Choose one bounded AI task in each design phase
Use a per-phase support menu for drafting, organising, or planning while people keep creativity, interpretation, research, and decisions.
Why
Why it earns its place
A small task menu is easier to govern than an open “help with design” request. It also exposes where AI would replace evidence or judgement.
How
Run it
Complete the human input
Gather real evidence, frame the current question, or create initial ideas before choosing AI support. Do not use synthetic people as research.
Pick one support task
Choose drafting questions, provisional organisation, workshop planning, method options, test-plan structure, or summarisation—not final insight or selection.
Bound the request
Supply the phase, verified input, constraints, output format, prohibited inferences, and the check that will follow.
Verify and decide
Trace factual claims, compare against the original material, repair omissions, and record what people accepted or rejected.
Ready to paste
Prompt
Current design phase: [Empathize, Define, Ideate, Prototype, or Test]
Bounded support task: [one drafting, organising, or planning task]
Verified input: [sanitised text, facts, or existing ideas]
Constraints: [constraints]
Human-only decision: [decision the model must not make]
Required check: [how people will verify]
Complete only the bounded task using the supplied input. Separate direct input from your suggestions. Cite input item IDs for summaries or interpretations. Mark missing context and uncertainty. Do not create persona facts, user needs, research findings, test results, or claims of effectiveness. Return the requested output plus a short “human checks required” list.- Which part of your output is suggestion rather than supplied evidence?
The AIdea layer
The designer’s call
- Human core
- People gather evidence, interpret context, originate creative direction, test with users, and make every consequential decision.
- The trade
- Smoother drafting and structure ↔ fabricated context and weaker creativity. Fix it with one bounded task and a human-first input.
- Skip AI when
- Skip AI when it would impersonate users, replace fieldwork, or perform the skill the learner is meant to practise.
Check before use
Watch out
Keep personal and confidential data out of general chatbots. Phase labels do not make output reliable; inspect source coverage, invented persona details, shallow summaries, and generic ideas before use.
Source of truth
Evidence
Bureau Sprankel recommends per-phase support and repeated human checking; the agency blog reports no classroom trial, client case, timing, or measured outcome.
Marije van der Star · Hoe ChatGPT jouw Design Thinking-proces kan boosten