There are two ways to put AI to work on local news. You can point it at what other people have already published, or you can point it at the town.
Patch, the national hyperlocal news company, chose the first, and chose it at scale. Its AI newsletter, Patch AM, "has been activated in fourteen thousand communities and has accrued nearly a million subscribers," the Columbia Journalism Review reported in March.
How much of that newsletter does a person touch? Peter Kafka put the question to Warren St. John, Patch's chief executive, on his media podcast, Channels, in September 2025. St. John said Patch had about 85 full-time reporters. Kafka asked at what point a human gets involved before the newsletter lands in your inbox.
"Not at all," St. John said.
Meanwhile, in Farmington, Connecticut, the police department takes its arrest log down every week, and as far as we can find, the only lasting copy is ours. Those two facts are the whole argument.
What Patch built
The machine reads town websites, social media and other outlets' stories, picks what looks newsworthy, and links out to whoever did the reporting. Kafka described it plainly at the top of the show: "one thing AI is quite good at doing right now is summarizing what's already on the web, and that's what this is."
St. John's own metaphor was food. A town with a Patch reporter gets "the five-course meal." The AI newsletter, he said, "is really a kind bar," as in the KIND snack bar. "It isn't actively bad for you."
He was candid about the seams, too. Plenty of American towns share a name, and the machine still mixes them up. "It's the wrong Springfield" is a complaint Patch hears "at least single-digit times a week," he told the Review.
And he named the deeper weakness himself. The newsletter runs on other newsrooms' work. "If local news goes away, that's bad for this product," he told Kafka.
That is the whole problem in one sentence. A summary of the local paper is only as good as the local paper, and the local paper is the thing that is disappearing.
The record nobody else is keeping
I publish a group of local papers on a platform I built, Mercury Local. One of them is The Farmington Mercury, in a Connecticut town where Patch also has a site. We lean on AI hard, and we have written down exactly what it does. The difference is what we point it at. The machine starts from the public record itself: the police department's arrest log, the recordings of town meetings, the reports committees hand the council. Nothing goes out until I have approved it, and every fact we publish is tied to the document it came from.
Take the arrest log. The Farmington Police Department posts one log for each weekday and replaces it the following week. The town keeps no archive. Since the start of September we save every one of them each morning, including the blank days, and we write them up. As far as we can find, ours is the only archive.
That is the part a summarizing machine does not do, however many towns it covers. The AI assistant on your phone can already summarize a news website. It cannot summarize an arrest log the town took down last week, unless somebody kept it.
When the log came back empty three days running at the start of September, we published that too: Seventy-Two Hours With Nothing on the Page. St. John described the opposite instinct. In a quiet town, he said, "we don't send you a blank email." I understand why. But a blank log is still a fact about a town, and a paper that only shows up when something happens is not keeping the record. It is waiting for one.
Even the staffed version
Farmington has a Patch reporter, which makes it the fairest test there is: Patch with a person in it, against our one town.
On September 8, Farmington's town manager brought the Town Council a sidewalk committee's report. About 48 miles of sidewalk. About $18 million to bring all of it up to the town's standard.
Our story ran the next day, and it led with what that number means for the people reading it: the bill is not the town's. Under Farmington's ordinance, the owner of the house repairs the sidewalk out front and answers for anyone who falls on it. The story carried the committee's A-through-F grades, its comparison against twelve other Connecticut towns, and the finding that Farmington has not run a routine sidewalk inspection program in 25 years.
Patch covered the same report on September 23. Its story was accurate and hit the main numbers.
The motto printed at the bottom of our Farmington stories is "Always last to breaking news." We mean it as a promise to check the arithmetic before we publish. On the sidewalk report we were first by two weeks.
Where he is right
St. John said something on that podcast I agree with completely. With machines able to summarize anything, a reporter's value now lies in "taking things that aren't available on the Internet, that nobody knows, that nobody can access."
Yes. That is the whole job. His AI newsletter is built from the other half, the part that is already on the internet.
The strongest case for Patch is reach. One sign-up, and a newsletter exists for a town that had nothing. A snack in fourteen thousand towns may beat a meal in a handful, and I will not pretend we are about to match that number. But the cost of opening a real paper keeps falling. The East Hampton Mercury, also in Connecticut, went from idea to a live paper in a single day, on September 24.
Patch taught a machine to read the news. We taught ours to read the town. When the paper down the street closes, the town is still there.