Patch’s AI Experiment: Thousands of Newsletters, Zero Humans

0:00Peter Kafka:Support for the show comes from Injun. Running a small business means every dollar has to work hard. But if your team is still booking travel the old way, it's costing you more than you think. Injun is the fastest growing travel and spend platform in the country, built specifically for businesses like yours. Book a trip in as little as two and a half minutes. Earn up to 10 % back on hotels. And in 2025, Injun customers saved more than$300 million on travel. with zero booking fees, no contracts, and no BS. More than 1 ,000 businesses join Engine every month. Join them and get$500 when your business signs up and starts traveling at engine.com slash Vox.

0:48Peter Kafka:From the Vox Media Podcast Network, this is Channels with Peter Kafka. That is me. I'm also the chief correspondent at Business Insider. And today we're talking about AI, of course, because it's 2025, specifically about AI and media, also because it's 2025. And we've gone from will media companies use AI to make content to here's how media companies are using AI to make content right now. I got to this one because of an earlier conversation on the show with Henry Blodgett. A lot of you like that one. Thank you for letting me know. And Henry was speaking approvingly about local news companies using AI.

1:26Peter Kafka:And that led me to look up Patch, the local news company, which earlier this year announced it was going to start covering news in thousands of communities around the country solely using AI. Zero humans involved. It may be helpful as you're listening to this conversation with CEO Warren St. John to actually see the product he's making with AI. You can sign up for it by Googling Patch AM. But even if you don't, you can kind of imagine what an AI-made newsletter looks like, because one thing AI is quite good at doing right now is summarizing what's already on the web, and that's what this is. So is this the future of media?

2:05Peter Kafka:It's certainly the future of a certain kind of media. And that's one of the things Warren and I talk about. Take a listen. Let me know what you think.

2:15Peter Kafka:I'm here with Warren St. John. He is the CEO of Patch. Nice to chat with you again, Warren. Hi, Peter. Good to see you. I think the last time we talked at any length was 2019. I wrote an article about how Patch, which had been this pioneering attempt to make local work and then kind of didn't work, was now profitable. And we're talking today because you guys are, is leading the charge the right term? I'm going to say it. I'm going to say you're leading the charge into AI-generated news. Are you comfortable with that description? Well, with a lot of nuances around that, I think, you know. It's a podcast.

2:53Peter Kafka:We got time. Let's do it. Yeah, great. So let's talk about what you're doing, why you're doing it, and how it's working. So the first thing we're doing is while we're experimenting and launching AI-powered products, we're continuing to do local reporting and hyper-local reporting the way we always have. So we've got our newsroom of 85 or so full-time reporters doing original reporting against the communities that more or less that we inherited from AOL when we spun the business out over a decade ago now. Yeah, we should do a tiny bit of background, right? So this was Tim Armstrong's baby. Prior to running AOL, he ran AOL, he bought Patch, did a big push into local, and then had to scale it back and eventually sold the thing to you folks.

3:45Peter Kafka:That's right. Which is still Hale. Correct. Hale Global and Verizon retains a minority stake, a sizable minority stake. So we inherited from them coverage of around 1 ,000 communities. And communities is, in hyperlocal, it might be a neighborhood. I think there are 29 New York City patches, what we call a patch. It's a local site. It's got its own local calendar. It has its own local coverage. We share coverage between communities. So, you know, if there is a fare increase, you know, on the subway or a blizzard in New York City, we're not going to do 29 versions of that story. We're going to do a single version and we're going to distribute it to the right geographical, geographically relevant community.

4:31And that's how you can do a thousand communities with 85 people. That's correct. And those efficiencies exist in lots of places. The common example at the really granular level is every year there's a big football game between Brick, New Jersey High School and Tom's River. And so we're going to do one story on that game and we're going to put it on both of those communities. We're not going to do, you know, the Tons River version and the brick version, which, you know, in a perfect world, we would. In a perfect world, every town would have its own dedicated hyperlocal reporter. And we just, you know, we haven't found a model that supports that.

5:12Peter Kafka:Right. This is the challenge for local news in general, right, is that the size of the markets, whether it's, you know, some combination of advertising and paid subscriptions just doesn't support newsrooms really of any size in lots of communities. And people have been trying to solve this for probably a couple decades now. You guys are one of those attempts. But as you're explaining, like there's, you know, you guys have 85 people. You don't have a thousand people, let alone, you know, multiple people in multiple newsrooms. Correct. So you're doing this, you know, you're stretching the string as far as it'll go already.

5:48That's right. And we're trying to add to that. You know, the more efficient we get, the more we can invest in the interregional reporting. And I would say that, you know, that's really our mission. That's what we all signed up for. We're an organization that's committed to working for a sustainable model for local news. We don't take, there's a lot of, you know, amazing philanthropic opportunity out there for hyperlocal independence and that sort of thing. We don't go that route. We do it all on our own steam. And that's because it's just sort of ideologically important to us to pursue a sustainable model that doesn't require largesse from the outside.

6:28Peter Kafka:So talk about you had this model where and you'd been using computers to assist you with your work for years. Again, back when you and I talked about it, some of the stuff was was generated with software. You guys have tried various attempts to make this work. And then this spring, you said, we're now publishing, you tell me, how many newsletters using AI? So we can produce a newsletter now for basically every community in the United States, which if you really round up and you go by what the census-designated places and all of these locales out there, it's around 30 ,000 communities. Currently, we're live in about 11 ,000 of those.

7:13Okay, because the headline says 30 ,000. and I wanted to make sure that I got it right. So it's really 11 ,000. We're capable of 30 ,000. And the reason we're not at 30 ,000 is, first of all, we don't wanna produce 20 ,000 things for which there's no audience. And so the way the system works is when someone signs up for a town, so if you went to patch.com, you punched in the zip code or the name of a small town. If we had no subscribers for that community, we would then start creating upon the first sign up. and so right now we have we've gone from around 1100 to 11 ,000 communities since we launched this product which we call Patch AM and it's a newsletter first product it's available on the web but it's primarily you know meant for your inbox you might think of it as sort of morning brew for your town and it tells you the headlines relevant headlines for the day upcoming local events which it turns out is a big pain point for people.

8:13We have, we've scaled our local calendar infrastructure. So now we can create a local calendar and have for, you know, any community in the United States. So we obviously pull that data and we supplement it. So we're telling people what's coming up. We look at social media sources, we look at town government sources, and ultimately, you know, our, the path we're proceeding on is to try to find any digital information we can about a community and then put it through our system, which essentially filters on newsworthiness, interest, a number of other things to try to surface what is the most valuable information about your community.

8:50One thing we have learned is that people don't want to know everything about their communities. Because if you look at everything about your community, one of the things that you might learn is there's a parent-teacher conference night tomorrow night for sixth grade well that's relevant to somebody but it's not relevant broadly to the community so that's not the kind of information we would include and so we built a system that's really around the curation and that's what we're using ai primarily for is to curate information um along the lines of relevance newsworthiness and that sort of thing so let's

9:24Peter Kafka:let's stop there and just and just talk about what the product is so i've been able to look at a few of them they're they're very straightforward right they're literally what you just described Weather events, smattering of news, the news stories you're not writing, they're written, created by someone, by another publisher, right? If there's a local newspaper or local TV, et cetera, you link, you summarize and link out to them. So at what point, if any, does a human get involved in this process before it gets sent to my inbox? Not at all. Not at all? Not at all. Now, we spent a year as humans working on the product.

10:03And a lot of that is putting rules in place. First of all, simple guiding principle, the simpler something is, the easier it is to scale. So simplicity has been kind of a North star from the beginning. Complexity makes, creates lots of edge cases, and then you get something that's suboptimal. So we launched this product first in a town, then ultimately in Georgia and Virginia to our communities there, gathered a lot of feedback. We've gotten tens of thousands of user responses at this point. And, you know, there are all sorts of complexities that have to be dealt with. One is simply a third of U.S.

10:45towns share a name with another town in some other place. So you can't just filter on keywords or something. You've got to build a taxonomy that says these publications are associated with this town. At the same time, if the New York Times or Wall Street Journal or some other Reuters, some big national publisher writes about your town, well, we don't want to miss that either. So we need to have a system that can look more broadly as well. And frequently, by the way, you know, if the New Yorker comes to your town and writes an 8000 word feature on something about your community, that's the kind of thing everybody's going to be talking about.

11:21And ultimately, if we miss that, we haven't done our job.

11:25Peter Kafka:But yeah, you want to do that. You want to make sure it's Springfield, Illinois, not Springfield, Missouri. I think there's 17 Springfields. There you go. I think maybe more, but but you would know better than I. But but you guys got this software to a point where you can set it and forget it and and are comfortable going out without having any kind of human eyes looking over it. So that's one question. The second question is you make a great point. You make a point of saying over and over that you've got 85 human journalists working for you. This is not meant to replace their work. if this product is working and it's built on publicly available information other people's reporting why why not make this sort of the future of patch i know it's an uncomfortable thing to discuss but everyone who's listening to this podcast is thinking through this discussion one way or another as a consumer as a creator as a manager of a business why wouldn't you use ai if you could.

12:26Sure. So I think it's important to understand that it does one thing, it doesn't do other things. So, and I think maybe an analogy that seems to have resonated for some people I've had this discussion with, we might think of it as, you know, when you go to a restaurant. So if you're in a priority patch town with a local reporter who's doing multiple local stories a day, we might consider that sort of the five-course meal of hyper-local coverage. You're getting a lot of local attention. I happen to, I live on the Upper East Side of Manhattan. We have a very strong audience here, and so we're able to support that patch with a lot of coverage.

13:09But we do that in places like Bel Air, Maryland, Doylestown, Pennsylvania, and Joliet, Illinois, lots of communities, Temecula, California. Lots of communities around the country get the five-course meal. The AI newsletter is really a kind bar. It's got some basic nutritious kind of components. It's tasty. It's - Isn't actively bad for you? It's not actively bad for you. And it's dependably sort of packaged and delivered. And so, you know, the goal would be to build on top of that. It certainly doesn't, first of all, it depends on, to some degree, other local news sources. But what is, I think it's another advantage we have there is that at the hyper-local level, not necessarily at the metro level, but the hyper-local level, what is considered news is it's a pretty broad, you know, it's a broad category.

14:08So news might be there's a new house on the market in your town or a new business opened or there's a food fair on Main Street this weekend in a really small town where not many things happen. Those things are, you know, news. They're not, you know, a ProPublica investigation or a 4000 word feature. Right. But those things have been elements of local news forever. We have this sort of notion. I grew up in Birmingham, Alabama. the Birmingham, we used to have two papers, the Birmingham News and the Evening the Post-Herald, and they were very thick. But when I, and I've looked at old versions of the paper from when I was a kid, there was a great deal of information in those things that was just local announcements, the society column, all of that.

14:57Peter Kafka:We'll be right back with Patch CEO Warren St. John, but first a word from a sponsor.

15:06Excuses are easy. An epic movie night? We don't have enough snacks. Dinner party with the girls? We'd have to decorate. Surprise date night? Nothing to wear. But Amazon's prime same-day delivery lets you say yes before the moment slips away. Try that new popcorn maker. Order those cheeky drink glasses. Get that new perfume. And turn that I wish we could into an I'm so glad we did. Visit Amazon.com slash Prime to find millions of items delivered fast. Same day delivery. It's on Prime. Available in select areas. Terms apply.

15:47Peter Kafka:And we're back. The stuff that you're talking about, house for sale, new restaurants open, this is stuff that people are going to generate information about on their own, either because they have a commercial interest, right? They want to sell the house or they want to talk about the new restaurant or it's just something they keep track of on their own. And so there is a digital file somewhere that you guys can scrape and analyze and pass along to me. You are also dependent on local newspapers and TV stations. Those are obviously threatened day after day. What happens as those close and you can't cite their reports?

16:29Peter Kafka:It seems like that That is the real limit to distributing these things, even in the kind-born metaphor. If there's no one to tell you that so-and-so resigned from city council after she was arrested for a DUI, which is one of the links you sent to me, I'm not going to get that in my patch AM. Correct. So a few things about that. So first of all, just in the model, just so your listeners know, when we include a headline from another publication in the newsletter, that link clicks directly to that publication. So we don't do that kind of what I think of as a smarmy thing. It's been done to us many times where people steal multiple clicks and then hand off, you know, the fourth click to the publisher.

17:09We don't do any of that. We send the click directly out to the publisher and cite them in the summary. So, and, and the, I think we've had one publisher opt out. Mostly we have publishers asking to be included because it's, it's not new traffic to them. So you're right. If public, if, if local news goes away, that's bad for this product. That's, I think that's bad for the world. The other thing, by the way, I want to point out is our number one complaint is, hey, you sent me to a link with a paywall and we let people know, you know, paywall. so you know, but we ideologically, you know, are going to stick with that because we want to support local pubs that have paywalls.

17:52And so, you know, that's part of the idea is go support your local publication for the reason you cite. Now, there are lots of sources of information, again, at the hyperlocal level, when you think of news as maybe having a slightly lower bar than something that a journalists wrote their, their town, uh, government websites. Last week, I attended a street life committee meeting for community board eight and on the Upper East Side, that meeting is available on YouTube. And again, these are, these are things that AI gives us an opportunity to, to kind of access in a new way, because our reporter for the Upper East Side, she's incredibly busy.

18:30The odds that she could sit and listen to a two hour meeting for just one or two nuggets It's about a sidewalk cafe that was approved. You know, that would not be the best use of her time. But if we can get the YouTube video, get the transcript of that, have a eye on it.

18:46Peter Kafka:And this was this was sort of the dream you heard people talking about years ago. I don't dream is the right word, but it's definitely the idea of like, well, why do we have to have city hall reporters? I mean, presumably there's a camera recording this stuff. Can't we just show people the stream and they can figure it out for themselves? And that's kind of what you're talking about here. Are you actually able to do that? Are you able to synthesize? So we have reporters who are doing that to kind of amplify their capability so they can be in more than one place at one time. And I think, I mean, this is really.

19:15Peter Kafka:But you're not doing it through AI. The AI is not saying I've watched this clip and let me tell you. The AI is essentially summarizing the transcript. So use AI to get a transcript of the YouTube video and then essentially to read through it and summarize it. And then a reporter can scan it to see if any. That's when you've got a human involved. But for this product, the AM, the newsletter product, that's not happening right now. Not right now. But we are looking at government websites on a regular basis for, again, news. You know, the leaf pickup schedule changed from Tuesday to Wednesday. Or City Hall is closed this week because of a plumbing issue.

19:53That kind of thing comes through all the time. And so, you know, I think that what's really interesting here and what's changing like the seam that is developing and morphing around us is sort of what is the chief value of a journalist with this technological capability loose in the world? and one other dynamic here is that these you know crawlers these bots are crawling all of our news websites and uh stealing our stuff and giving the answer in full whether it's google or open ai to the end user and they're clicking through at a much lower rate we all know that this dynamic is happening. So then, which is, you know, a very inconvenient dynamic.

20:46And we experienced that with our original content. But I think one thing that is clear is that if it's out there already, it's of a lot less value. And, you know, I've been doing this long enough that I remember this sort of aggregation game, you know, the days of Gawker and early Gothamist, when you were, you know, there's a lot of summarizing, people summarizing other journalists' work and that was your blog and it was sort of digestification of the broader, you know, news on your beat. Some of us may still be doing some of that work as we speak. And it's got value. But if a machine can do it much more broadly and much faster, then what is the value of the reporter?

21:30And, you know, I would submit, and I think we're all trying to figure that out in journalism, but I think, you know, So a clear area of value is taking things that aren't available on the internet that nobody knows, that nobody can access, and presenting those and finding those things. I mean, reporting the stuff that people don't know. So if I have a reporter and that reporter, I have a choice to encourage that reporter to go comb through real estate transactions to find out what houses sold in a community, which is something people want to know, or go out and make phone calls and try to find a story that's not available to people because no one knows it yet.

22:14But, you know, I'd rather them do the latter and have a machine go through the real estate news and summarize that than have a person. Because, first of all, the machine is going to do it faster and the machine is going to do it more thoroughly. And the value of that information, because it's already out there available to others, is it's much lower than the things people don't know yet, the things that aren't online. And then there's this sort of time of decay immediately once something's published. Now bots are going after it. They're going to process it, chew it up, spit it out, sell ads against it on other platforms.

22:50So you have this sort of like, you know, the amount of time that something has value is shorter, I think, than it's ever been because other people just rush in and grab it.

23:04Peter Kafka:Right. And meanwhile, the stuff that you guys create or the humans create at Patch, like you said, gets aggregated by your competitors and other folks as well. And so it puts even more strain on that to make sure that you're really providing something that's unique. So we go back to this product where you're saying we're aggregating stuff that someone else has published, whether it's a government site or someone on social media or another publication. We're assembling it for you. It has value, right? It's free to the consumer. There's obvious value for you guys. It generates a place for you to sell advertising.

23:36Peter Kafka:Where does it go from here, though? If it's something that is working for you guys, then presumably it's available to anyone else. You guys are not a tech company. You're a publishing company. It's kind of tech forward. But in theory, a Google or an OpenAI or anyone else could create this product, right? How are you thinking about sort of where this goes in the future? Yeah. So a few things about that. So first of all, and this touches on a question I asked earlier, which is, so why not do that? It is something that we are doing everywhere. everywhere. So we offer Apache on our communities that have a full-time reporter.

24:11We see it as entirely supplemental. But I think for us, we have a product roadmap around enriching this product and making it better and better. Some of that's on our own platform. I mean, we have a lot of work that we do on our own local calendars. And again, I've written a couple of books, And between my first and second book, the recession of 2008 happened. And I went on my book tour for my second book. And I heard this refrain at all the bookstores I went to, which was, we don't know how to get word about, a word out about author events anymore. The local paper stopped doing, you know, the author features, the, you know, the art section's gone, so forth.

24:59And so now the papers are gone. And so there is this persistent struggle at the town level to just get the word out about important community events. And when we include these events, this product has a 94 thumbs up rating from its subscribers, which for us, that's incredibly high. We're very happy about that. It speaks to the problem it's solving, which is not just tell me what happened. And it's tell me what's coming up as well, which is a thing people really struggle with. And this just gets, you know, my view is we've resisted personalization of the news at Patch. We like to kind of be the shared thing, the shared community board where the things that everyone in town should know about can be found.