Grow the idea pool after a human-only round
Generate ideas independently first, then use AI to search for genuinely different mechanisms before the team curates one combined portfolio.
Why
The workflow problem
AI can increase individual idea output while making different people's ideas more alike and weakening their sense of ownership.
A device-free opening round protects independent thinking and makes the team's starting patterns visible. AI enters later as a bounded contrast generator, not as the author or judge of the solution space.
Concrete output
What you will produce
A deduplicated idea portfolio with human-origin and AI-origin labels, mechanism-level clusters, a diversity check, and a human-authored shortlist.
How
Run the workflow
Generate independently
Close every AI tool. Each person responds silently to the evidence-backed How-Might-We question, one idea per note. Share only after everyone has produced a starting set.
Map the mechanisms already present
Cluster the human ideas by the change mechanism they use, not by wording or visual polish. Name each cluster and identify needs, constraints, or perspectives that are still missing.
Ask AI for deliberate difference
Give AI the anonymised challenge, constraints, and mechanism map. Request alternatives that avoid the existing mechanisms. Treat every answer as a provisional stimulus, not a recommendation.
Generate contrasting mechanisms
Act as a bounded idea-expansion partner. Our existing mechanisms are an exclusion set, not a style to imitate. How-Might-We question: [insert question] Intended users: [non-identifying description] Verified needs and constraints: [insert evidence-backed list] Mechanisms already present: [paste cluster labels and definitions] Generate 12 one-sentence candidates that use materially different mechanisms or user moments. Return: candidate | core mechanism | closest existing cluster or “none” | exact difference | assumption that must be tested. Replace near-duplicates before answering. Do not rank, claim an idea will work, or invent user research.Curate and decide as a team
Add AI candidates in a different colour, kill paraphrases, and rewrite survivors in the team's own words. Humans apply the same evidence, desirability, feasibility, and responsibility criteria to the combined pool.
Human–AI partnership
Who contributes what
AI contribution
Propose mechanisms outside the visible pool and make assumptions explicit. AI supplies contrast, not validated needs or decisions.
Human responsibility
Frame the challenge from real evidence, produce the first-round ideas, protect minority directions, judge meaningful difference, and make every keep, combine, test, or reject decision.
Stop rule
Stop using AI when an independent first round is incomplete, suggestions converge on the same mechanism, or generated polish begins to outrank user evidence and accountable judgement.
Check before use
Watch out
Do not paste identifiable research data into an unapproved tool. Keep provenance visible and inspect whose needs, cultures, abilities, and constraints remain absent.
Evidence base
What supports this workflow
Peer-reviewed studies support the individual-output versus collective-diversity trade-off. They support a human-led, reflective sequence more strongly than model-led rewriting; this exact workflow remains an AIdea synthesis.
- Anderson, Shah and Kreminski · Homogenization Effects of Large Language Models on Human Creative Ideation
Controlled study showing more individual ideas alongside greater similarity and lower ownership across people using ChatGPT.
- Anil R. Doshi and Oliver P. Hauser · Generative AI enhances individual creativity but reduces the collective diversity of novel content
Peer-reviewed evidence for the individual-quality versus collective-diversity trade-off.
- Stanford d.school · Idea Expedition
Canonical guidance for divergent ideation and disciplined selection.