The town wasn't as divided
A Town Looks at Itself: What Bowling Green Found Under the Noise
In February 2018, Americans were about as sure of their divisions as they had ever been. Two countries on one map, red and blue, neighbors who supposedly could not stand each other. So when the local newspaper in Bowling Green, Kentucky, a city of around sixty-five thousand, invited residents to say online what they thought should change about their town, the reasonable expectation was a fight that would split along the usual lines.
It did not turn out that way, and the reason it did not is the point worth taking from it. The exercise ran in three parts over about three weeks: first an online phase using the listening tool Polis, then a town hall in person, then a community workshop. It was organized by the Bowling Green Daily News together with the American Assembly at Columbia University and the media researcher Sam Ford. The online prompt was deliberately open. What should change in Bowling Green and Warren County to make it a better place to live, work, and spend time. Anyone who lived or worked there could add short statements and vote agree, disagree, or pass on everyone else's. The turnout was unusually high. Around two thousand people took part, roughly three percent of the population, where a deliberative exercise of this kind might normally draw a tenth or a third of one percent. Between them they wrote some six hundred statements.
What came back was not a battlefield. Of the hundreds of statements, only a small number produced any real disagreement. The rest, on traffic and development and internet access and parks and the cost of cable and the openness of city government, drew broad and rather boring agreement, some of it with only two or three percent of participants objecting. The organizers had braced for a brawl and kept remarking on what they found instead. The things people chose to raise were mostly the things they agreed about. Disagreement was real, but it was the exception rather than the rule.
The point here is not the comforting one, that underneath it all we are really the same. People in Bowling Green disagreed about plenty, then and now. The sharper point is that how divided a community feels depends a great deal on the instrument through which it encounters itself, and we rarely notice this. National news and social platforms are built to surface the small fraction of things people fight about, because conflict holds attention and travels well. The large majority of ordinary shared concern is invisible to them, not because it is absent but because it does not perform. Give the same people a different instrument, one built to surface agreement rather than reward outrage, and the proportions invert. The neighbors are the same. The machine showing them to each other is not, and so the picture is not.
None of which means the division is fake. It means that much of what we take to be the basic temperature of a place is really a property of the rooms we have chosen to argue in. Change the room and the reading changes.
There was a practical result as well, beyond the better mood. Because the agreement was written down and laid out plainly, residents had something concrete to hold their officials to. Not a vague sense of the public will, and not a poll a politician could wave away, but a documented account of what the town itself had said it wanted. A record like that is harder to ignore. The approach worked well enough that it was tried again elsewhere, including in Louisville.
There is a postscript worth mentioning for anyone curious where this leads. The Bowling Green conversation became a kind of reference case for researchers. The Computational Democracy Project, which builds Polis, worked with Anthropic to use this very dataset to test whether a language model could take on some of the slow work of summarizing these conversations, drafting readable reports in minutes, while keeping people in the loop to catch the machine's mistakes. The conversation that let a town hear itself also became a place to learn how to do that hearing faster and more cheaply, without handing the judgment over to a machine.
For the same thing at national scale, with a government on the other end of it, see Taiwan's fight over Uber. Bowling Green does not prove that division is an illusion. It suggests that a good deal of it is manufactured by where we choose to fight, and that the common ground we assume is gone is often just somewhere we have not yet built a way to look.
Sources and further reading: the American Assembly's account of the Bowling Green Civic Assembly (https://americanassembly.org/projects/bowling-green-civic-assembly); the Bowling Green Daily News on the consensus the assembly revealed (https://www.bgdailynews.com/news/civic-assembly-reveals-consensus-on-many-issues/article_837ada9d-f281-516a-9a01-7b592d2b2397.html); the Computational Democracy Project's case study (https://compdemocracy.org/case-studies/2018-kentucky/); and the research that later used this dataset, Opportunities and Risks of LLMs for Scalable Deliberation with Polis (https://arxiv.org/abs/2306.11932).