EMMANUEL DZOTSI: Today’s episode is brought to you by that moment when you get to a subway platform at night and you’ve just missed the train and you think back to all these moments in your day where you wasted 30 seconds like if you’d just not played Candy Crush for five minutes in bed when you woke up you’d be on your way already. Instead, you’re sitting there telling your date you’re gonna be late to a place you picked that is far away from where they live. I’m never gonna get married if I keep playing Candy Crush.
Okay, here are the real ads.
From Gimlet, this is Reply All. I’m Emmanuel Dzotsi.
For the last several months I’ve been fascinated by one specific Twitter account. It belongs to a woman named Christina Pushaw who in May became Governor Ron DeSantis’ latest press secretary.
Christina Pushaw is in her early 30s. Her Twitter bio will tell you that she’s here to debunk false narratives about Florida and DeSantis, that she’s anti-Communist. She’s not from Florida. She’s lived there less than a year but you wouldn’t know that from the gator next to her display name.
If you put her name into Google, you’ll see that she has been very busy since starting her job in June. Busy embarking on an 18-hour-plus, 200-tweet spree against an AP journalist that led to death threats against him. Busy tweeting about Kyle Rittenhouse post acquittal saying, “welcome to our free state and enjoy your time here.” Busy tweeting out anti-Semitic conspiracy theories about vaccine passport apps –– she’d later say she was being sarcastic about that last one.
Anyways, it’s a lot of tweeting and a lot of constant fighting, all day every day against people who oppose her or DeSantis. The classic Trumpian deal.
But the thing that really had people in Florida freaked out was that Christina wasn't doing this fighting alone, she seemed to have assembled some sort of digital army to come to her aid in a way they’d never seen before, a formidable force she’d summon even for the most boring of political fights.
ANNA ESKAMANI: Hi, is this Emmanuel?
EMMANUEL: Yes, yes, this is he.
EMMANUEL: Hi, is this Rep Eskamani?
ANNA: Yeah, it is. It’s Anna. Hi. And then turn on Concord Street, yeah. Turn right there. I’m in an Uber right now.
EMMANUEL: Oh, you’re okay.
ANNA: Um, this could be your B-roll.
The first person I spoke to about Christina was State Representative Anna Eskamani.
EMMANUEL: Obviously I'm calling you because of, of Christina Pushaw. And—
EMMANUEL: … um, I spent, like, a couple hours on Twitter. [laughs]
EMMANUEL: And saw, like, a lot of interaction. I was like, oh, it's like—it’s kind of wild on Florida political twitter!
ANNA: [laughs] Wild is like, a nice way to put it.
Anna’s battle with Christina was over a budget item. Governor DeSantis had vetoed a bunch of money for LGBTQ+ youth and for survivors of the Pulse nightclub massacre, which happened in Anna’s district. The governor was getting a lot of blowback and Anna felt like Christina was trying to spin the story on Twitter.
ANNA: I tweeted back, and I, and I basically schooled her on, you know, the process of how the budget works and what the governor did, and it was really interesting. While this was happening, someone actually DMs me, sends me a direct, direct message, and warns me, and says, "Hey, now that you've responded to Christina, I just want you to know that like, they're all gonna come for you." And I was flooded [MUSIC] with all these accounts coming after me, and accounts with no real name to them. Like, they [crosstalk].
EMMANUEL: Oh, like what kind of stuff?
ANNA: Like, they’re anonymous accounts. Like, there’s—like, there’s no photos of people on these accounts. Like, they just, again, as far as I can tell, they almost—they, they really present themselves as like, anonymous accounts because, um, you don't know who really operates them.
Anna was seeing accounts with these generic, like, Team DeSantis-y kinda names.
ANNA: You know, DeSantis is the best, like, 2000 or something.
And they were all using the same messaging points. Like, to Anna, it felt weirdly like it was scripted by Christina.
ANNA: And it's just so funny because, in this one circumstance, Christina's perspective is very, very narrow and also wonky. Like, those messaging points are not something that an everyday person would even, uh, really resonate with. ‘Cause it’s so, like, insider baseball.
When Anna replied to one of the accounts asking, "Are you Christina Pushaw?", the account didn't respond. Instead, it deleted its tweets.
And it wasn't just Anna. Like, this was the thing a lot of regulars on Florida political Twitter, folks from all over the state, were noticing.
JOHN FOX: There's a level of coordination that certainly, I've never seen in Florida politics.
PAT DELUCA: If you tweet something negative about her, all of a sudden, you will get a whole bunch of replies that seem inauthentic, to say the least.
PAT: Yeah, meaning they’re, they’re pictureless profiles. The names of the profiles are, you know, strange. You click on the profile, you find that it was created a month ago.
EMMANUEL: But that doesn't necessarily mean that they're not real people, as much as that just like, they’re not typical accounts, right?
PAT: It doesn't mean that they're n– it's, it’s just—it’s a, it’s a very bizarre pattern MUSIC if you look into it. And it’s something that didn't happen in Florida politics before Christina Pushaw showed up.
Sometimes it just seemed to be keywords that set these accounts off. Like, God help you if you were a journalist or a Democrat who said “DeSantis,” “COVID,” and “failure” in the same sentence.
Like Christina might come along and say something... like call you a media activist or “ Blue Anon,” but even if she didn’t show up, it still felt like there was somebody or something out there just like, scanning the web for critics.
And so, people started asking a question that just months ago would have felt totally absurd: Could it be that the governor’s press secretary was running Twitter bots?
PAT: I have a feeling that she has botnets that they work with. And by “they”, I mean her and the DeSantis administration, to try to promote whatever agenda best suits Governor DeSantis at any given time.
The idea of a press secretary for a sitting governor using fake accounts to scare her opponents? That seemed wild to me.
Like, for one thing, most other press secretaries for republican governors, they’re not commanding big presences on Twitter, let alone brazenly using some sort of digital army of suspicious accounts.
But as people in Florida kept pointing out to me, Christina wasn’t like a lot of those other press secretaries; she’d gotten a lot of her political experience in eastern Europe, working with a former president of Georgia, you know, the country.
And now whatever Christina was doing, she was not apologetic about it, she was out in the open following these accounts and retweeting them.
And from what I could tell, all of this was having a real impact ––intimidating and silencing critics and political opponents. Why join a debate about, say, DeSantis’ COVID policies when you know you might get death threats?
Christina wouldn’t agree to a recorded interview for this story, but I spoke to her. She denied that she runs bots. She doesn’t condone anyone getting death threats and says she hasn’t ever encouraged it.
For me, though, a journalist obsessed with political Twitter— I just wanted to know, like, what’s actually possible here? So, I decided to look into it and figure out exactly what was going on.
But first, I needed some help.
That’s after the break.
– BREAK –
[PERCUSSION IN AND OUT]
Welcome back to the show.
So, if I’m being honest, when I started reporting this story, I didn’t know that much about bots.
Like, I didn’t know how I was gonna possibly figure out whether Christina was using them, unless she was like, openly paying for bots on Venmo or something.
And just to say, when I talk about bots, I don’t mean like, trolls. I mean non-human, automated accounts programmed to tweet specific things and reply to people.
How I was gonna differentiate the two was not exactly clear to me. So, I decided to call another reporter, someone who spends a lot more time in this space. And who’s actually been on this show before.
JOSEPH COX: I'm Joseph Cox. I'm a senior staff writer at Motherboard, which is the technology section of Vice.
EMMANUEL: Nice. I feel like there are Reply All listeners who are listening to your voice and listening to my voice and being like, “That guy's English accent is much better.”
JOSEPH: No, I wouldn't say that. [laughing]
When I told Joseph about Christina and her alleged bot activity, he told me about an incident that sounded very similar to what I’d heard about in Florida.
Back in 2017, Joseph wrote an article about some researchers who’d been reporting on Russian interference online. Essentially, these researchers who’d written about Russian bot activity in the lead-up to Charlottesville were now being attacked by those same bots on Twitter.
Joseph wrote the article, published it, tweeted it out, and went about his day. But then he started getting a whole slew of Twitter notifications—like, way more than usual.
JOSEPH: I started seeing something like hundreds and thousands, and it was going up and up and up. And I simply—I couldn’t keep up.
EMMANUEL: What did you think when you, when you saw that you had a hundred? Did you just—yeah.
JOSEPH: I mean, I assumed that, oh, maybe I got tagged in, in, in somebody’s popular tweet. You know, maybe a big news organization or a big commentator had tagged me, or something like that. I didn’t think it was going to be one of, one of my tweets. You know, they’re not that funny. [laughing] Like, that was—that, that, that was definitely not my thought.
So, Joseph went to go look and see what was going on.
JOSEPH: It’s people following me, unfollowing me, and retweeting, uh, this article just incessantly.
EMMANUEL: Wow, they were—so wait, accounts were following you, and then subsequently, those same accounts were unfollowing you?
JOSEPH: Yeah. It was the same accounts—
JOSEPH: … following and unfollowing. But then, with, I think, my tweet of my own article, uh, about, uh, the, the researchers’ work, it got rapidly retweeted, like, thousands of times.
For Joseph it was clear. The bots that had attacked the researchers, the bots he’d written about, they were now going after him.
And then, just as he was looking at all of this stuff, Joseph got logged out of Twitter, and for some reason, he couldn’t log back in. For a second, he was afraid—thought maybe he’d been hacked. But it turned out Twitter had temporarily suspended his account. Twitter never actually confirmed why they did this, but Joseph thinks it’s because they saw all of this activity and thought that maybe he’d bought accounts and artificially boosted his article. He thinks it’s possible that may have been the goal of the bots in the first place—to get him kicked off Twitter.
The whole episode was really unsettling for Joseph. It was clear he’d been targeted for writing the article, but it also taught him something about what bots could do.
JOSEPH: They were pointing at me, and they wanted me to know, “Hey, we read that, and we know your account, and this is the sort of, um, power that we have on social media.” I just found that as a, as a very interesting flipside to the common idea of bots, which is, oh, let’s just—put a message on blast. No, no. It can be for a very, uh, personal or political and targeted reason as well, potentially.
EMMANUEL: That's really creepy.
JOSEPH: Yeah. Yeah, it, it, it is.
But as creepy as all that was, it happened back in 2017. Like, Joseph could figure out those accounts were bots pretty easily. But since then, bots have gotten a lot more sophisticated.
JOSEPH: Nowadays, they are much better. They're much better looking. You know, it, it will have a profile picture actually of a person.
JOSEPH: Uh, and if it's not one that they've stolen, it's one that they may have algorithmically generated, you know, with a website such as This Person Doesn't Exist, which uses machine learning to generate a new image of a person that didn't ever exist before. You know, and plenty—
EMMANUEL: Wait, wait, wait. There's a website called This Person Doesn't Exist that just like, creates fake people?
JOSEPH: So, if you go to ThisPersonDoesNotExist.com?
JOSEPH: You visit that.
EMMANUEL: Let me do that right now.
JOSEPH: And then you will immediately get an image—
JOSEPH: … uh, of a person.
EMMANUEL: Yeah, I'm looking at, uh, an image of like, a pretty—yeah, a pretty normal looking, very friendly-looking white guy.
JOSEPH: So, that person's not real. [laughs]
EMMANUEL: Wait, what? [laughing]
JOSEPH: The, the program, uh, underlying this website has used machine learning to generate entirely new images every time somebody visits. And of course, it's doing this by processing images of real people finding their, you know, identifying features that, oh, this is what lips look like, this is what hair looks like. Uh, this is where eyebrows are supposed to be. And then it's using all of that information to make a new face. And this is what at least some bot armies are certainly doing now. I mean, the reason that they—the researchers have managed to pin that down is because sometimes, albeit the one I'm looking at right now, of a woman with black hair and brown eyes, like, it looks super realistic.
JOSEPH: Um, sometimes there can be some telltale, uh, signs that it's wrong.
JOSEPH: Sometimes the ears look a little bit off. Or sometimes—
EMMANUEL: I was gonna say, as—
EMMANUEL: … as you say that, now that I'm looking at this man, like, I am noticing that like his ear lobe is—
EMMANUEL: … separated from his ears. But I would never have noticed that on first glance. I would never have even [JOSEPH: Yes — ] looked at it if you hadn't pointed it out.
JOSEPH: Right. But this is what bots are using. And then if someone would have made a fake account on LinkedIn or Facebook or Twitter and used one of these images, I mean, come on, like eight, nine times out of 10, somebody is not going to be able to notice.
EMMANUEL: Oh yeah. I wouldn't—yeah, I wouldn't have thought to notice.
JOSEPH: Right. And that's going to make automated attempts to police this sort of activity from sites like Twitter or other social, uh, media companies harder because you can't just do, you know, a reverse image search and say, oh, this is a stock photo taken from the AP or Getty or something, because this is an entirely new photo. And, and it is—it's a legitimately produced new photo as well. It's just the person doesn't exist.
Joseph told me that what he knew about bots was just the tip of the iceberg. So, I set about looking for bot researchers.
And you might know this; I didn't—there are services that claim to be able to tell you whether an account is a bot or not. It's a whole industry.
And the gold standard is this online service called Botometer. It kinda felt like we could just like, run the accounts in Florida Twitter that seemed suspicious through it and then I’d know for sure which accounts were bots.
That’s what I was thinking..but then I called an expert, this guy Darius Kazemi.
EMMANUEL: Hi Darius
DARIUS KAZEMI: Hey.
EMMANUEL: How you doing? Uh that is quite the…
Darius is a computer programmer and researcher, knows a lot about bots, For years he specialized in making what he calls artbots –– bots that do things like tweet out metaphors all day long. And he gave me some bad news about Botometer. He says that in his experience it's not a sure thing.
DARIUS: I, I would try on a bunch of my own Twitter bots and it would be like, "Oh, likely not a bot," and it's like, "No, I know this is a bot. I made this bot. It, it is an automated account."
Darius told me that when Botometer was created, what they did was they had a bunch of grad students sit down and basically flag Twitter accounts as bots or not. And the algorithm learned from the students.
DARIUS: I started digging into that lab's research. And as best as I could tell, there were a lot of false positives and false negatives on this. Way more often, it was humans being marked as bots.
DARIUS: And sometimes it was just—it was often just people who didn't use Twitter in the way that, uh, you know, a, uh, a grad student at a California university would use it.
DARIUS: Um, so it was people in foreign languages, or people posting from games they were playing, or what have you.
I reached out to Botometer about this, and they defended their work. They said that the students were given a very extensive rubric to help identify accounts, and that Darius’ criticism is misleading, in part because it largely speaks to older versions of the software.
But even the folks at Bot-o-meter admit that bot detection is not an exact science, they knew that when they built it. It’s why the software doesn’t actually tell you outright if an account is a bot or not. Instead, it just gives you a probability and leaves it up to you to decide the rest.
Talking to Darius it just felt like, I dunno man, if bots are this hard to identify, if they’ve gotten this advanced, it actually feels possible that what was happening on Florida political twitter could be bots.
But here’s the thing. Darius told me that when it comes to the sorts of things Christina was accused of doing, like bots attacking opponents en masse, responding to every tweet that mentions “DeSantis,” “COVID,” and “failure,” Twitter’s pretty good.
EMMANUEL: Could you program a bot to basically scour the internet looking for something and like, attack it in that way?
DARIUS: Yes, you can. Uh, I think Twitter would catch it very quickly.
EMMANUEL: Oh, really?
DARIUS: Twitter catches that really fast ‘cause it's spam.
It’s not that spam pushed by bots on Twitter doesn’t still happen, Darius said. But if were to happen in a really big way—say, a coordinated automated bot army consistently going after critics of DeSantis—Twitter would probably flag the activity.
Still, though, I’d seen a bunch of accounts that looked suspicious to me. I had a spreadsheet, like, full of them, and I started showing some to Darius.
EMMANUEL: So, I just want to share—I'll share a doc with you. We have a bunch of different documents… [FADE OUT]
We talked about the sorts of accounts I’d seen how there were accounts that only seemed to respond to tweets with, like, weird statements. Like, there was this one account that would only respond to stuff saying, like, “Hmm” or “Sad.” There were accounts that had just been made and seemed to only tweet about Christina or DeSantis. There were also kinds of accounts that just had a string of random letters and numbers as usernames.
I showed him one of those: the account @d33p782047. I’d actually seen it get accused of being a bot in real-time, and I’d thought its response—“I’m just a pissed off Hoosier that admires Christina"—and its statement that: "You cannot believe people like me are real” seemed pretty contrived.
Darius squinted at the name.
DARIUS: I think the handle is actually, uh, meaningful. Uh, D33P is deep, and then I think the 7h2047 might be like, um, uh, uh, uh, uh, thoughts? Like T—maybe. I don't know. But—
DARIUS: I, I, I think, I think there's actually something, um—there seems to be—
EMMANUEL: There’s something symbolic about that, actually?
DARIUS: There seems to be something symbolic or some kind of—oh, uh Deep Throat! That's what it is. Yeah—
DARIUS: OK, well, D33P is deep.
DARIUS: Because E, because three is E, right? Um, and then 7—if, if you—there's an H in there, right? 7H, right?
DARIUS: Um, uh, 7 could easily be a T. So if we say that that’s TH.
DARIUS: But yeah, Deep Throat, right? Like, like if I see deep and then TH, and it's a political account, it's likely to be a, a Deep Throat reference, [EMMANUEL: Right.] essentially.
DARIUS: Um, so I, I wouldn't—that's not, that's, that's a—that is a human, handpicked handle right there.
DARIUS: That's not random numbers and letters. And, you know, honestly, this tweet about “I'm just a pissed off Hoosier”? I, I, I believe this person. And honestly, I think they're kind of right, too, you know? Like, I think a lot of people don't want to believe people like him exist.
DARIUS: But they really do. [laughing]
Darius told me that the surefire way to know if an account is a bot or not is to just DM the account. So I DMed good ol’ Deep Throat, asked them, “Are you a bot?” And they got back me, told me they were real.
“I’m not really interested in an interview, but thank you for the offer,” they wrote, before also thanking me for following them.
I was feeling a little bit like an idiot. But Darius said it totally made sense that really anyone — me, the people in Florida, you,— might think accounts like deepthroat’s were bots. Maybe it's even easier for us to think that.
Like we’ve all been hearing for years about how our elections have been impacted by foreign interference, about how disinformation is spreading like wildfire on the internet and ensnaring seemingly reasonable people, and to Darius, that's exactly why bots have become the perfect boogieman.
DARIUS: I mean, I think it's a classic moral panic. Um, just like people think that, uh, you know, there's a, a massive increase in crime in crime all the time, even when you look at the numbers and there's—the crime numbers have been going down in many places. It's an easy out. It's an easy way to point at some nefarious other, like the, the bot makers and like, oh, they're the reason democracy is failing.
Obviously, I can’t definitively rule out whether Christina is using bots. But the more I looked into Christina Pushaw herself, the more I became convinced that bots don’t explain her effectiveness in Florida.
I’ve been following Christina n Twitter for months now, seeing her casually pop up in my feed morning and night. I’ve sifted through whole days of her tweets. And at this point, I don’t think it’s just that she’s another culture warrior.
Here’s my theory –– I think Christina is someone with a very specific skill, a skill that any natural politician has. She has this knack for noticing people who otherwise are invisible to the rest of us on Twitter.
Take this one 18-year-old high-school kid from New York. Back in February of this year, months before she got the gig as DeSantis’ press secretary, this kid DM’ed Christina on Twitter. He’d seen an article Christina had written defending DeSantis and blasting his media coverage.
And that kid, Samuel, he was trying to write his first real article about politics. So she helped him—gave him advice. And even after that, she stayed in touch with him as he started a troll account to mock DeSantis’ political opponents.
EMMANUEL: Is this fun for you?
SAMUEL BRAVO: Like, running the account and posting on Twitter and stuff?
SAMUEL: Yeah. I love it, I love it. It's, it's great. It's just, it’s entertaining. The, the content and, and when it blows up, and every so often, the people who I'm mocking will see the tweet and like, respond, and so I have fun with that.
Samuel’s account specialized in posting screenshots of tweets where people criticized Governor DeSantis or complained about him. It seemed to operate almost like a homing beacon, like, letting its followers know, “Hey, come mock or confront this person.”
He named the account DeSantis Derangement Syndrome. And Christina was a fan. She took Samuel seriously, treated him like an adult, as a peer. Gave him a front-row seat to a world he wanted to join really badly. Like, one day in May…
SAMUEL: I just remember her messaging me and being like “I have big news, you're gonna like it. And then she tells me that she’s gonna be like, the press secretary. And I was like, “Oh.” Like, whatever I said, it was like, in all caps. It was like, “No fucking way."
In her new job with the governor’s office, Christina would boost the account, retweeting him when he had a funny tweet. When she wanted to troll someone on Twitter, she’d tag him and be like, “Here’s some content for you.”
And often she wouldn’t have to tag him...over time Samuel actually developed a twitter list of a bunch of DeSantis’ critics that he’d monitor. Anna Eskamani was on it, you know, the representative who’d argued with Christina about funding for the pulse nightclub massacre survivors. Turns out. It was actually Samuel’s account, DeSantis Derangement Syndrome, that Anna had tweeted at originally, asking if it was Christina.
And as Samuel’s account grew more and more popular, growing to over 50 thousand followers. it was as though he’d gained entry into some cool club of people who all had connections to Christina on Twitter.
Suddenly, this high school kid was in group chats with all these other troll and anonymous accounts, and they moved like a clique. When Christina got into it with someone, they’d all often descend.
EMMANUEL: I have to say, from the outside, it all looks really coordinated. Like, all of you are kind of like, in league. It's very clear you all know each other, all of these accounts—
EMMANUEL: … and that like, you're in league with each other?
SAMUEL: What, like Nordau and Redfern and Christina and my account?
EMMANUEL: Yeah, and like, also just like, the way in which, like, it seems like the anonymous accounts, like DeSantis Derangement Syndrome or Pushaw and Co, the way in which like, you guys revolve around Christina online.
SAMUEL: Yeah, yeah, we talk, we talk. Um, I talk with Christina, I talk with each of them. I've been on a Space every so often.
EMMANUEL: Oh, like a Twitter Space?
SAMUEL: Yeah, that’s—recently I've, I’ve done that a few times.
Talking to Samuel made me wonder like, just how many people feel connected to Christina like this. From what I’ve seen, that number is really high.
Like, just a few days ago, I saw Christina reply to an account I’d actually once seen people accuse of being a bot—the sort of account that is always in the replies to Christina’s opponents’ tweets, always parroting what she says or retweeting Christina’s content.
The account, I’ve got to say, is a real person. Her name is Carolyn. And she tweeted that her mother had just died. And Christina wrote to her and offered her condolences.
It was really obvious how meaningful this was to that woman. When big people like Christina notice you and go out of their way to interact with you, that’s an awesome feeling.
Makes you feel like you’re part of something with people who get you—people you’re willing to go to bat for. Whenever and wherever they need you, you’ll be there.
That’s Christina’s secret. That’s what I think has unleashed this wave of so-called bots on Florida political Twitter. She is someone who spends too much time on Twitter, develops fierce connections with other people who spend too much time on Twitter, who all believe the same thing—that if you fight, you win.
[MUSIC CONTINUES, FADES INTO REPLY ALL END THEME]
This episode of Reply All was produced by Noor Gill, Sanya Dosani, and Phia Bennin. It was edited by Tim Howard. And, of course, the episode wouldn’t have happened without the rest of the Reply All team. Our intern is Esperanza Rosenbaum. It’s actually Esperanza’s last episode with us and I just wanna say, working with Esperanza has been an absolute pleasure. She brings such great energy and insight into every room she’s in. Seriously, thank you so much Esperanza.
The show is hosted by me and Alex Goldman. This episode was mixed by Rick Kwan, with fact-checking by Isabel Cristo, music and sound design by Luke Williams. Additional music by Breakmaster Cylinder, Tim Howard, and Marianna Romano.
Special thanks to Jacob Perry, Sarah Rumpf, David Quinones, Mac Stiponovich, Rustam Yulbarisov, Sopo Japaridze, Bryan Gigantino, Kathleen O’Connor, Susan Roll, Khatuna Chomakhidze, Reyhan Harmanci, Lydia Polgreen, and Hannah Chinn.
Thank you all for listening, see you next week.