT]."

[00:05 - 00:15] Individual Action (Assumption Stripping):
Prompt for participants to copy into their LLM:
---
I am working on the following problem: [INSERT CORE PROBLEM STATEMENT].
Identify 5 foundational assumptions that standard corporate teams make when solving this problem. 
Then, rewrite the problem statement three times, removing each assumption completely.
---

PHASE 2: HIGH-VELOCITY VOLUME SPRINTS (00:15 - 00:45)
Facilitator Announcement:
"We are running six 5-minute sprints. Total silence. No cross-talk. Do not read through your outputs yet. Your target is 30 raw solutions in your private chat window."

Sprints 1-2 Prompt (Extreme Simplification & Automation):
---
Given the stripped problem: [INSERT STRIPPED STATEMENT FROM PHASE 1].
Generate 10 distinct solution angles that require:
1. Zero human intervention after setup.
2. An interface simple enough for an 8-year-old.
Give each angle a 3-word title and a 1-sentence operational mechanism.
---

Sprints 3-4 Prompt (Channel & Asset Inversion):
---
Given the problem: [INSERT STRIPPED STATEMENT].
List 10 ways to solve this without using software, screens, or internet connectivity. 
Translate the core value proposition into physical, spatial, or paper-based systems.
---

Sprints 5-6 Prompt (Speed & Scale Extremes):
---
Given the problem: [INSERT STRIPPED STATEMENT].
Generate 5 solutions that must execute in under 3 seconds per transaction, and 5 solutions designed to take 10 years to complete. Focus on value creation at both extremes.
---

PHASE 3: ORTHOGONAL CONSTRAINT INJECTION (00:45 - 01:15)
Facilitator Announcement:
"Now we force the model out of standard industry patterns. Run these three prompts in your current chat session."

[00:45 - 00:55] Lens 1: Biomimicry
---
Translate our problem: [INSERT STRIPPED STATEMENT].
Find 3 organisms or ecological systems that solve an analogous resource or communication challenge. 
Extract the biological mechanism and turn it into a concrete business model concept.
---

[00:55 - 01:05] Lens 2: Historical Parallels
---
Analyze our problem: [INSERT STRIPPED STATEMENT].
How would a logistics coordinator in the Roman Empire, a master builder in 14th-century Venice, or a telegraph operator in 1880 solve the root mechanic of this dilemma? 
Provide 3 concepts adapted to modern markets.
---

[01:05 - 01:15] Lens 3: Impossible Economics
---
Analyze our problem: [INSERT STRIPPED STATEMENT].
Concept A: You have a $100M budget and zero regulatory oversight.
Concept B: You have an absolute $0 budget and must launch in 48 hours using existing public infrastructure.
Provide 3 concrete execution paths for each scenario.
---

PHASE 4: SEMANTIC DISTANCE CLUSTERING (01:15 - 01:30)
[01:15 - 01:20] Output Transfer:
Facilitator: "Review your chat history. Pick your 4 most unusual, non-obvious solution concepts. Copy and paste them as individual sticky notes on our shared workspace board."

[01:20 - 01:30] Group Sorting (Silent Categorization):
Facilitator: "Do not group by department (e.g., Marketing, Product, Tech). Drag sticky notes into three columns:
1. Low Semantic Distance: Familiar approaches we could ship tomorrow.
2. Medium Semantic Distance: Novel approaches borrowed from adjacent industries.
3. High Semantic Distance: Radical approaches that challenge our operating model.
Begin sorting now."

Once your team organizes these concepts across the three semantic zones, you must establish clear evaluation metrics to filter the viable breakthroughs from outright operational liabilities.

Four Facilitator Prompt Archetypes for Creative Volume

Facilitators can break default large language model outputs and generate over 100 distinct concepts in 15 minutes by cycling participants through four structured prompt archetypes: the Inversion Prompt, the Cross-Domain Hybridizer, the Zero-Resource Provocateur, and the Future Historian.

Creative divergence is an ideation method where participants deliberately generate a high volume of non-obvious concepts across wide solution spaces before applying any evaluation criteria to filter them.

Standard generative AI queries yield standard corporate consensus. When a workshop participant asks OpenAI ChatGPT or Anthropic Claude to "brainstorm marketing strategies for a retail app," the model predicts the most statistically probable tokens, returning safe ideas like loyalty points or push notifications. To force genuine variance, facilitators must inject specific structural constraints into the prompt architecture.

1. The Inversion Prompt

The Inversion Prompt forces the language model to optimize explicitly for catastrophic failure modes, systemic collapse, or customer alienation. Instead of asking how to solve a problem, you command the model to construct the most efficient way to cause it.

This technique builds on the pre-mortem framework developed by cognitive psychologist Gary Klein. Research published in Harvard Business Review shows that prospective hindsight—imagining an event has already failed before it begins—increases a team’s ability to identify real failure reasons by 30%. Reversing the objective strips away corporate optimism bias and exposes hidden operational risks.

Consider an enterprise team working on client retention. A standard prompt produces generic customer service tips. The Inversion Prompt runs in two consecutive steps:

Step 1 Prompt:
"List 10 decisive actions our account team 
could take over the next 30 days to guarantee 
that 80% of our enterprise clients cancel 
their contracts immediately."

Step 2 Prompt:
"Take the 3 most destructive mechanisms 
from that list. Invert each one into a 
counter-intuitive protocol that protects 
client retention."

When participants run this in small breakout groups, the AI exposes latent vulnerabilities that normal brainstorming misses. Teams discover that rapid churn often stems from over-communicating low-value project status updates. Inverting that failure point leads to asynchronous, exception-only client dashboards. Using this archetype helps teams unlock creative potential by challenging confirmation bias before committing engineering resources to flawed premises.

2. The Cross-Domain Hybridizer

The Cross-Domain Hybridizer instructs the AI to resolve a target challenge using the exact operational mechanisms, vocabulary, and physical principles of an entirely unrelated industry.

In a 2023 working paper on generative AI and ideation, Christian Terwiesch and Karl Ulrich at the Wharton School documented that large language models generate raw concepts at a rate roughly 40 times faster than human teams. However, raw volume lacks value if the conceptual variety remains flat. Cross-domain prompts deliberately shock the model’s latent space, bridging semantic networks that do not normally overlap in business strategy.

Direct participants to select an external industry completely removed from their market:

• Deep-sea commercial salvage
• Intensive care unit (ICU) patient triage
• Commercial poultry farming
• Formula 1 pit stop telemetry

Hybridizer Prompt Syntax:
"Act as a chief systems engineer for Formula 1 
pit stop operations. Solve our problem: 
[User onboarding drop-off after 48 hours]. 
Map the exact telemetry, sensor verification, 
and sub-3-second handoff mechanics of a tire 
change onto our digital onboarding flow. 
Provide 5 concrete feature mechanisms."

A financial services firm addressing loan approval delays can map its pipeline directly to hospital emergency room triage protocols. The AI response substitutes standard linear review queues with real-time vitals tracking, dynamic reassignment, and immediate bedside verification. This lateral transfer exposes fresh architectural options, showing participants the role of divergent thinking in creative breakthroughs when standard industry approaches hit performance plateaus.

3. The Zero-Resource Provocateur

The Zero-Resource Provocateur strips away every standard business asset, commanding the model to solve the problem with zero operational budget, immediate execution within 24 hours, or complete absence of digital technology.

Default AI responses consistently prescribe resource-heavy tactics: custom software development, paid ad campaigns, or dedicated hiring. These recommendations mirror the bloated initiatives that often derail projects inside large organizations. By removing money, time, and code from the model’s solution space, the facilitator forces the tool to surface lean, structural interventions.

Zero-Resource Prompt Syntax:
"We need to increase employee engagement with 
our internal compliance portal by 40%. 
Constraint set:
1. Capital budget: $0.
2. Software development time: 0 hours.
3. No new software licenses.
4. Execution must happen within 8 business hours 
   using existing email, paper, or Slack.
Generate 6 unconventional execution tactics."

This archetype mirrors the core discipline of lean problem-solving described in Eric Ries’s The Lean Startup, where progress depends on rapid validation rather than capital scale. For enterprise teams accustomed to six-figure project scopes, running this prompt produces jarringly practical mechanisms. Teams frequently discover manual interventions they can test by noon the following day, bypassing multi-month software roadmaps. Facilitators can combine this approach with a 60-minute ideation workshop agenda to compress solution testing from quarters into single afternoons.

4. The Future Historian

The Future Historian positions the generative model at an imagined point 30 years in the future, looking back to document how an unconventional, controversial decision made today created an industry-defining standard.

When prompted in the present tense, AI models lean toward cautious consensus. Framing the prompt as historical analysis changes the model’s narrative parameters. The tool stops listing risks and instead writes a retrospective case study that explains how early market skepticism dissolved into mainstream adoption.

Future Historian Prompt Syntax:
"The year is 2055. Our company, [Insert Name], 
is globally recognized for radically transforming 
[Insert Problem Area] through a bizarre decision 
made in 2025 that competitors initially mocked. 
Write an excerpt from a business history textbook 
detailing:
1. The exact weird mechanism we introduced.
2. Why industry analysts in 2025 predicted it 
   would fail within 6 months.
3. The structural tipping point that made it 
   our primary commercial moat."

This chronological displacement frees workshop participants from defending short-term quarterly metrics. A logistics team examining package delivery might uncover AI-generated scenarios where residential customers manage local consolidation depots in exchange for utility subsidies. The scenario sounds improbable on a 12-month horizon, but perfectly rational across three decades.

Divergence Prompt Sequence
       │
       ▼
 [1. Inversion] ──────────► Invert failures
       │
       ▼
 [2. Hybridizer] ─────────► Cross industries
       │
       ▼
 [3. Zero-Resource] ──────► Strip budget
       │
       ▼
 [4. Future Historian] ───► Project 30 years

Applying these four archetypes systematically converts generative AI from a predictable copywriter into a high-variance concept generator. The challenge shifts immediately from generating ideas to scoring them—which brings us to the real-time evaluation matrix we will use to filter this raw volume into validated pilot projects.

Live Facilitation Rules to Prevent Semantic Groupthink

Live facilitation must prevent semantic groupthink by isolating individual prompt generation before team members share any model outputs. Semantic groupthink is an ideation trap where participants cluster around the exact vocabulary, metaphors, and conceptual frames returned by early AI responses, collapsing creative variance.

A 2023 Harvard Business School study by Fabrizio Dell’Acqua and colleagues examined 758 consultants using OpenAI models. The researchers found that while individual output quality rose, collective diversity plummeted: consultants using AI produced 41% less variation across their solutions compared to control groups. When one participant shares a standard generative AI response out loud early in a session, the room anchors to that semantic pattern immediately. To preserve the role of divergent thinking in creative breakthroughs, facilitators must enforce specific operational controls throughout the live workshop.

The Silent Sprint Protocol

Run an 8-minute silent sprint at the start of divergence before anyone speaks. Each participant sits at their own terminal, enters prompts individually, and records their outputs without sharing screens.

Verbal discussion during initial prompting creates instant anchoring bias. Team members inevitably adopt the phrasing of the first person who reads a high-probability model response aloud. During the 8-minute window, require participants to document at least 3 distinct conceptual directions inside their personal notes. Nobody discusses ideas, pastes text into common Slack channels, or projects their interface onto shared screens until the sprint timer hits zero. Applying structured constraints this way supports unlocking creative flow: bias-free ideation techniques across cross-functional groups.

Intervention Techniques for Repetitive Answers

Teams running standard prompts on default settings will hit identical, middle-of-the-road answers within 15 minutes. When you notice three consecutive tables pitching minor variations of the same product or workflow, step in with 3 direct mechanical resets:

1. Shift the Temperature Parameter: If participants use developer playgrounds or interfaces that permit configuration, instruct them to move the temperature setting from the default 0.7 up to 1.1 or 1.2. The temperature parameter is a numerical setting that controls how randomly a language model selects its next word, with higher numbers producing less predictable language. Lower temperatures yield conservative, expected consensus phrases; higher temperatures force the model to select low-probability tokens that break conventional phrasing.
2. Alter the System Role: Standard prompts like "Act as an innovative product director" produce corporate cliches. Force teams to re-assign antagonistic or non-obvious roles. Command the model to respond as a forensic auditor searching for fraud, an 18th-century naval quartermaster, or an aggressive safety regulator. Changing the operational persona changes the vocabulary distribution immediately, similar to methods used for generative AI for creative writing prompts.
3. Enforce Hard Word Bans: Ban the five most common industry buzzwords from both the user prompt and the required output. For a corporate fintech sprint, ban "seamless", "frictionless", "dashboard", "ecosystem", and "empower". Forcing negative constraints denies the model its highest-probability predictive pathways, compelling it to formulate alternate semantic links.

The 60-Second Plausibility Challenge

Facilitators must distinguish high-variance breakthrough ideas from impossible model hallucinations quickly. Ethan Mollick at the Wharton School of the University of Pennsylvania notes that creative AI use requires working at the frontier of competence, where bizarre outputs often contain viable unconventional logic. Do not dismiss an odd output immediately.

When a team presents a high-variance, seemingly absurd output, run a 60-second plausibility challenge:

[ AI-Generated Output ]
          |
          v
[ Extract Core Mechanism ] (20 sec)
  Isolate physical or economic lever
          |
          v
[ Apply Reality Filter ] (20 sec)
  Check real-world friction & laws
          |
          v
[ Categorise & Decide ] (20 sec)
  Label as "Viable Seed" or "Hallucination"

Spend the first 20 seconds isolating the underlying mechanism rather than the literal surface pitch. If the model suggests "telepathic customer feedback", strip the fantasy: the core mechanism is zero-effort, passive behavioral feedback collected at the point of experience. Spend the next 20 seconds testing that stripped mechanism against basic regulatory and economic realities. Spend the final 20 seconds making a clean binary call: keep the underlying seed for the next round or discard it as non-actionable fiction.

With semantic guardrails established and low-value noise filtered out, you are ready to move these raw outputs into the structured 4-step convergence rubric detailed in the next section.

The Facilitator Run Sheet and Copy-Paste Prompts

A structured 90-minute GenAI divergence workshop forces teams past consensus ideas by pairing time-boxed human prompts with large language model counter-perspectives. In a 2023 study by researchers at Harvard Business School, consultants using GPT-4 completed creative product innovation tasks 25.1% faster and produced results scored 40% higher in quality compared to a control group without AI. To get those results in a corporate setting, facilitators must run a tightly controlled clock and issue explicit boundary conditions.

Paradigm divergence is a measurement of how radically an idea departs from an industry’s established operating assumptions, revenue models, or delivery methods rather than making simple feature tweaks. Teams default to familiar patterns unless forced out of them, which is why the role of divergent thinking in creative breakthroughs relies on active cognitive friction.

Minute-by-Minute Facilitator Run Sheet

Keep the room moving with these timed stages, spoken scripts, and participant directives.

00:00 - 00:10 | Setup & Baseline
00:10 - 00:30 | Reverse-Assumption Prompting
00:30 - 00:55 | Cross-Domain Collisions
00:55 - 01:15 | Red Teaming & Edge Cases
01:15 - 01:30 | Divergence Scoring

00:00 – 00:10 | Setup and Baseline Calibration (10 Minutes)

• Facilitator Announcement: "Open your assigned LLM workspace—either Anthropic Claude or OpenAI ChatGPT. Do not generate ideas yet. In the next 3 minutes, write down the 3 most common, obvious solutions our industry always proposes for this problem. We are setting our baseline to eliminate low-hanging fruit."
• Time Cue at 00:07: "Three minutes left. Post your obvious ideas to column 1 of our shared board. Everything in that column is now officially banned from our generation prompts."
• Participant Directive: Document the status quo. These banned concepts prevent confirmation bias from polluting the AI prompts.

00:10 – 00:30 | Stage 1: Reverse-Assumption Prompting (20 Minutes)

• Facilitator Announcement: "We are breaking industry assumptions. Copy Prompt Template 1 from your handout. Plug in our banned concepts and execute the run. Your goal is 10 non-traditional product architectures in 8 minutes."
• Time Cue at 00:20: "Halfway through Stage 1. If your model produces standard SaaS models or typical advisory services, increase the constraint pressure. Ask the model to eliminate customer onboarding entirely or operate with zero synchronous communication."
• Participant Directive: Run 3 iterations. Select your 2 weirdest, high-potential concepts and paste them into column 2.

00:30 – 00:55 | Stage 2: Cross-Domain Collisions (25 Minutes)

• Facilitator Announcement: "We are taking our column 2 concepts and smashing them into unrelated legacy industries: industrial agriculture, deep-sea salvage, and commercial aviation. Use Prompt Template 2."
• Time Cue at 00:45: "Ten minutes left in Stage 2. Do not argue with the outputs yet. If an idea feels operationally uncomfortable, keep it. Focus on functional mechanisms, not cosmetic themes."
• Participant Directive: Extract the single strongest operational mechanism from the output and combine it with your Stage 1 concept. While a 60-minute ideation workshop agenda moves fast, this 25-minute block provides the focus needed to force deep system crossovers.

00:55 – 01:15 | Stage 3: Extreme Edge-Case Stress Testing (20 Minutes)

• Facilitator Announcement: "Now we break the concepts. Take your top hybrid concept and feed it to Prompt Template 3. We are instructing the model to act as an aggressive hostile competitor and an adverse economic environment."
• Time Cue at 01:05: "Ten minutes remaining. When the model points out fatal business model flaws, do not abandon the idea. Demand that the LLM preserve the core mechanic while altering the cost structure."
• Participant Directive: Document the survival pivots. Refine the concept into a 3-sentence summary: Mechanism, Unfair Advantage, and Radical Delivery Vector.

01:15 – 01:30 | Stage 4: Scoring and Selection (15 Minutes)

• Facilitator Announcement: "Pencils down on LLM prompts. Open the scoring rubric. Every participant has 6 minutes to score their assigned partner’s concepts across Novelty, Feasibility Tension, and Paradigm Divergence."
• Time Cue at 01:25: "Final five minutes. Plot your top scoring concept onto our 2×2 matrix. Any concept scoring below 3 on Paradigm Divergence is archived."
• Participant Directive: Complete peer evaluations and map concepts. Hand the top 3 workshop outputs to the project lead.

Copy-Paste Prompt Templates

Distribute these templates directly to participants. Each prompt uses variable brackets ([VARIABLE]) for instant customization.

Prompt 1: The Assumption Inverter (Stage 1)

Role: Principal Product Strategist specializing in disruptive business models.
Context: We are solving [CORE PROBLEM] for [TARGET AUDIENCE].
Standard industry solutions rely on these 3 assumptions:
1. [ASSUMPTION 1]
2. [ASSUMPTION 2]
3. [ASSUMPTION 3]

Task:
Generate 8 radically different product or service concepts that solve [CORE PROBLEM] by actively violating all three assumptions above. 
Rules:
- Prohibit any mechanism that relies on [BANNED MECHANISM, E.G., MONTHLY SUBSCRIPTIONS, MANUAL DATA ENTRY].
- Every concept must specify: (a) The inverted operating principle, (b) The user interaction model, (c) How the unit economics function without standard models.
Format: Bullet points, 3 sentences maximum per concept.

Prompt 2: Cross-Domain Mechanic Collision (Stage 2)

Role: Systems Architect.
Concept to Mutate: [INSERT WINNING CONCEPT FROM STAGE 1]
Target Domain: [CHOOSE ONE: DEEP-SEA SALVAGE / INDUSTRIAL AQUAPONICS / AIR TRAFFIC CONTROL]

Task:
Deconstruct how [TARGET DOMAIN] handles [PARALLEL PROBLEM: E.G., RESOURCE SCARCITY / HIGH-STAKES ROUTING / ASYMMETRIC RISK]. 
Map that exact operational mechanism directly onto our concept.
Output Requirements:
- Name the specific engineering or operational principle borrowed from [TARGET DOMAIN].
- Describe how our target audience consumes this solution.
- Identify the counter-intuitive metric this new hybrid system optimizes for.

Prompt 3: Adversarial Red-Teaming (Stage 3)

Role: Hostile Competitor and Risk Officer.
Concept Proposal: [INSERT HYBRID CONCEPT FROM STAGE 2]

Task:
Identify 3 structural vulnerabilities that would cause this business model to collapse within 180 days of launch. Focus on:
1. Economic unviability at scale
2. Behavioral inertia from [TARGET AUDIENCE]
3. Regulatory or legal friction points

For each vulnerability, force a pivot: Rewrite the product mechanic so the vulnerability becomes an operational moat, without reverting to [ASSUMPTION 1].

Printable Divergence Scoring Rubric

A study published by researchers at the Wharton School found that raw ideation with modern language models drops concept generation costs below $0.01 per idea while producing higher average novelty than human-only groups. However, raw volume introduces noise. Evaluate your workshop concepts against this rubric to isolate viable divergence from empty novelty.

Target ideas that score at least 4 in Novelty, 3 or 4 in Feasibility Tension, and 4 in Paradigm Divergence. Ideas scoring 5 on all three categories usually fail immediate business cases; ideas scoring below 3 on Paradigm Divergence are not worth workshop time.

Take your current product bottleneck, drop your three default assumptions into Prompt Template 1 right now, and review the structural alternatives your LLM returns.

Sources & Further Reading

Structured generative AI divergence workshops rely on empirical creativity research and human-computer interaction studies rather than unconstrained brainstorming. Grounding workshop prompts in established cognitive frameworks prevents teams from settling on generic model outputs.

Divergent thinking is a cognitive process used to generate spontaneous, non-linear ideas by exploring multiple possible solutions rather than converging on a single correct answer.

Research from the Wharton School at the University of Pennsylvania revealed that GPT-4 generated raw product ideas at roughly 40 times the speed of human university students, with external panels rating the top generative concepts significantly higher in purchase intent. However, a 2023 Harvard Business School study involving 758 Boston Consulting Group consultants demonstrated that while AI access boosted creative task performance by 40%, individual participants working without forced divergence prompts produced outputs with 41% higher collective similarity. You can examine this dynamic in detail through research published by Harvard Business Review, which tracks how unguided prompting narrows collective solution spaces. Without rigorous lateral constraints, automated tools default to statistically central, homogenized suggestions.

The agenda and facilitator prompts in this guide draw on cognitive psychology and classic innovation literature to force the model—and your team—away from those central tendencies.

Featured image by Ro han on Pexels