Co-ideate through a fixed question protocol
Use a question-led AI exchange to develop a designer-authored concept from an evidence-backed How-Might-We question.
Why
The workflow problem
Open requests to “give me ideas” produce shallow lists rather than a traceable process of reflective concept development.
A fixed questioning protocol can create productive prompts for reflection without handing concept authorship to the model. Its trade-off is that a coherent transcript can conceal repetition or an unexamined model frame.
Concrete output
What you will produce
A concept sheet showing the question trail, alternatives considered, evidence constraints, and designer-authored concept rationale.
How
Run the workflow
Frame the design territory
The designer writes one evidence-backed How-Might-We question, intended user, non-negotiable constraints, and decision horizon. A reviewer confirms that no preferred solution is embedded.
Start the question-led exchange
The designer asks the configured co-ideator to ask one question at a time. Answer in notes or sketches; redirect questions that assume facts not in the brief.
Open the co-ideation dialogue
You are a co-ideation partner, not a concept author. Ask exactly one concise question at a time. Move across present, past, and future perspectives; occasionally ask for an analogy, a constraint reversal, or an overlooked stakeholder. Design question: [insert evidence-backed HMW] Verified context and constraints: [insert] Do not propose a finished solution, invent research, rank options, or ask more than one question in a turn. Wait for my answer before the next question.Make and branch concepts
The designer turns useful answers into at least three rough concepts, marking which decisions came from evidence, professional judgement, or an assumption. Compare branches for genuinely different mechanisms rather than different wording.
Challenge the concept trail
A human reviewer reads the sheet without the transcript first, then checks whether the rationale is intelligible, evidence-linked, and not dependent on model authority. Revise in the designer's own language.
Select a next experiment
Humans choose one concept and a proportionate next test. Record rejected alternatives and the uncertainty that the test must resolve.
Human–AI partnership
Who contributes what
AI contribution
Supply paced perspective questions and occasional reframing prompts. Its questions are stimuli, not evidence, requirements, or a decision procedure.
Human responsibility
Supply domain knowledge, answer critically, sketch and author concepts, check evidence, and select the concept and experiment.
Stop rule
Stop the exchange when the model begins deciding, repeats a pattern, introduces unsupported facts, or the designer cannot explain the concept without relying on the transcript.
Check before use
Watch out
A question can still steer the work. Inspect loaded assumptions, culturally narrow analogies, and false trade-offs. Do not use confidential research or identifiable participant material in an unapproved system.
Evidence base
What supports this workflow
A controlled study with 48 design students evaluated a reusable co-ideation protocol; it is promising but comes from one educational setting. The protocol supports structured reflection, not a claim that AI-generated concepts are better in professional practice.
- TU Delft, The Hong Kong Polytechnic University and Imperial College London researchers · Enhancing designer creativity through human-AI co-ideation
Controlled classroom study (n=48) and source of the fixed opening co-ideator protocol.
- Stanford d.school · Idea Expedition
Provides canonical divergent-thinking and concept-selection method context.