Ideation & Brainstorming
Overview
This is the phase where AI is most misused. LLMs are trained on the statistical average of human output. When you ask "give me 10 ideas," you get the 10 most expected, most averaged, least surprising ideas possible. That is not ideation - that is retrieval. Real ideation requires divergence, discomfort, and unexpected connections. Use AI as a provocation engine and a clustering tool, not as the idea generator. The human creative phase must come first.
AI workflow
7 steps
- 01
Human-First Brainstorm
Whiteboard / Miro
Run a 25-minute human-only ideation session before touching AI. Individual ideation first (10 min), then share and build. This prevents the AI anchoring effect where everyone defaults to the first plausible AI suggestion.
- 02
Cross-Industry Provocation
Claude / ChatGPT
Prompt: "How would a hospital solve this problem? How would a game designer approach this? How would this work if the constraint [X] did not exist? What would the most counterintuitive solution look like?" Use these as provocations, not final ideas.
- 03
Divergent Prompt Sequences
Claude
Use chained "yes and" prompting. Start with a concept, then: "Yes and - take this further." Then: "Now invert the core assumption." Then: "What would the version for a completely different user segment look like?" Build idea trees, not idea lists.
- 04
Post-Session AI Clustering
Miro AI
After the human session, paste all sticky note content into Miro AI for automated thematic clustering. This surfaces patterns humans miss when they are close to the material. Treat as input to discussion, not final groupings.
- 05
Concept Stress-Testing
Claude
For the top 3-5 concepts: "What are the three most likely failure modes of this concept? Who would hate this and why? What assumptions must be true for this to work? What is the cheapest way to test the riskiest assumption?"
- 06
Concept Naming and Framing
Claude
Generate 10 names and one-sentence descriptions for each concept. Good naming accelerates stakeholder alignment. Prompt: "Name this concept in a way that makes the value proposition immediately clear. Avoid jargon."
- 07
Decision Matrix
Claude + Notion
Evaluate concepts against: user value, technical feasibility, business viability, time to prototype. Claude can populate the matrix from concept descriptions. Human team finalises scores and selects direction.
Tools
| Tool | Use | Free tier | Client-safe |
|---|---|---|---|
| Claude | Divergent prompting, stress-testing, concept framing | Free tier available | Yes |
| ChatGPT | Cross-industry analogies, unconventional reframes | Free tier available | Yes |
| Miro AI | Post-brainstorm clustering, affinity mapping | 3 boards free | Yes |
| QoQo | Persona and journey map generation to fuel ideation | Free tier available | Yes |
Constraints
4 constraints
- Honest Assessment: AI ideation produces competent, safe, expected ideas. It is optimised for the average. If your goal is truly novel concepts, use AI for provocation and stress-testing, not primary generation.
- Anchoring Risk: If you show AI ideas to a team before human ideation, the team will unconsciously anchor to those ideas. Always human-first.
- AI ideation must follow, not replace, human creative sessions in any client-facing deliverable.
- Monitor for premature convergence: AI naturally narrows options. Build in explicit divergence gates before any convergence moves.