AI-Clustered Consultation (Talk to the City)
People answer an open question in their own words. An AI pipeline extracts the individual arguments, groups similar ones into clusters and sub-clusters, and links each back to the original quotes. The organisers check the clusters against the quotes, share the map back with participants to confirm they recognise their views, and use it to structure a live discussion or decision.
Use when you have far more open-ended input than any group could read, and need to see the full range of views without reducing it to a multiple-choice poll.
Solves: Large consultations collect thousands of comments that nobody reads, so the summary reflects whoever wrote the report.
- 1
Step 1: Write one open, neutral question and collect free-text answers or interviews over one to three weeks.
- 2
Step 2: Run the responses through the clustering tool to extract arguments and group them.
- 3
Step 3: A small team checks clusters against the original quotes, fixes labels and splits merged views (60-120 min).
- 4
Step 4: Share the map with participants and ask: is your view here, and is it described fairly? Adjust (async, one week).
- 5
Step 5: In a live session, walk through the main clusters and points of agreement, and choose which disagreements to work on (60-90 min).
- 6
Step 6: Publish the final map with the decisions taken.
- Always let readers click from a cluster to the real quotes. That is what makes the map trustworthy.
- Budget human time for checking; the authors say current models are not good enough for fully automatic processing.
- Watch for nuanced answers that the AI split into contradictory pieces.
- Use it on transcripts from several parallel table conversations to merge what was said in each.