How We Listened
We took it to the Juneteenth Family Reunion, Get2KnowNoke, and Hoot ‘n Holler. We sat down with arts and cultural leaders in roundtables. We shared experiences at story circles and had conversations at GO Fest.
The window
The survey collected 617 responses. A separate idea form, where anyone could submit a single idea without answering anything else, collected 61 responses. Of the 617, 346 people took the full survey and 271 took the short version. The short version asked for ratings and rankings but no writing.
How people found out about it
Outside of in-person events, we hosted 26 Days of Plan 26, a post a day about the plan and the people in it. It drew almost 95,000 views across Facebook and Instagram. As part of this effort, we partnered with RVTV to produce five videos about the impact of an arts and cultural plan:
Plan 26 was also covered by the media, including:
- WDBJ7: Roanoke Arts Commission seeks community input on ‘Plan 26’ arts and culture strategy
- WFIR: As Roanoke Arts Commission crafts “Plan 26,” they ask residents to weigh in via an arts and culture survey online
- Valley Business FRONT: There’s Something Happening Here
How the input was read
617 responses and 61 ideas is more text than a volunteer commission can read closely in the time available, and skimming it would have meant the longest and loudest answers set the agenda. So the reading was done with AI assistance, under a method designed to be checked rather than trusted.
- Step 1. Every response was read line by line and each distinct ask was pulled out in the respondent’s own words. Nothing paraphrased or polished. Each response was read twice, independently, so that a single pass could not lose something.
- Step 2. Anything that could not be traced was thrown out. Every extracted item had to match the respondent’s actual text exactly or it was deleted automatically. 0.4% failed that check.
- Step 3. Asks were separated from obstacles. The reading produced 1,916 recommendations (things somebody wanted done) and 649 barrier reports (things stopping people from taking part). Only recommendations become policy. The barriers explain why a priority matters and how it fits into the plan’s structure.
- Step 4. Duplicates were merged, leaving 1,059 distinct recommendations from 337 people.
- Step 5. Recommendations were grouped into 457 distinct ideas, and those into 63 recurring types, things like affordable studio space or free and low-cost youth arts classes.
- Step 6. Each type was placed into a chapter by the barrier it removes, as the respondent described it, not by what it is about. Most sorted cleanly, but any that sat between two rules were decided by a person.
Several different AI models did the reading, blind to each other’s work, so that agreement between them meant something. Where they disagreed, a person ruled. Where they agreed but the item sat on a boundary, a person ruled anyway, because agreement is not the same thing as being right.