Maintain a what-we-heard / what-we-changed prototype log
Pair AI-assisted prototype drafts with an auditable record of de-identified input, participant feedback, changes, rejections, and accountable ownership.
Why
The workflow problem
AI-generated civic or participatory prototypes can obscure who shaped what, why a change was made, and whether dissent or minority input was lost.
Fast provisional visuals and narratives can make feedback more concrete, while a public-facing trace makes influence and non-adoption inspectable. The trade-off is privacy and performative transparency: a log can expose participants or imply consensus where disagreement remains.
Concrete output
What you will produce
An auditable prototype package containing a de-identified input brief, versioned drafts, what-we-heard / what-we-changed log, rejected suggestions, decision rationale, and ownership record.
How
Run the workflow
Approve a de-identified input brief
The facilitator separates consented, de-identified input from restricted material and records the scope, source groups, unresolved disagreement, and publication boundary. A privacy or community-governance owner approves what may enter the tool and log.
Create a deliberately rough draft
Use AI to produce a provisional visual or narrative only from the approved brief. Tag it with version, inputs used, generation date, and limits so it cannot be presented as community consensus.
Draft from the approved brief only
Create a deliberately rough prototype draft from the approved, de-identified brief below. Brief: [insert] Input groups and stated differences: [insert] Required uncertainty labels: [insert] Prohibited representations or details: [insert] Return the draft plus: source-brief sections used | assumptions | questions for participant review. Do not invent community views, personal details, quotes, or claims of consensus.Record feedback, change, and rejection
Participants critique the draft and can propose changes in their own words. The facilitator logs each material input as heard, changed, deferred, or rejected, with a reason, owner, and any dissent preserved without identifying individuals.
Publish or share the trace responsibly
The accountable owner checks for re-identification, misrepresentation, and unsafe disclosure, then shares the appropriate prototype version and log. Keep a non-public audit record where public disclosure would create risk.
Human–AI partnership
Who contributes what
AI contribution
Produce fast, provisional visual or narrative drafts from an approved brief. It does not represent participant voice, determine consensus, or decide which feedback matters.
Human responsibility
Protect voices and data, define consent and publication boundaries, facilitate power-aware decisions, preserve dissent, document changes, and own what is published.
Stop rule
Stop AI use if output is presented as community consensus, participant contributions cannot be traced to a governed record, or public logging creates a material identification or safety risk.
Check before use
Watch out
Visibility is not the same as accountability. Check who can access the log, who can challenge it, whether dissent is being compressed, and whether visual realism could be mistaken for an agreed plan.
Evidence base
What supports this workflow
A named civic sprint documents a five-step legible-AI process with concrete outputs and qualitative outcomes. It supports the value of traceability in a participatory context but does not establish general effectiveness or resolve privacy trade-offs in other settings.
- Steve Diasio · Prototyping Civic Futures with Legible AI
Direct harvested civic-sprint case supporting a what-we-heard / what-we-changed approach and legible prototype outputs.
- IDEO.org · Build & Run Prototypes
Canonical method context for making a prototype, gathering feedback, and iterating a human-centred design intervention.