Co-ideate through present, past, and future questions
Give a custom GPT your design problem, then let it interview you one question at a time before you shape and judge concepts.
Why
Why it earns its place
Structured how, why, and what-if questions can deepen a designer’s own reasoning instead of replacing it with an instant idea list.
How
Run it
State your problem first
Write an 80–100-word, sanitised design problem plus verified user evidence and constraints. Add your own current observations before opening the dialogue.
Let AI interview you
Ask for one question per turn: current experience first, then relevant past experience, then a future what-if. Answer from evidence or label uncertainty.
Push beyond the first thread
When the exchange circles or repeats, request a tangential what-if. Keep distinct branches visible rather than polishing one answer too early.
Shape and judge concepts
Ask for a provisional synthesis only after the interview. Humans edit, sketch, test feasibility, compare variety, and make every keep or kill decision.
Ready to paste
Prompt
Act as a questioning co-ideator for one designer. Do not begin by proposing solutions.
Design problem: [80–100 words]
Verified user evidence: [sanitised excerpts with IDs]
Constraints: [insert]
My current observations: [insert]
Ask only one concise question per turn. Start with my present experience of the problem and use how or why probes to uncover mechanisms, not motives you invent. Then ask about relevant past experiences. Finally, introduce one plausible what-if about a preferred future.
Use my answers and evidence IDs only. If I cannot support a claim, record it as an assumption. Do not impersonate users, fabricate research, or force agreement. If my answers repeat, ask a tangential question and preserve the abandoned branch.
Only when I type SYNTHESISE, return five provisional concepts with: mechanism | evidence IDs | assumption | feasibility question | difference from the other concepts. Do not rank them or claim they will work.- Ask one tangential what-if without proposing its answer.
- Which branch did we abandon too early, and what question would reopen it?
The AIdea layer
The designer’s call
- Human core
- The designer supplies experience, interprets evidence, authors the concepts, and decides what survives; AI controls only the questioning scaffold.
- The trade
- Deeper reflection ↔ repetitive dialogue and no proven variety gain. Keep branches visible and compare outputs before converging.
- Skip AI when
- Skip AI when the designer cannot verify its questions against evidence or when the custom configuration has not been tested safely.
Check before use
Watch out
Sanitise research before use. A custom GPT can repeat familiar patterns, misunderstand the brief, or lose context; keep evidence IDs, save branches, inspect variety manually, and stop if questions become leading or invented.
Source of truth
Evidence
In a randomized 48-student Delft study, custom-GPT co-ideation scored higher on novelty and feasibility-based quality, not variety; effect magnitude and broader transfer remain unknown.
Pan Wang, Yash Khinvasara, Geesje Josine Creijghton, Tessa Scholing, Yihua Wang, Zhibin Zhou, Peter R. N. Childs, and Yuan Yin · Enhancing designer creativity through human–AI co-ideation: a co-creation framework for design ideation with custom GPT