Using AI to make sense of nearly 8,000 residents' ideas for a 25-year community plan

The problem

Bowling Green, Kentucky, and the surrounding Warren County are growing quickly. The county's population is projected to reach about 233,000 by 2050, up from roughly 143,000, a growth rate local leaders see as a once-in-a-generation planning challenge. As Judge Executive Doug Gorman put it, growth can either “happen to us, or for us.”

To prepare, Gorman launched the BG 2050 initiative at the start of his first term in 2023 to set out a long-term vision for the area. It drew in around 100 community leaders from across the county, who worked through eight themes: economic development, housing, infrastructure, public health, quality of life, storytelling, talent development, and tourism. That produced useful input, though from a fairly small group.

The harder task was hearing from far more people and making sense of what they said. Public consultations can produce large volumes of open-ended feedback, and that is usually where they stall. Sorting thousands of written responses by hand and drawing out themes is slow, often taking days or weeks. The patterns that matter most, where residents agree, where they differ, and what they care about, are hard to pick out reliably at high volume. For local governments, the bottleneck is rarely gathering views. It is turning them into something they can act on.

The solution

Bowling Green set out to close that gap and to open the conversation to many more people at the same time. Its approach paired two tools: one to gather views at scale, and one to make sense of them. Working with Innovation Engine, Google's technology incubator Jigsaw, and the non-profit Computational Democracy Project, the county government built the “What Could BG Be” website.

The site invited anyone who lives, works, or spends time in Warren County to answer one open question: what do you want to see in the community over the next 25 years? Residents could post their own ideas and vote on others' ideas. The voting ran on Polis, an open-source platform for large-scale online deliberation. The conversation was open from 14 February to 17 March 2025.

After it closed, Jigsaw's Sensemaker tool, which runs on Google's Gemini AI, processed the responses. It grouped them by theme, highlighted the most prominent ideas, measured how much agreement or disagreement each attracted, and produced a web report written to be readable by planners and the public alike. Work that a team doing it by hand would have spent days or weeks on was done in minutes. It was Jigsaw's first large-scale use of Sensemaker in a live public consultation.

The impact

1. Nearly 8,000 residents took part in a single month
The website reached far beyond the original 100 volunteers. Nearly 8,000 residents, about one in ten Bowling Green residents, took part, posting 3,940 ideas and casting more than a million votes on each other's suggestions over a single month. Gorman said the project had “blown up so much bigger than I ever imagined.” Sensemaker sorted the ideas into 12 broad topics and around 70 subtopics.

2. Residents agreed far more than they disagreed
The analysis found broad common ground. Over half the ideas drew agreement from 80% or more of those who voted on them. Among the most widely supported were expanding local access to specialist care so residents need not drive to Nashville, protecting historic buildings, and making fuller use of existing commercial space before putting up new buildings. It was noted that public debate often rushes to find points of disagreement when there is plenty to act on that people already agree on.

3. The analysis took minutes and saved an estimated month of work
Sensemaker turned the full set of responses into a structured, navigable report in minutes. The project team estimated the tool saved 28 days of analysis, synthesis and writing on average. The degree of agreement across the data would have been hard to see at all without this kind of computational analysis.

4. The findings are starting to shape decisions
The 25-year roadmap is still being written, but some results are already in use. Gorman said riverfront redevelopment emerged as a clear priority, and on that basis, the county approached the federal government for funding. In surveys afterwards, 83% of participants said they believed the exercise would lead to a better plan, and 96% of local leaders said the report made it easier to represent their community.

Lessons learned

• The bottleneck was analysis, and that is what the AI removed. Bowling Green already had ways to gather public input. What it lacked was a way to process a large volume of open-ended responses fast enough to be useful. Sensemaker did not change how residents took part. It changed how their feedback was handled once it arrived. For governments that already run consultations but get stuck at the analysis stage, AI may add the most value.
• Local knowledge made the AI useful. The tool could sort and summarise thousands of responses in minutes, but turning that into something actionable still took people who knew the place. Local volunteers supplied context that the data did not carry, such as why riverfront ideas mattered, given developments already underway. Jigsaw described keeping a person in the loop as the surest path to useful insight.
• The exercise surfaced an agreement that usually stays hidden. When feedback is read aloud, attention tends to gravitate towards the loudest or most contentious voices. Measuring agreement across all responses showed that residents shared far more common ground than a typical public meeting would suggest, giving planners a base of broadly supported ideas to build on.
• The findings are an input, and the plan still has to be written. The report provides a detailed picture of what residents want, but translating that into planning decisions is still in its early stages. Local leaders have committed to working with the feedback, and the value of the exercise will ultimately rest on what they do with it.
• Open tools and open data kept the process accountable. Both Polis and Sensemaker are open-source, and the project published its full dataset for anyone to examine. The report also marks which summaries were written by AI and notes that they should be checked before use. For public trust in AI-assisted engagement, being able to see the inputs and the method matters.

This case study was written with assistance from artificial intelligence.