00:00:53.240and I've spent my life exploring how other people survive what should have destroyed them.
00:00:58.760I'm Elizabeth Smart, and these are The Survivor Files.
00:01:02.380Every week, I'm with survivors who live through the unthinkable, abducted, stalked, controlled, and nearly silenced.
00:01:09.920These are stories about what it takes to make it out alive.
00:01:13.740Listen to The Survivor Files with Elizabeth Smart on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.
00:01:21.360If your bookshelf and your For You page are equally important to your personality, welcome home.
00:01:27.100Pro Society is a weekly podcast that's part book club, part group chat for anyone who thinks Pride and Prejudice and Love Island deserve the same level of discourse.
00:01:35.740Each week, we're connecting the dots between books, the internet, and pop culture with your favorite writers, book talk creators, and plenty of overthought opinions.
00:01:56.700We host the podcast Hollywood Handbook.
00:01:58.860Every episode, we're trying to help our guests improve their careers by mainly focusing on how they can help us improve our careers.
00:02:04.740The show is famously super accessible, so you can easily jump into any of our 650 episodes and understand what's going on.
00:02:10.880We recommend some of our recent episodes with Ben Stiller, Danny McBride, Riz Ahmed, and Mary Steenburgen.
00:02:15.840Listen to Hollywood Handbook on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.
00:02:25.340There Are No Girls on the Internet is a production of iHeartRadio and Unbossed Creative.
00:02:33.780I'm Bridget Todd, and this is There Are No Girls on the Internet.
00:02:39.240A pop star accused of being a secret Nazi.
00:02:42.800A family restaurant accused of going woke.
00:02:46.160Now, at first glance, these may seem like two completely unrelated internet controversies.
00:02:51.180But dig a little bit deeper, and they share something important in common.
00:02:55.340Both controversies were amplified by coordinated bot campaigns that sought to rile up ordinary
00:03:00.940users and draw them into conversation.
00:03:03.700In this episode, we're breaking down two reports on coordinated bot campaigns, one
00:03:08.940aimed at Taylor Swift, another aimed at Cracker Barrel, to talk about what happens when online
00:03:14.400outrage is manufactured, amplified, and weaponized, as well as what it means for the authentic
00:03:19.860human voices trying to be heard against the algorithmic roar.
00:03:23.920Because if bots can create the appearance of consensus and generate real backlash or
00:03:28.660controversy at scale, how do any of us know when we're reacting to real people or just
00:03:33.680reacting exactly the way somebody wanted us to?
00:03:36.960This is what Molly Dwyer has spent her career researching.
00:03:40.280She describes herself as a kind of professional internet vibe checker, but her real title
00:03:45.160is Director of Insights at Peak Metrics.
00:03:47.360And before the internet was talking about bots manipulating conversations around Taylor
00:03:51.780Swift, Molly had been looking into how similar kinds of inauthentic behavior was driving the
00:03:57.060discourse around the old-timey country-themed chain restaurant Cracker Barrel. Now, I know it
00:04:03.380may seem wild to think that anybody would care deeply about this, but Molly says it's yet another
00:04:09.220instance of how easily the internet can be manipulated to change people's views. Her company,
00:04:14.780Peak Metrics, was not involved with the report about Taylor Swift. But a few months earlier,
00:04:19.600they published a report about the Cracker Barrel controversy that had a lot of similar findings.
00:04:24.560So we asked her about what was happening with the Cracker Barrel discourse
00:04:27.540and got her take on the recent Taylor Swift report.
00:04:31.460I spent a number of years living in Russia on various U.S. State Department programs
00:04:37.720to help young people study languages that are critical for national security.
00:04:41.500I say that's relevant because my first kind of experience of how the Internet can manipulate
00:04:47.400and change people's view on the world was my experience sort of pre and post 2014 in Russia,
00:04:53.460where I watched people that I knew personally go through quite a transformation on their
00:04:58.720perspective on Ukraine and the US in a very short period of time. So that really kicked off kind of
00:05:03.320my interest in this idea of like information operations, how did the things that we consume
00:05:08.380on the internet impact our worldview, whether we're aware of it or not. Immediately out of that,
00:05:15.580I went to work for a small startup that does open source intelligence, which is really just another way of talking about all of the data that is publicly available on the Internet.
00:05:26.080There's a pretty big tool set out there nowadays for different technologies that will help you quantify and qualify what is being said, what is being researched on the Internet.
00:05:36.200And I've had a decade of experience in that open source intelligence tool space at this point.
00:05:41.700Now, the internet of we're going into 2026, the internet of 2026 is very different than the internet of 2016. When I first started out in this industry, traditional metrics that we were used to, like volume or sentiment, or even just taking for granted the idea that, you know, a verified page was who they say they are, are just no longer applicable nowadays.
00:06:06.540And I think that the internet has fundamentally changed from under us, I think, especially, obviously, since the rise of LLMs and agentic AI. And we're all just sort of grappling and trying to figure out what this means for us. And my role at Peakmetrics is to help companies and other organizations figure out what the internet means for them.
00:06:27.980Before we started recording, you said that you basically were an internet vibe checker that I mean, I it's one of the reasons I wanted to talk to you is that I have had the experience of seeing a conversation kind of take hold of the internet very quickly, much more quickly than other kind of organic conversations.
00:06:48.340and maybe you don't know something is going on, but you suspect, right?
00:06:52.600Things where I would think, who really is dedicating their time to writing a bunch of posts about this?
00:07:21.160And like, what are the motivations for people to participate in this conversation?
00:07:24.760I'll give you an example of maybe sort of the algorithmic bubbles that we live in.
00:07:30.600Last year, I remember there were a lot of media inquiries to look at the impact of,
00:07:35.820it was the first or second presidential debate with Kamala Harris as the candidate.
00:07:40.780And I was looking at the scale of conversation to that relative to something that I had not heard of, but someone on my team brought forward to me, which was the plight of this squirrel that was like a pet squirrel that was going to be put down in New York State that Trump entered the conversation on.
00:07:58.200And I, in my algorithmic bubble, had not heard of this. I was focused on reporting on the metrics around discussions around the debate. And when I actually went to go look at just the overall volume of conversation around those two topics on X, the volume of conversation around the squirrel was multiple X larger than the volume of the conversation around the presidential debate.
00:08:20.560So I think that anecdote illustrates sometimes both our algorithmic bubbles and just like how big these conversations can get quickly when certain players are involved, like when certain big influencers enter the conversation.
00:08:35.280And then also thinking about like when bots are involved, right?
00:08:39.700Like what type of content are they incentivized to post about and to amplify for, you know, their ultimate goals probably of monetization.
00:08:50.560Molly's background in Russian makes sense here because there was a time that when you talked about bot campaigns online, the assumption was it was being done by foreign bad actors.
00:09:00.120The kind of sophisticated manipulation campaign that we associate with an adversarial foreign country.
00:09:05.480But today, things have really changed.
00:09:08.220So I think if you take the Occam's Rager approach to the Internet, which is just assuming that everything is a money making grift, you'll probably be correct in like 90 percent of cases.
00:09:18.060And I think that that applies here. I'll back up a little bit to talk about like bot networks in general. And I think that this is important to cover before we get into who is behind it.
00:09:30.200So traditionally, when we would talk about bot networks and who was operating them, you know, the degree of sophistication required to set up these campaigns to run these accounts simultaneously to get them to stay on topic with their posting, you were looking at a relatively limited set of essentially state adversary actors like the big bads that we're used to, Russia, China, Iran, North Korea, right, that had the sophisticated like cyber teams that could run these campaigns.
00:09:57.520And that's why I think in our minds, we're still a little bit stuck in this idea of like, okay, big, bad foreign actors manipulating the conversation. I think there are still big, bad foreign actors manipulating the conversation and they have a lot of incentives to do so.
00:10:11.940But the world changed with with AI and it's now easier than ever to run these types of campaigns.0.55
00:10:20.300The other side of that was you had maybe like traditional non-state cyber actors like Threat Network, you know, a non type people who were super sophisticated could do this.
00:10:30.080Right. And now, you know, smaller governments can like run a campaign to prop up their dictator.0.53
00:10:35.960Right. Average people who are not part of like a threat network can stand up campaigns like this, thanks to AI, you know, run them simultaneously.
00:10:45.520And I think that's where we get into this explanation of, OK, well, what if it's not ideological the way that like a state adversary actor would be?
00:10:53.360Well, why are they doing this? I think shits and giggles is always a reasonable guess.0.90
00:10:58.400But I think that money is the thing that makes the most sense. And I'll tell you why I think that worldview, I think, is bearing out recently.0.86
00:11:05.380So I don't know if you saw in the past few weeks that X made an update where user locations were published.
00:11:15.460And what we saw, this is, again, anecdotal.
00:11:17.820I don't necessarily have the data on this, but what we saw anecdotally was that a lot of accounts that were posting incendiary political content on both the left and the U.S. right were coming from places like India, the global south.
00:11:31.720with places where, you know, are relatively low income, the money that you could make from this
00:11:38.840is not insignificant. We don't necessarily think that these people have ideological reasons to
00:11:44.900fan discourse in the US. It just so happens that that's maybe a good way to make a side hustle
00:11:49.780online because of the way that, you know, engagement is monetized. So that's my overview
00:11:57.580answer of like the types of people who could be behind these. It's a much larger set of actors
00:12:03.800than it was maybe four years ago. But I still think that money is probably the guiding principle
00:12:11.680of what's going on here. Yeah, we've been having a lot of conversations about that change at X. And
00:12:17.500I think it really helped me see, I knew this sort of inside, but it helped. It was just like
00:12:23.640another way to sort of crystallize exactly what you said, that we're so used to thinking about
00:12:29.160nefarious actors, bad actors who are manipulating our online discourse to foment chaos and confusion
00:12:37.080and political division. But also in making this, in changing the financial payouts of X,
00:12:44.360they certainly have incentivized people to post inflammatory, incendiary content, rage bait,
00:12:50.080Right. Content that we know works as a revenue stream. And part of me can't even really blame them for being like, oh, they set up this this incentivized reason for me to do this to make a little money.
00:13:12.820If you haven't been on a long car trip in a while, I bet you haven't spent a lot of time
00:13:18.900thinking about the restaurant Cracker Barrel. So here's the controversy. Earlier this year,
00:13:23.740Cracker Barrel rolled out a new logo. Previously, the logo featured an old white man in overalls
00:13:29.900casually leaning over a barrel. But this new logo removed the old man and just kept the name
00:13:35.840Cracker Barrel in an ever so slightly sleeker font, one that maybe jived with a younger audience.
00:13:42.280To be sure, this is the kind of boring corporate marketing change that most people wouldn't even notice.
00:13:48.300But in today's climate, where every little thing can be turned into evidence of a woke agenda, it became an entire news cycle.
00:13:55.840Okay, so you've really set the stage wonderfully.
00:13:58.280And I think I mentioned this to you, that independently, we had been wanting to do an episode about bots and sort of the general grip that they have on our discourse.
00:14:06.720And one of the conversations that I saw that I was like, this is a very weird conversation happening, was Cracker Barrel changing their logo.
00:14:16.280I grew up in the South and my parents went to Cracker Barrel.
00:14:22.700But other than my parents, I don't think I've ever heard anybody mention Cracker Barrel.
00:14:27.600It wasn't like a big part of the discourse.
00:14:29.960So imagine my surprise when one day I wake up and everybody's freaking talking about Cracker Barrel.
00:14:34.440Then I saw the peak metrics report about the way that bots might have helped shape the way that conversation spread online. How did that come to be something that peak metrics was looking at?
00:14:45.840Yeah, I too can't believe that we're still talking about Cracker Barrel, but here we are. I think why media latched onto it when we put out some initial findings was this sense of why are we talking about this?
00:15:00.400And we gave them some numbers to chew on to help them understand what degree of this conversation is potentially inorganic or automated that might explain why it's staying in the discourse longer than we would expect it to be or why it rose to the top of the discourse in the first place.
00:15:19.740And I think our key finding there was that, you know, within the first 24 hours of this like rage cycle, we found that the original posts that seeded this idea of a boycott.
00:15:34.620And you got to rewind it back because I think that's also another principle that's hard to do on the modern Internet is how did something start?
00:15:42.200There's not necessarily an easy answer to that.
00:15:44.180You have to bring in a lot of data, search for the right stuff to figure out how a trend even started.
00:15:49.740So from what we could see, how the idea of a boycott started, it did start from what our tech would classify as organic accounts.
00:15:59.220But very quickly, those calls for a boycott were amplified by accounts that we flagged as automated or inorganic.
00:16:08.600So there was this sort of like seeding introduction of a claim that immediately got amplified.
00:16:15.900And at the height of the discourse, like within the first 24 hours, we found that basically half of the posts that we're calling for a boycott were coming from accounts that we flagged as automated.
00:16:27.820Half is a big number, but we also don't necessarily, at the time that we published that report, we didn't have a lot of comparison points.
00:16:34.640I think that's key to maybe understanding what's going on with the Taylor Swift discourses.
00:16:38.980I think we're all across the industry, like finding these numbers and trying to figure out what they mean.
00:16:44.240If I were to give you a spidey sense now, several months after Cracker Barrel, is that when there's a incendiary conversation online, I would consider like a normal baseline for the amount of automated activity in that conversation to be somewhere between 20 to 30 percent of the conversation.
00:17:01.120And it typically will go higher than that when we hit like a crisis point. So like the Cracker Barrel changes its logo, organic accounts seed this call for a boycott, and then it jumps up. But that may just be a symptom of how the internet works and not necessarily a symptom of a coordinated campaign to target chain restaurants that are abandoning traditional values, right?
00:17:32.620We're putting a number to this thing for the first time.
00:17:34.740And so we're trying to figure out what is baseline at this point.
00:17:38.560So what's the takeaway that you all found from the Cracker Barrel report?
00:17:41.580I think the takeaway here is, you know, one and two being automated.
00:17:46.100That's a big number, but that's still a lot of people that have potentially big, real feelings about this.
00:17:51.700The other element that's at play here, though, is because of how algorithmic boosting works,
00:17:56.500How many of those people who were real people posting outrage would have even seen this to post their outrage had it not first been amplified by inorganic accounts that like brought it to the top of their feed?
00:18:10.620So even within that that number of people with big, real human feelings, there is an element of being shaped by this inorganic discourse, bringing that conversation to them in the first place.
00:18:23.380I saw an analogy to this, that if you were cutting vegetables in your kitchen and you cut your finger and you ask somebody next to you, hand me that towel so I can apply pressure.
00:18:32.060If they squeezed it instead so that you bled faster, they didn't start the cut, but they certainly made it worse.
00:18:37.720They certainly brought the problem, you know, to a different level.
00:18:41.200And that's sort of a good way to think about how bots and inauthentic activity can shape a conversation online that might actually be seeded with organic people and their big feelings they're sharing on the Internet.
00:18:51.420Now I'm just thinking in my head of like what all of the potential bots would say if you ask them to hand you a towel. Because I find that you can spot them in the comments section of like they'll take things too literally sometimes. So it would be, you know, like something about the towel, something about like not being able to squeeze your finger.
00:19:11.200I think there was a key point in time where we saw some of the bot networks like Short Circuit with their instructions about what to post at the peak of the Trump Epstein files controversy where in real time, I'm not the only person who saw this.
00:19:26.160I don't know if there's any anything written about this, but other people were observing this, that like a lot of these seemingly MAGA accounts were turning on Trump in the context of like calling for the Epstein files to be released, because you can tell that they're like pre-programmed instructions were to like nonstop call for the Epstein files to be released.
00:19:47.560And they short-circuited a little bit based on their previous instructions for the world before, you know, they got updated to the version of the world that existed at that point in time.
00:20:08.640I think if people are wondering, you know, like, what is a good way to confirm your spidey sense of something in serial?
00:20:14.000I mean, I think we we're all used to like those account handles that are like John 24560, whatever. Right. Like we know that that's a that's a common way that those account names are formulated. We do still see that. But I think even without any other technology to flag automated behavior, what you can do is if you're like on your phone and just scroll a couple of times.
00:20:35.960If you're still scrolling and you have not reached yesterday in that person's posting, you know, that's probably a sign that they are posting at a frequency that is just not probably humanly possible and is probably, you know, an automated posting rate that's going out there.
00:20:53.760So that's my tip of one of the simplest ways to be like, eh, is it a bot?
00:20:59.380Yeah, just scroll back and see how far it takes you to get to yesterday.
00:21:02.300That is a great tip. And yeah, don't spend don't dedicate your time and energy to getting into a back and forth with a bot or like worrying yourself with what a bot is saying, unless you're doing it because you're a researcher and you're interested. Don't have it like fuck up your day. What what a bot left in your comments on Instagram or something?
00:21:21.640Yeah, exactly. I mean, I think we're talking about X and I do think that X is pretty central to like the discourse in the sense that it really is like the most fertile ground for bot behavior. Like it's very text based, which means it's like easier to, you know, produce content there.
00:21:38.820Like it's a little bit harder to automate like posting of pictures, if you think about it, like just logistically. So I think we do still see a lot of bot activity on X where I think researchers are less familiar with what bot activity looks like on other platforms.
00:21:54.580I will say that, you know, I'm seeing a lot of folks going to more niche communities like Facebook groups or Discord or Reddit, where I think that there's a desire for people to like be in conversation with real people on the Internet.
00:22:12.560And when they're finding that that's not possible in certain forums anymore, they're moving to different places to try to be assured that when they're engaging in conversation with someone that, yeah, it's not like a bot who's arguing back at them.
00:22:28.200Especially in this age of AI, I want to know, even if we're having a spicy conversation or an argument, I want it to be with a real person, like in Des Moines, Iowa or something.
00:22:37.520I don't want to have the feeling of like, what am I, wasting my time going back and forth on Reddit with a bot?
00:23:53.580If your bookshelf and your For You page are equally important to your personality, welcome home.
00:23:59.520This is Prose Society, the weekly podcast that's part book club, part group chat for thought daughters,
00:24:05.940pop culture obsessives, and anyone who thinks Pride and Prejudice and Love Island deserve the same level of discourse.
00:24:11.900I'm Eli Rallo, and every week we're connecting the dots between books, the internet, and the conversations everyone can't stop having.
00:24:19.080I'm going to have to look up this story. I'm obsessed. I'm obsessed.
00:24:22.660From best-selling authors and your favorite book talk creators to the latest pop culture moments, nothing is off the table.
00:24:28.620It's like if you can hide some real messages inside compelling characters, then that is a Trojan horse.
00:24:36.080Whether you're looking for literary deep dives, smart pop culture conversations, or a community of readers who love to think a little too much, you're in the right place.
00:24:45.680Listen to Pro Society on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.
00:26:27.040and you make it on you make your own that's great i call him tarot reed
00:26:31.200it's a good name they're all named after cast members of american pie yeah i call i don't
00:26:39.360i never looked up if she was in this but i call him lychee sobieski yeah i don't think she was
00:26:44.420an american pie but i haven't looked it up she was prominent around the same time yeah then
00:26:49.660mary steenburgen why are you guys talking to me like i'm worried why are you talking so down to
00:26:55.240me that you think I don't understand your little play game. I just really am trying to set you up
00:27:02.460for success here. Listen to Hollywood Handbook on the iHeartRadio app, Apple Podcasts, or wherever
00:27:07.540you get your podcasts. And we're back. Since Molly is an internet vibe checker, I wanted to know her
00:27:21.380thoughts on the report around the controversy surrounding Taylor Swift. Things like accusations
00:27:26.140that Swift is a secret trad wife or even secretly a Nazi. So I want to talk about the report that
00:27:34.440I'm sure you've seen by now that was put out by a company called Gadea that was then reported on
00:27:39.720in Rolling Stone that looked into social media commentary around Taylor Swift's latest album
00:27:44.300and whether or not that commentary was in part driven by inauthentic coordinated accounts.
00:27:50.240What was your reaction, just as somebody who puts together reports like this, who's somebody who is in this space, what was your reaction to that report?
00:27:57.680I mean, I think it was a it was a spidey sentence check.
00:28:00.580I am not a Swifty. Please don't come after me, Swifties. My wife is a Swifty.0.97
00:28:06.400But I, as a person on the Internet, had had seen the reactions to Taylor Swift's album.
00:28:13.240Now, at the time, I was not discriminating whether I thought that those reactions were organic or inorganic.
00:28:18.120But I was aware of the pushback to the most recent album. And looking through the methodology, it makes sense to me. I think it's sound in certain aspects of it.
00:28:29.640And I think maybe where sometimes the nuance gets lost is, you know, we talk about how
00:28:35.740one part of a conversation is automated or manipulated, and that's not to discount the
00:28:43.580other part of the conversation that has real people having real big human feelings here.
00:28:49.160And inherently in whether it's data sampling or, you know, the way that your technology,
00:28:55.640whether it's by keywords or something else, is identifying the parts of the discourse.
00:29:00.480You're going to more easily identify the parts of the discourse that like sound automated to begin
00:29:06.800with because they're flagging like the right keywords or the right markers for what's going on.
00:29:11.700And inherently, you're going to miss the more nuanced conversations because people, when real
00:29:17.920people talk about issues, they're all using slightly different language to describe what's
00:29:22.140going on. They're maybe not referencing Taylor Swift's full name in the conversation. They're
00:29:27.380replying. So inherently in the work that we do based on how you're going to collect this data,
00:29:33.760it's already going to veer a little bit more towards missing some of those more organic
00:29:37.300conversations because of how humans talk about things. So something got lost down the road
00:29:43.380in terms of the real people having real human feelings about this. I don't disagree based on
00:29:50.240the methodology, that there are certainly indicators of automated activity here.
00:29:55.040But we don't necessarily know, going back to the earlier question, you know, these bot
00:30:00.920networks that were, you know, jumping on the bandwagon here that also were attacking Blake
00:30:05.900Lively, you know, we don't necessarily know what the incentives are of these bot networks.
00:30:11.220Are they just hopping on the latest celebrity trend to gain traction and monetization?
00:30:17.960or are they targeted against certain celebrities?
00:30:21.140Like these are unknown questions at this point.
00:30:23.660So I think we want to make sure to not jump too many steps forward
00:30:27.220to claim that billionaire Taylor Swift is being targeted
00:30:31.520and we need to protect her from attacks online.
00:30:34.040There are a lot of other people who are being targeted
00:30:36.160who don't necessarily have Taylor Swift's PR arsenal at their back.
00:30:42.220Yes, I'm so glad that you brought that up because, I mean,
00:30:45.540I don't know if you've I mean, I'm sure you've seen the way the conversation around this report has spread online.
00:30:51.240And one of the things that I've been a little frustrated to see is people saying, oh, well, this report confirms that everybody who was critical of Taylor Swift was just a bot.
00:31:03.000They can they can or people who were, you know, manipulated by bots.
00:31:06.700All of that discourse can be dismissed.
00:31:08.680So, first of all, the report does not say that.
00:31:10.800Second, not even a little. And in fact, something that I think is getting lost in the sauce of the report is that, in fact, the report is talking about authentic accounts having authentic discourse alongside what they have what they have seen as like perhaps inauthentic discourse.
00:31:25.400So like nowhere in that report are they saying everybody who was critical of Taylor Swift can just be dismissed as a bot.
00:31:31.500And I think that there's something about the way it was reported that is not giving enough credence to the fact that there were there were and are real people in this conversation who are not bots, who have been talking about Taylor Swift critically for a long time.
00:31:47.480And so even though, you know, I'm no data scientist, but looking at their methodology, I don't think I don't think they've made this up.
00:31:54.980I don't think Taylor Swift paid them to put this reporting together or something like that.
00:31:58.860But I understand what people are saying when they're saying this report does not actually reflect the fact that there are so many people authentically being critical of Taylor Swift on the Internet.
00:32:09.760We're not bots. I think something about the way the report was framed sort of gave credence to this idea that you could just be dismissive of all of these critical voices.
00:32:17.900Exactly. I mean, I think on the flip side, we could look at any other recent controversial issue and perhaps come to a similar conclusion of, wow, this conversation is bot driven, because I think that that may just be a symptom of like how the Internet functions nowadays that may not necessarily be specific to the dynamics of Taylor Swift.
00:32:40.320I think the other thing that you can do, and I mean, I'm taking lessons from this, you know, working and researching in this space is, you know, one of the one of my favorite ways to set folks up to analyze conversations online with peak metrics technology is I'll say, you know, look at the data and ask the same question filtered to the organic activity and the non-organic activity so that you can see the nuance and the difference and maybe like the specific themes or narratives that they're talking about.
00:33:09.460So an interesting question to ask of this data might have been within the controversial conversation around the latest album, which ones specifically were the bots trying to push? And then when it comes to real human people, which aspects were they focused on?
00:33:29.100presenting, I think, maybe those two things side by side gives us better context for what those
00:33:36.080people who were having real human feelings about this were thinking, which aspects of the album
00:33:40.260controversy were they latching onto the most? And how did that perhaps differ from what the
00:33:44.860bots were focused on? And I think there's also I mean, I can't not talk about the way that this
00:33:50.540seems to the conversation seems to be happening along some clear racial lines to me, where a lot
00:33:56.480of the voices who were critical of Taylor Swift, not all, but a lot, were women of color, Black
00:34:02.020women, people of color. And a lot of the Swifty community appeared to be, again, not all, but a
00:34:07.420lot of white ladies, white people. And so I think it's just one of those issues that will always0.92
00:34:12.820sit at these tension points of the tensions that we know exist in our society. Something about the
00:34:19.200report giving credence to the idea that you could just discount a largely, like the critical voices
00:34:28.180of largely minoritized people because you love Taylor Swift and you don't have to think critically
00:34:33.260about the points these people are making because they're bots. I can see why that just hit people
00:34:37.460sideways. Yeah. And I think to keep the organic voices anonymized here, I think that might have
00:34:44.460been a great opportunity to look at the people on the organic side of the conversation and see
00:34:49.180Like who were the biggest influencers in the space? So within people who were criticizing the album or the aesthetics of Taylor Swift and they were organic, you know, who were some of the biggest accounts that weighed in?
00:35:00.320Like there are, you know, I'm aware that like black Twitter is a thing, like who were some of the biggest voices that were that were shaping the conversation from the human side?
00:35:08.780What we may uncover if we looked at like, you know, there might have been a post from someone who has like 500,000 followers and arguably that might have shaped the discourse a lot more. That single post from that person with a lot of reach might have shaped the discourse more than the, you know, 50 posts from bot accounts that have like 100 followers.
00:35:32.900So maybe there's some nuance to be parsed out here, too, that volume doesn't necessarily equal impact or influence on the Internet.
00:35:41.720And what bots are inherently trying to do is boost the volume of the conversation.
00:35:47.980But where humans have an impact, and there are also like automated accounts that look like influencer accounts that have lots of followers.
00:35:54.840But I'd like to pull the thread a little bit more about who are the influential people on the human side of the conversation.
00:35:59.900That makes a lot of sense. And I guess I would have liked I also think that, listen, we have said this on the show a million times when you talk about something, there's something about Taylor Swift that is like, once you start talking about her, big feelings come out on all sides.
00:36:14.600Even people who don't like Taylor Swift, it's like there's just something about her that gets people talking.
00:36:19.020And so I think if you're going to be putting out a report that is about Taylor Swift, it would behoove you to make some of this clear.
00:36:25.560Right. It would behoove you to spend a little time explaining like, oh, well, who are the voices that we're seeing on the like,
00:36:33.700who are the authentic voices that we're seeing that are talking about her in this way and weighing in this way?
00:36:38.020Because you have to imagine it's going to get a lot of eyes on it.
00:36:40.920By now, everybody knows that when you talk about Taylor Swift, it's something that gets a lot of eyes or a lot of ears or whatever medium you're using.
00:36:46.940Yeah, if I if I maybe have three guideposts that I'm making up on the fly as an as an Internet person, having been on, you know, Tumblr back in like the early 2010s, you knew not to mess with the Swifties.
00:37:01.500Right. Like that's so the guideposts of the Internet is like, what do we establish?0.97
00:37:06.000Like, assume it's a grift. What is the other one from that Netflix documentary? Don't fuck with cats. And I think maybe the third guidepost of the Internet is do not touch the Taylor Swift discourse with a 10 foot pole. If you follow these things, you can't go wrong on the Internet.
00:37:26.300What do you think about the methodology that they used in this report to define bot accounts?
00:37:30.100I think it's challenging that we're using a common word that you could define in a lot of different ways. I think that there's no, you know, set methodology. I think a lot of different tech companies have different definitions and have their own sort of secret sauce when it comes to, you know, what available data they're using to come to these conclusions.
00:37:53.200I can tell you that, you know, from the way that we think about it coming from sort of a framework of, you know, degrees of confidence that you would see in like the intelligence community, right?
00:38:05.040Like you're never going to say, I'm 100% sure that, you know, something is something you're going to say, okay, well, the available data that I have shows, you know, it's highly likely that this is a bot account.
00:38:16.580Even when you're getting into the highest degree of confidence that something is a bot account,
00:38:22.520the way that we frame it is that it's almost certainly a bot account, which is still,
00:38:27.520you know, by verbiage, a little bit short of saying, you know, for sure. Because I think the
00:38:33.460only way to know for sure that something is a bot account is to do a lot of honestly, like manual
00:38:39.240forensic analysis of that account. And the issue is that you can't do that at scale. So when you're
00:38:44.280looking at large sets of data and you're looking at, you know, a limited availability of metrics
00:38:49.480that you have on these accounts. Honestly, I don't think it's necessarily the way that,
00:38:54.960you know, the bots are classified. It's the language that you're using to describe the
00:38:59.120confidence that these are bots. And again, I think there's something that's getting lost
00:39:03.380in translation between like a methodological report, journalistic reporting, and then where
00:39:08.020that journalistic reporting goes in the discourse. So we have, you know, things like account history,
00:39:13.020Like when was the account created? We have the posting frequency. You have like the, you know, profile image, right? Is that like a recycled image or an AI generated image? You know, we talked about like the obvious indicator of the way that the username is formulated. Are there other usernames like that that appear on other social media sites?
00:39:33.060It's what type of content is this account posting?
00:39:36.760Is it mainly acting as an amplifier and just resharing, reposting?
00:39:41.980Or is it doing a lot of original posts, which is just like mechanically and also computationally
00:39:48.660from like a energy perspective going to be harder for like a bot run account to maintain
00:39:55.120So I throw out all of these metrics to say that I think that the answer lies in probably
00:40:00.160a combination of all of them to get to the best answer. But I think we need to be really careful
00:40:06.240about the language that we're using when we say like the confidence level that we have that
00:40:11.320an account is automated. One of the criticisms, and I mean, you've been, I think, helpfully kind
00:40:18.960of critical, good critical of the report. But some of the criticisms I'm seeing of people sort of
00:40:23.340just trying to trash this report and say that it's a lie is that on their site, they have a
00:40:29.060like a disclaimer that's like oh we cannot we're not saying that we're not guaranteeing that what
00:40:34.800we say in these reports is true and i thought to myself well this seems like standard cover your
00:40:39.780ass language to me that like of course they're not going to say with 100 certainty every like0.54
00:40:44.860what we're saying you can take it to the bank take it take it to your grave 100 true i didn't feel0.92
00:40:49.700like that was a necessarily a fair criticism of saying this report is full of lies it's bunk
00:40:55.300Yeah, I mean, I think when you're stepping foot into measuring the internet, I think the issue is just that like, what I'm finding is that media outlets are really hungry for any data that quantifies what's going on, don't necessarily have folks in house who can fact check or verify that data.
00:41:16.560So you're really being, you know, it's a pretty big responsibility to be the person who is doing the research and the person who is presenting that research to an organization that, you know, doesn't necessarily have the same tools at their disposal to interrogate it.
00:41:34.560And I feel that responsibility, someone working in this space, which is why I will not touch the Taylor Swift discourse with a 10-foot pole.
00:42:24.780credit visit bell.ca for details and to check availability bell connection is everything run a
00:42:30.600business and not thinking about podcasting think again more americans listen to podcasts than ad
00:42:36.120supported streaming music from spotify and pandora and as the number one podcaster iheart's twice as
00:42:41.500large as the next two combined so whatever your customers listen to they'll hear your message
00:42:45.940plus only iheart can extend your message to audiences across broadcast radio think podcasting
00:42:51.240can help your business? Think iHeart. Streaming, radio, and podcasting. Let us show you at
00:42:56.580iHeartAdvertising.com. That's iHeartAdvertising.com. If your bookshelf and your For You page are
00:43:02.780equally important to your personality, welcome home. This is Prose Society, the weekly podcast
00:43:09.640that's part book club, part group chat for thought daughters, pop culture obsessives,
00:43:14.120and anyone who thinks Pride and Prejudice and Love Island deserve the same level of discourse.
00:43:19.040I'm Eli Rallo, and every week we're connecting the dots between books, the internet, and the conversations everyone can't stop having.
00:43:26.160I'm going to have to look up this story. I'm obsessed. I'm obsessed.
00:43:29.740From best-selling authors and your favorite book talk creators to the latest pop culture moments, nothing is off the table.
00:43:35.680It's like if you can hide some real messages inside compelling characters, then that is a Trojan horse.
00:43:43.200Whether you're looking for literary deep dives, smart pop culture conversations, or a community of readers who love to think a little too much, you're in the right place.
00:43:52.760Listen to Pro Society on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.
00:46:59.580that's not even really what it said it's so many steps too far um and i don't know how many people
00:47:06.360are aware that right like headlines are written by different people who write the article you
00:47:10.700know i it's it's it's almost like we circled back to the the thing that people have said for you
00:47:17.660know decades that they don't like about the media right like so this isn't necessarily an internet
00:47:21.500problem at this point that we're talking about this is a problem of like media and the common
00:47:27.520grievances that people have about how things are clickbaity yeah i don't know that people know how
00:47:32.640pieces like this come to be and i also think rolling stone like i saw people going back and
00:47:37.700saying oh well when her album came out taylor swift did a rolling stone takeover they gave
00:47:42.400her album a five out of five like be and i don't think that that's i think that that's fair to be
00:47:49.120part of the conversation is what relationship did rolling stone have with taylor swift before
00:47:53.440And then, like, why were they chosen to get this exclusive report about her being the target of this, quote, coordinated attack?
00:48:01.640I don't think those questions are totally out of pocket, but I think it really goes to show, like, why you need to be so intentional and so careful when you're going to be rolling out a report like this because people are going to run with it.
00:48:14.340Yeah. I mean, coming from this side, from the industry side, I would say, you know, there are two sides of how this could work. You know, sometimes we find something internally that's interesting and, you know, we shop it out to various media outlets and we say, like, is anyone interested in this cool data that we found?
00:48:35.340And sometimes it's the flip side. Sometimes it's an outlet coming to us that they're asking this question of a lot of different tech companies or researchers in the space and saying, hey, does anyone have an answer to this question?
00:48:47.040So that would be kind of my follow up from like the mechanics of the industry perspective is was this report shopped around to a number of different outlets and just it happened to be that Rolling Stone was the one who picked it up or was Rolling Stone calling different companies asking if anyone had any data on this.
00:49:06.340We didn't get a call from Rolling Stone asking if we had Taylor Swift data.
00:49:09.960So I can tell you that they did not call us for a consultation, but it is something that happens frequently.
00:49:15.960And sometimes I will exhaustively research something and, you know, work with a media organization about how to, you know, write an article about it.
00:49:25.020And sometimes that article doesn't even get published. Right.
00:49:27.800So, you know, that's kind of the background of how this all works.
00:49:31.780So I'm not sure which dynamic was at play here.
00:49:33.880If I can jump in with another question here, I think this is, you know, raises some good questions about the role of industry in making this kind of data available to the public.
00:49:44.240So, like, I come from a public health background where resources are pretty limited, right? And with the platforms shutting down API access to researchers a couple years ago and, you know, the Trump administration shifting resources away from a lot of funding mechanisms that were there in the past, the capacity of, like, nonprofit public health researchers to do these kinds of investigations is really pretty limited.
00:50:13.880And so, you know, I've seen criticism online about this Taylor Swift report that like, oh, it's a for-profit company. We can't trust anything that they say. But given like just the state of the world and how complicated the internet is and the limited resources and capacity of people in public health, I'm really like skeptical of an approach that says like, well, we just can't trust anything that comes out of industry whatsoever.
00:50:43.240It just feels like so limiting to shoot ourselves in the foot when, you know, I wish that we had companies like yours doing these kinds of analyses of public health topics instead of brands or celebrities.
00:50:59.300But, you know, it's just not there. And so I'm just curious what you think about, you know, what is that relationship of, you know, for-profit companies that are acting in private space, but then contributing to public conversation? How should members of the public think about that?
00:51:18.380What do you think about that? That's a great question. We're getting like into the nitty gritty of this industry and I really like that. So I can tell you that I'm working on public health problem sets for our customers, right? But that's not necessarily something that we're being asked to report publicly about.
00:51:35.900So I think maybe to like lift the hood a little bit for the industry side is, you know, I'm working on a huge number of different customers from Fortune 500 to government to commercial to big PR and entertainment.
00:51:50.840And the only things that I end up being able to talk in public about are if that customer wants to, you know, do a public facing report or if media asks us.
00:52:02.340And so that's why in the industry we'll say yes to media asks, because that's a way to highlight our capabilities. And what does media typically ask about? It's typically politics or celebrities.
00:52:18.580So if you were to look at, you know, what are the issues that I've commented on in my in my current role over the past two years, you would come to sort of like an odd conclusion that I am very into only pop culture and politics and this is 100 percent of my work.
00:52:36.260Those just happen to be the issues that I'm able to talk about publicly. Because to be frank, a lot of these tech companies were small. I know Peakmetrics is still in a startup role. And so we don't necessarily have the time or resources to pull together the most interesting public-facing report on the most pressing issue if we can't guarantee that it's going to get placement somewhere.
00:53:00.820That's a lot of energy and resources to expend on something. So that's why companies, you know, answer the call from media. So media gets basically this free data and companies get free, you know, advertising of the types of things that they can do so that, you know, customers who have other types of problems that are not maybe pop culture or politics related call us up to say,
00:53:28.900hey, can you take a look at this issue for me? So that's the mechanics of like how this works.
00:53:34.080And I think the incentives, I don't think there, I think everyone should be skeptical of things
00:53:38.060that are coming always from private industry versus like academia and researchers. But we
00:53:42.520also don't really have like an established discipline of internet researchers, I think,
00:53:48.060in academia to even, to even call from. I mean, you know, there's like the, am I thinking correctly
00:53:54.360of the Shorenstein Center at Harvard, you know, love all of their stuff that they publish. But
00:54:00.600I mean, it's an emerging field. And I think that's maybe why you have a sense that industry
00:54:06.560is leading the way is because I think that academia has not figured out where to carve
00:54:11.520this out yet. Oh, Molly, you just hit my I'll just say I agree because I'll if you get me going,
00:54:17.840I so formerly I was a research fellow at Brookman Klein, which is sort of a cousin
00:54:24.000to the shorenstein center at harvard uh people who do research on the internet in from in in
00:54:32.560an academic way attached to universities are having a rough time i'll just put it that way
00:54:38.160and they're used it used things used to be better there used to be more funding but if you do that
00:54:42.800kind of work and you're doing it today in 2025 god bless you i am happy that you've got it figured
00:54:47.940out but we have really hollowed out an entire space of researchers who are interested in what's
00:54:53.920happening on the internet. It's not a robust field any longer. I can definitely confirm that.
00:54:59.900But I'm really glad that you made this point about private companies versus academia and
00:55:06.640other spaces, because one of the criticisms I've seen online about this report was that
00:55:11.960it's essentially Taylor Swift PR, right? And they're talking about this in an almost nefarious
00:55:19.640way that, oh, this is just a company whose whole job is to get good, positive press hits for
00:55:26.840brands and celebrities and things like that. And what's interesting to me is that from what you've
00:55:32.840said, it doesn't sound like that criticism is like totally wrong, because in a kind of way,
00:55:37.980that's how these companies get their name out there so that they can continue to fund the other
00:55:41.880important work that they're doing. But that like, it's sort of missing the forest for the trees of
00:55:46.720what's actually going on in terms of how this research comes to that to get to the public.
00:55:52.020A hundred percent. And I mean, I think if they were successfully doing PR plans for Taylor Swift,
00:55:57.600they would have a much more, you know, robust company at this point. And we would be able to
00:56:02.780tell that the roof would be in the pudding. I think I guess one other element to talk about,
00:56:08.820like while we're just talking about, you know, the vibes of the Internet is, you know, you talked
00:56:14.220about the hollowing out of, you know, academia on this. There's, you know, maybe fewer people
00:56:21.140in-house that news and media organizations have that are like internet experts and are giving
00:56:27.940kind of like the time and resources to dig into this. I mean, if we're talking about like
00:56:32.960internet vibes and reporting on the internet as a beat, I feel like I have to mention
00:56:37.460And Brandy Zadrozny, who I respect her work so greatly, getting to pitch peak metrics data to her and interact with her is like been one of the highlights of my career, because that was that was the first time that I really saw the Internet being reported on as like, hey, we're sending a reporter out to this location.
00:56:56.060No, we're sending a reporter like into the bowels of the internet to come back and tell us what's going on. And, you know, not every news organization has a brand new Zedrosny. And so that's why they call up companies like Peakmetrics to, you know, give them a sense of what is going on on the other side, which is, you know, this thing that we're all a part of.
00:57:20.480But you can't just walk away and, like, put your finger up in the air and say, like, as a matter of fact, this is what's happening.
00:57:26.880One of the questions I was going to ask is, you know, what's the solution to this, to the way that bots are disrupting our discourse?
00:57:33.200And I was thinking of it as a question of, like, what should platforms be doing or not doing?
00:57:37.180But it sounds like stepping back, the problem is so much more layered than platform X needs to do a specific thing.
00:57:45.680It's a it's a problem of media outlets.
00:57:57.360Yeah, I mean, we're going to continue probably to be in like an outrage clickbait cycle of, hey, the bots were this big of a part of this conversation.
00:58:06.280And what implications does that have for the subject of this conversation?
00:58:09.540but you know until we get a sense of like what is the normal baseline level of bot activity
00:58:16.280um we're just going to kind of keep spinning on this wheel so i i'm hopeful that maybe this time
00:58:22.540next year i feel like this is this bot activity discourse has really hit maybe in in 2025 um we're
00:58:30.600all a little bit more aware of it now we're we're building out you know a repertoire of research on
00:58:34.700You know, hopefully if we if we were to be coming back together in December 2026, maybe we'll have like a better understanding of this that will get us less locked into these outrage cycles.
00:58:46.780Molly, is there anything else that I have not asked you about that you want to make sure gets included in this conversation?
00:58:53.640I guess like, is there a part of the Internet that makes you happy?
00:58:56.640Like, why are we all why are we all still here on the Internet?
00:59:03.760What makes what fills you with hope for the future and or just you just like spending time online?
00:59:08.660Yeah, well, I mean, I guess I mentioned that I was like a 2010 Tumblr girly and I still have close friends that I met on Tumblr that I am talking to, you know, 15 years later.
00:59:23.280So I think that that's like the power of Internet sub communities when they're working well.
00:59:30.340And I think that people use the Internet to find connections like that.0.74
00:59:36.580I'm so used to talking about the fucking dregs of the Internet that I forget that actually it's a place that I like and I'm there every day voluntarily.
00:59:44.920And like it was hugely informative to me as a youth.0.96
00:59:48.560And, you know, that's why I keep coming back.
00:59:50.200It's a good it's good reminder to be like, you actually enjoy technology, right?
00:59:53.780Like you actually like voluntarily keep it in your life.
01:00:53.840It turns out there's a lot more use cases than the limited one we had in the nonprofit, so we then spun out into a company.
01:01:01.160Keith's company, Gadea, was all over the internet thanks to a report they put out looking at how coordinated bot campaigns influenced discourse around the release of Taylor Swift's latest album, Life of a Showgirl.
01:01:12.700You might recall that after the album was released, Taylor put out a necklace featuring lightning bolts, an image that some felt bore a striking resemblance to Nazi SS iconography.
01:01:23.140The report examined what sparked those claims and how those claims spread across the internet like wildfire.
01:01:29.400The report was published in Rolling Stone magazine.
01:01:32.200We like to consider ourselves the storm tracker for the internet.
01:01:35.220So as a meteorologist uses patterns to predict the weather, we use patterns to predict information online.
01:01:41.540So we take in information from over 480 different platforms.
01:01:45.960And then we have a patented process where we run grapheneal networking through that data to see how people behave with it.
01:01:52.340and then use AI to lift out those behavioral patterns and summarize for our clients.
01:01:57.740I have seen people say that what your company does is essentially PR,
01:02:01.660that celebrities and brands like Taylor Swift just give you all money to print nice things about them.
01:02:07.720What do you say to somebody who feels that way about what you're doing?
01:02:11.500I would say, one, this report we did on our own.
01:02:38.320And so this was a way for us to get something out there so that people could have a better understanding of how online information environments are actually being manipulated.
01:02:47.020It's true that most of the social listening work happening in 2025 stays behind closed doors.
01:02:51.800And even though the headline in Rolling Stone made it seem like something extraordinary had happened with the Taylor Swift narrative, to analysts who work on this kind of stuff all the time, this kind of influence campaign was basically just another Tuesday.
01:03:05.440It's funny that you say that this report started with just sort of a gut feeling.
01:03:10.460Months ago, back in October, we did an episode about the conversation linked to the launch of Taylor Swift's latest album.
01:03:21.060I'm not a data scientist. My producer is a data analyst. But like I do that is not a skill set that I have. But I have been on the Internet for a very long time. And I'm particularly pretty plugged in with how conversation about marginalized people, so like women, especially women in the public eye, I'm pretty clued into how those conversations move.
01:03:42.340And the way that I would describe the conversation specifically around the Taylor Swift Nazi necklace thing, it was like it came on strong out of nowhere, was such an intense conversation.
01:03:57.220And then it kind of stopped just as abruptly as it started.
01:04:14.480There are some telltale signs of things are fishy, right?
01:04:20.240And this is when you see such huge surges out of nowhere, that is normally a leading indicator that there are some sort of some sort of coordinated activity to get that information out there to then impact how a normal person is going to talk about it.
01:04:37.420So get them to interact with that negative or illicit content.
01:04:41.900right? It's the goal. They're rage baiting. So is that really kind of how this report came to be?
01:04:47.400Just this inkling of something fishy is going on here. Let's find out what it is.
01:04:52.000Yeah, truly. I know in the article, Georgia Paul says, you know, she just had a feeling. It really
01:04:59.200came from that. It was, we were like in a, you know, one of our early morning meetings and we
01:05:04.000were like, hey, let's do it. Let's investigate and see what's going on there. So what did the
01:05:09.080report find? So essentially that the Nazi, that Taylor Swift is a Nazi and that the necklace
01:05:16.820with the lightning bolt necklace, and that was alluding to the SS from Germany. And what they0.60
01:05:21.980were doing is, so they laid down that content, a large quantity of that content to then impact
01:05:28.220influencers who would then pick that up and then spread it more broadly to normal individuals.
01:05:32.720At that point, the narrative transforms to actually then being a comparison of Taylor to Kanye. Having something like that happen is the end goal, right? Because that was actually real people now having that conversation, not driven by this inauthentic activity.
01:05:49.600Yeah, thank you for the summary there. I'm kind of jealous of the position that you guys are in, like having this data, because like Bridget was saying, we had a similar like feeling and conversation that something felt off about this conversation.
01:06:02.980uh but that's where it ended with us right because we did not have the tools readily available to
01:06:09.520to look into it and one of the things that has uh come up for me as we were researching this
01:06:16.420episode and talking to people is just how valuable it is for the public to get glimpses like this
01:06:23.500uh using data to like actually bring some data to help us understand uh what can seem very random
01:06:30.940But but often is not in terms of like conversation happening on the Internet.
01:06:35.080It is almost kind of like a black box, the Internet, especially the social media platforms where there's millions of posts, billions of posts being made daily.
01:06:46.100Lots of different coordinated activities to either, you know, from crypto schemes to just influencers wanting to get clicks.
01:06:55.120There's a lot of competing priorities that just you can't see normally.
01:07:00.080Like you can't get that bigger picture, especially as a normal person.
01:07:04.400Right. We can only really see what our feeds are.
01:07:06.720Yeah. And I always make this point on the show.
01:07:08.720There are so few spaces or industries where the public is coming into contact with it regularly, where you have such limited information or data about how it's impacting people.
01:07:20.960Right. If if if there was a car company that was killing people and we weren't able to get that information, if there was a pharmaceutical company,
01:07:28.280There aren't really a lot of companies or industries where we've just accepted this is a black box of information where the public is interacting with this every day, but has no idea what that interaction actually means or looks like.
01:07:41.340On our end, this industry is really new to just trying to actually understand how the Internet works or information moving on the Internet or how it's impacting people.
01:07:50.760There's some academic research that's been happening recently, you know, and then companies like ours that are trying to figure this out to help society.
01:07:59.340But it's still a pretty small pool of people trying to do this.
01:08:04.040I'd like to get into the methodology of your report just a little bit.
01:08:07.140Like one of the things that you that you did was classify accounts according to whether they were typical or atypical and then further subdivided the atypical.
01:08:16.680And that seems like an interesting approach to this problem of trying to make sense out of all these different actors and like different types of actors where, you know, the binary of ordinary human versus bot.
01:08:32.000As we've been researching this, it's I'm starting to feel like that's not perhaps a super useful distinction, a binary distinction.
01:08:39.100And so you guys used, I think, five different categories. Could you talk a little bit about that decision and how you defined those?
01:08:46.680Yeah, and you're kind of hitting the nail on the head for how we were thinking about this, that one or zero, if you're looking at what are people trying to do online as either bot or not bot, typical or atypical, you're kind of missing the point.
01:09:06.860All of these actions are coordinated to make information go viral, right?
01:09:27.380And so we've observed and modeled five different behavioral patterns from the typical user all the way down to what you would consider the bot.
01:09:39.300and are working to understand how they're used online
01:09:45.020to actually get information to go viral.
01:09:47.840So, you know, an influencer has their own specific behavioral pattern
01:10:04.160to make that information get in front of people.
01:10:06.740How do you know when this behavior is coordinated? That was one of the points, the takeaways in the report was, these are not just inauthentic accounts acting on their own. They're all coordinated in a kind of way as a network. How did you determine that?
01:10:20.800Yeah. So that's part of our patented process where so that graph neural networking through the information reveals patterns of behavior. And so it really does. It's like a fingerprint within the data. It actually has a very distinct pattern that we can then see and lift out and then make those determinations.
01:10:40.840And so a lot of that can be, you know, how are they interacting with like accounts? How are they interacting with peers, the language that they are using, the timing? There is a slew of clues that we have identified that we then use to make these determinations.
01:10:57.540One of the criticisms that I've seen of the study is that the data set, the data that you used for this is was not made available. Right. And so if I wanted to take that data set and crunch the numbers myself, I could not do this.
01:11:10.900We had a conversation with somebody from a social listening organization, and they said, oh, well, that's actually kind of commonplace for private companies.
01:11:20.500They might have proprietary things they don't want out there.
01:11:23.580How much did that impact not making that data set available for the public?
01:11:29.260That twofold. So proprietary processes and then like data licenses that we're not actually can make some data available. Right. Because we do have API access to various platforms. So you just can't give away their information.
01:11:48.080Okay, this is such a good point because, and this is a frustration of mine, it used to be that if you wanted API access to X, right, that was not something that would be difficult to get. In 2025, things have really changed. And so, yeah, I don't know if people who are not sort of in this world really know the ways that a lot of the internet and how it works is has been turned into a black box.
01:12:12.800And so it's incredibly difficult to get that kind of access for most of us these days.
01:12:17.480And we simply, you know, just don't have it.
01:12:20.540And just to underscore that, you know, it's it's not a black box because it is so mysterious.
01:12:25.700It is a black box because the people who run the platforms have decided that they want the box to be black.
01:12:32.060Yeah. You know, monetizing the information that they have.
01:12:36.160And I think it's become even probably even worse now, you know, with training LLMs, you know, a good place to get information is from these social media platforms, which then, you know, aren't they're not getting any value off of that, those companies training off of their data.
01:12:54.600So I think it's kind of become a positive feedback loop.
01:12:58.400I was also really interested in the sort of temporal analysis that you guys did, looking at not just the mix of the types of accounts, but how it changed over time pretty rapidly in the immediate days after the beginning of the study period, which I think was like the day after the album dropped.
01:13:18.080Can you talk a little bit about what that temporal analysis and how the shifting mix of actors, what that tells us that wouldn't be available if you just looked at a single snapshot in time?
01:13:30.820It's actually important to look at these as snapshot in time.
01:13:33.960So we can go down from like the microsecond all the way up to, you know, centuries.
01:13:38.440We don't have that much data yet, though.
01:13:41.180But the key here is, so let's say the Taylor Swift report.
01:13:46.220It's actually about 3.7% of the total users were these inauthentic type users that contributed
01:13:55.020That's the big snapshot, but that doesn't tell you the whole picture when you then look
01:13:58.960at it, how it happened over the course of hours, where that 3.7% at the first instance
01:14:04.960of this narrative spiking contributed, it was about 15 to 20% of the total audience
01:14:10.800and contributed, I think it was 78% of the total volume of, right?
01:14:14.780So if you only look at one overall snapshot, you are going to you're not going to see the forest from the trees.
01:14:21.080I'm sure you know that the reaction to this is quite aware.
01:14:26.100Yeah, it's been big. I'm sure y'all have had a had a wild week.
01:14:30.300And it's been really interesting and I think kind of weirdly telling to engage with some of the criticisms that people have made or like just responses that people have had.
01:14:39.860They've been like deeply emotional responses, I guess, is how I will put them.
01:14:43.640And one of the things I've seen, I've seen the takeaways from the report. I don't want to say misrepresented, but I would summarize it as one camp being like, see, 100% of the people who pushed this narrative were bots. And this was a narrative that did not exist for real. If you thought this was real, you got taken.
01:15:05.360And then I saw people like black women on social media being like, I'm not a bot. I'm a real person. I felt X, Y, Z about Taylor Swift. And what's interesting to me is that when you actually read the report, the report does not make either of those claims. And so I guess I wonder, you know, what do you think accounts for the fact that people are using the report to say something that I think the report patently did not suggest?
01:19:14.420how did the report end up in being covered in rolling stone we went out and we shopped it
01:19:20.800around and rolling stones was interested in writing about it like again we were also equally
01:19:25.840giddy about that one we did not anticipate such uh reaction i'm gonna be honest yeah i mean the
01:19:33.540reaction has been absolutely wild and i under i understand that like companies like yours part of
01:19:39.500publishing studies like this like especially like flashy ones ones that people are going to actually
01:19:43.840you know want to be reading uh part of it is like getting publicity for the work that they
01:19:49.520are able to do and so like to that end do you consider this to be a success like i don't know
01:19:56.300the last time that a report i mean we look internally we're reading reports about the
01:20:01.840internet all the time typically my cousin is not texting me about them you know what i'm you know
01:20:07.020yeah uh so yes definitely would consider it a success uh to that point i've had family that
01:20:15.920i haven't spoken to in years and i'm like oh my god yeah uh it was crazy but you know it was
01:20:24.540i guess to the point we were just trying to demonstrate the the ability that we have right
01:20:29.400that was really what we wanted to show people what is happening online and how we can help
01:20:35.100I mean, one possible takeaway here is that, like, there is a lot of appetite among the public and demand for this kind of information.
01:20:44.240Yeah, that's I was kind of kind of were thinking that, too.
01:20:47.120And then what else could we look into that could be helpful for people?
01:20:50.560Can you give us a sense of when somebody gets on the Internet, how how much conversation online is perhaps being impacted by inauthentic behavior?
01:21:01.220Because, you know, from Cracker Barrel to Blake Lively to Taylor Swift, it seems like these conversations that you might have thought of as innocuous are now actually being manipulated by inauthentic actors.
01:21:13.280Yeah. So my whole quote, the Internet is fake, it kind of really is. So since we've been doing this, we have yet to find a narrative that didn't have bot activity or inauthentic activity, like not a single one.
01:21:31.320I mean, it's sort of, I mean, I guess when I read that quote, the internet is fake, I didn't realize you meant it quite so literally.
01:21:40.380It's just kind of scary. Like we, you know, we were a little bright eyed and bushy tailed when we got into this. And then over the course of, you know, doing this work, we've seen just how much inauthentic activity there really is happening online.
01:21:54.480how are we meant to use our social media platforms and and platforms for discourse
01:22:01.080how are we meant to use them effectively if that's the case like like yeah can they be used
01:22:06.180effectively anymore that's a really hard question too and so if i could wave a magic wand the like
01:22:14.000how i would think or fix this um so humans have had thousands of years to figure out the etiquette
01:22:21.480of behaving with each other in real life, right?
01:22:24.360There's so many different norms that we have
01:22:26.800that have just come from those centuries.
01:22:29.760The internet, like in the current state that we use it,
01:22:47.200Yeah, the Wild West is a good way to describe it.
01:22:49.500And especially for these conversations that, frankly, I think even 15 years ago did not feel this heated online.
01:22:59.360It's incredibly difficult to have a conversation, even a conversation about something as simple and innocuous as an album, a movie, something like that, without it feeling vitriolic.
01:23:09.160And I have to imagine this inauthentic activity is adding to that.
01:23:21.920It's actually been really, really interesting and meta for us to watch us become the conspiracy.
01:23:29.180Yeah, that one's going to be a fun white paper we're going to do on ourselves now.
01:23:34.420Oh, my God. You absolutely should. We'll have you back.
01:23:37.480And I mean, that's the thing. When I was preparing for this, I was going through social media and I was trying to pull out some of the, I guess, criticisms maybe isn't the right word.
01:23:45.240But some of the things people have been saying about the report, about your company, some of them from my own background, I know to be incorrect, right?
01:23:52.620Like saying, oh, this is just PR. Taylor Swift probably paid to have them print this.
01:23:56.380And I'm like, well, it's not really what these companies do.
01:23:58.300And I guess, how could I even ask this? Let's say I understand why and how a lot of people who are saying, hey, I'm not a bot. I feel offended and unseen and erased by a report that makes me feel like my voice isn't a real voice.
01:24:17.620Right. That's not what the report said, but I get how that's the conclusion that they're coming to.
01:24:22.400My thing is this. If you are someone who really can't stand Taylor Swift, you think that Taylor Swift is a literal Nazi.
01:24:30.080Let's say that for the sake of argument. I would think that you would want reports like this to make clear how difficult it is for your message to break through online.
01:24:40.520How much that message, even if that's a message that you authentically feel, how easily exploited it is and the fact that it is being disrupted.
01:24:48.880And so I would imagine even people who want to use Internet platforms to authentically engage in, like, critical discourse, they, more than anybody, should want to have a media landscape where that is possible without this kind of inauthentic interference and manipulation.
01:25:06.180No, I completely agree. Not only that, but those individuals that do have those true feelings and, you know, want to use it as a platform to voice those feelings often get taken advantage of by the inauthentic activity or accounts.
01:25:21.740So a very, very common thing is they're not creating the narrative that they want to push.
01:25:27.360They're pulling it from these small populations that they're like, oh, that one, that one will really make people mad.
01:25:32.900And then they push it inauthentically.
01:25:35.080So essentially, they're actually getting taken advantage of by these accounts.
01:25:39.260I've actually seen this or at least suspected that I've been a target of this kind of thing, because you could just be being a garden variety hater on the Internet.
01:25:48.060You know, like, oh, I didn't like this. I didn't like that.
01:25:49.960then you're going to comment from someone who's like yeah let's boycott it or like yeah let's
01:25:53.880like they take it to a level where you're like well i was just trying to engage in a little
01:25:57.740low-level snark i didn't mean it like this but if you're already sort of riled up and you're not
01:26:03.460necessarily thinking super critically about it it is this difficult to not engage i guess is what
01:26:08.980i'm saying no exactly and again that's the point it's really hard not to engage with this when
01:26:15.880either you're really do believe in it or you're really really opposed to it you want to say
01:26:20.760something that's just human nature do you see any solutions to this is there something that
01:26:25.840i mean i don't it's like like what like is there something that platforms could be doing that
01:26:30.740they're not or you know shouldn't be doing what they're currently doing yeah and that's a really
01:26:36.780tough question that um i don't know if we're even equipped to say like what the solution would be
01:26:42.660here there are certain things that are really clear indicators that somebody is doing something
01:26:47.700nefarious um like uh people that are really trying to push uh illicit content will change their
01:26:53.540handle us a lot over the course of just a short time span right how many times did you change
01:26:59.820your handle like never rarely people right that's and these accounts are doing it you know three
01:27:07.180times a day like you know there are there's key indicators that they could be looking for that
01:27:15.000would allow them to mitigate some of this uh inauthentic activity my favorite thing is when
01:27:21.000i see an account that says it has a that that image is of like a white person and then the
01:27:26.620post will be like well as a black woman and i'm like oh did someone switch up their grift
01:27:31.600but what happened yeah those are our favorite two we pass those around where it's like wow
01:27:36.820your your LLM model here really failed you yeah so it's it's tough I mean it sounds like there's
01:27:45.980a lot platforms could be doing but obviously I don't have a direct line to Elon Musk or anything
01:27:51.020but what about individuals like while we are in the absence of platforms really doing what they
01:27:58.100can to to crack down on this kind of thing do you have advice for people especially as we navigate
01:28:03.840what feel like increasingly volatile times where conversation just reflecting the times like it
01:28:10.780just seems more volatile do you have advice for folks as they wade through that to not be
01:28:16.280impacted or or manipulated by this kind of inauthentic behavior yeah um first like take
01:28:23.780a breath, right? Don't respond immediately. But then I think it's, it is because people are still
01:28:32.300going to want to respond. So how do you do that, right? How do you engage with this content without
01:28:35.780actually engaging the algorithm that then is going to boost it to even more eyes? So our goal is to
01:28:43.480try to help people understand how to navigate these narratives. And so in this instance, it's
01:28:48.220kind of like a three-step process. So observe, but don't interact. So, you know, you can see the
01:28:53.440content, but don't like it, don't share it, don't do anything that would boost the algorithm.
01:28:58.620Then if you want to counter, you post separately, but don't reply in the comments, right? So you can
01:29:05.420make your own post about it, but don't be replying to it in the comments, because again, that's going
01:29:09.680to boost it algorithmically. And then try to redirect to a more positive narrative. So in your
01:29:16.180post, talk about, you know, don't engage what they're talking about directly, try to redirect
01:29:21.300into a more positive sense without using any of the same hashtags or the like of that original
01:29:28.360negative content. That way you can be trying to change the conversation instead of boosting
01:29:33.700something algorithmically. What is next for you all? Can you give us a little preview about what
01:29:39.820the next big inauthentic conversation online might be? Sadly, no, right? Because you never know what's
01:29:50.060going to happen online i mean we definitely are going to do one on ourselves because man did that
01:29:54.820go crazy uh but i don't know like if we wanted to do another one maybe k-pop could be a fun one
01:30:01.840oh we were just having a conversation with one of our producers joey about just k-pop fandom in
01:30:09.780general i had no idea i had no idea oh we have we've worked with clients where like they're the
01:30:17.360they're just their online activity have broken our systems so the k-pop community is real and large
01:30:25.100and it just goes to show exactly what you were talking about that these might sound like quote
01:30:30.340celebrity stories but you know k-pop it's it's such a big fandom that it says so much about how
01:30:39.320not just celebrity and fandom and how we live but but like how we live our lives politically
01:30:43.240class issues gender issues these are real things that that really motivate people in our world and
01:30:49.240so it's not just celebrities and fluff it's conversations that have actual influence
01:30:55.120when we see big fandoms getting involved with something normally it's actually in a positive
01:31:02.180sense uh really lifting up a new album or saying how much they like something actually the the
01:31:10.700fandoms really do tend to be like, I want to say wholesome community, like they, they share nice
01:31:17.660content. As someone who studies the internet, and this does spends a lot of time making reports
01:31:22.280about what's happening there. Are you does it give you hope like this, like the conversation
01:31:27.180about fandoms and them making wholesome content? You had a little smile when you mentioned that
01:31:31.760talking about technology and the internet that like, these are things that we like spaces that
01:31:35.680we like. I find myself hating on them and lifting the bad stuff. But do you feel hopeful and good
01:31:41.920about the internet, given all this? No, I think there's still so much to be gained from ever
01:31:48.540increasing connections, right? Because in all of these social media platforms really were about
01:31:55.680connecting individuals together that had like interests. And I think when it became like
01:32:03.020social media and it's now about media content that has driven a little bit more of the negativity
01:32:08.640but you know it really at its core it is about people engaging with communities that have like
01:32:14.300interests and i think that's great our whole goal is to help people have a better understanding of
01:32:19.500the world they live in and especially that information environment where
01:32:23.280it's just really hard to know what's real or not