Most of the time it's hard to know how and why a narrative went viral. Not just to wave your hands at it, but to really know.
There are all sorts of ways to try, of course. Cluster analysis and network graphs and various ways of measuring linguistic similarity and betweenness and mutation, most of which amount to a spoonful of intuition and a wheelbarrow full of bullshit. We think we've discovered some other ways that aren't. But it is a rare thing indeed to look at a narrative virus and know precisely where it began, why it began, and how it spread.
I'd like to tell you the story of precisely that kind of narrative virus.
Our story begins in the afternoon of October 10th in the town of Hoffman Estates, Illinois. It's a pretty nice town about an hour from downtown Chicago on the Kennedy on a very good day. Closer to two hours most days. Hoffman Estates is a classic middle-to-upper-middle class American place with some neighborhoods above that and some below. Think Plano, Texas or Chandler, Arizona if Chicagoland isn't familiar to you. There are a lot of foreign-born and second generation immigrant families there, some from Mexico and many from east and south Asia. There's a little bit of geographic stratification - lower-middle income skewing east and south, higher-middle income skewing west - but not that much. Not really a town with an obvious wrong side of the tracks.
On this day, however, ICE had deployed agents to Hoffman Estates on immigration enforcement operations. They set up in the parking lot of the local police department, which would make a statement later that day attesting to their presence and reaffirming that the department is compliant with Illinois laws preventing the use of local police resources from assisting with or participating in federal immigration enforcement actions. At some point in the course of the afternoon, some American citizens (and other residents, presumably) showed up at the parking lot to protest ICE's presence. A young woman and her boyfriend ended up being pursued in their vehicle by officials to a home on the older, more middle income side of town near its border with Schaumburg. One of those young women ended up with the knee of a large law enforcement officer against her neck before being detained for an extended period. If you managed to miss it, here is the video.
This video was first posted on Facebook at 9:38 PM ET on Friday, October 10th by a woman purporting to be the Aunt of Evelyn, the young woman dragged from the car. She had posted another video earlier at 3:35 PM ET detailing the alleged circumstances of the pursuit and detention. It would have been hard to be certain about very much beyond what's in the video. We see what appears to be a vehicle fleeing from pursuit, but we wouldn't have known if the young woman or her boyfriend had broken the law in that act or before. But by the evening of the 10th, there was generally common knowledge in the Facebook communities connected to Hoffman Estates, Illinois that there was an interaction between federal immigration officers and local residents, confirmed by the local police department and amplified by the videos posted by the woman identifying herself as Evelyn's aunt.
The story took a while to break containment from Facebook and Instagram, where other accounts with a Chicago-area focus slowly picked up on it on October 11th. It wasn't until the video was posted on on X.com at 2:17 PM by a much larger account - that of Project Liberal Executive Director Joshua Eakle - that it really began to spread. In the three days or so that followed, it would rack up more than 11 million views, with millions on top of that from reconstructed versions posted by others. For the first several hours, the response looked more or less like you would expect. A lot of people objected to what seemed like excessive force. Others assumed that the person fleeing was almost certainly an illegal migrant and this is what they voted for. A lot of others asked Grok - xAI's anime waifu-avatared AI model - what it thought about the whole affair.
But there are three unusual circumstances in this case that offer us a rare and special opportunity to see with confidence precisely how narrative viruses spread on social networks in 2025. The first is that Eakle's viral video post included his own transcription of Evelyn's shouts while being removed from the vehicle. "I'm 15," he hears her to say, which many others would repeat. It appears after the fact and after an interview with her parents that she was saying "I'm not resisting." The second comes from the date in question. The video was posted on X.com on October 11, 2025 and concerned an event which took place on October 10, 2025. The third is that Eakle describes the events as taking place in Chicago proper, when in fact they took place in the northwest suburb of Hoffman Estates. I'm not nit-picking - you'll see why this matters in a moment.
Because that's where poor Doug comes in.
Doug's probably a good guy, so I'm going to black out what may be the last name in his handle. But here's where I'm going to ask you to give me license to fill in what will immediately become apparent to you is an obvious gap. Doug searched for "Chicago 15 Year Old Arrest October 11." No, I don't have his search history. And no, I don't know if he did it on Google or with an LLM. But at 5:53 PM ET on October 11, Doug posted the below link as a reply to Eakle's viral post.
That is the first time anyone had posted that link on Twitter in a full year. It's a 30-second read - go take a look. It is a local news article published on October 11 which discusses a 15-year old girl who was arrested in Chicago on October 10 for aggravated assault. See what happened here now? There's just one problem: The news article was published about the arrest of a 15-year old girl in Chicago on October 10th...of 2024. Doug the ham radio enthusiast was doing his best, searched for support for an emerging news story, thought he found it, and shared it. He couldn't have anticipated what would happen next.
There's a thing that happens on X.com when a post goes viral - the reply guys come out. High engagement posts are the perfect opportunity to get into the replies and get exposure to a whole new audience. That's as true for would-be influencers as it is for a guy like Doug. And it didn't take much. Even now as I'm writing this, Doug's tweet has 38 views. Thirty-eight. But it was enough. Again, remember that until this very moment, nobody in the world had posted the link to this article for a year. A minute later, another account used Doug's link to reply to the viral thread. Two others chimed in at 5:56 PM, just two minutes later. Both of these liked it enough to start adding it to other areas of the replies to Eakles's post. One of them, a nurse from New York, added the link as a reply no fewer than six times by 6:02 PM. By 8:45 PM ET, more than two dozen other copies of the link Doug found were littered through the original post with commentary of increasing confidence and stridence. We've solved the mystery! They're twisting the narrative again! That's when the first big account - a 200,000+ follower account focused mainly on Irish cultural debates - promoted the narrative that this video actually portrayed the 2024 arrest of a 15-year old violent criminal.
Throughout this time, no official had made a statement. No new information had emerged. No reverse video search linking the viral clip to the prior event had been located. It would have been a bit of a trick, too, since the location of the posted video is very easily traceable to a certain intersection in a certain neighborhood in southeast Hoffman Estates, while the news article specifically references an address 48 miles away on the south side of Chicago. Beyond that, all we had was a Facebook video from an aunt, some chatter on local Facebook groups, a viral tweet, a guy named Doug and some reply guys who saw an opportunity to spin the story on its head. It didn't take long for other big accounts to join in on the fun.
If that were where our story ended, we might call it a pretty typical case of misinformation on social media and be done with it. But the world has changed since the world of 2020 and 2021 in which every big news organization was launching a "fact checking" arm. The emergence of widely available AI models may be among the biggest of those changes. The integration of xAI's Grok LLM with its social media platform - what most people still colloquially know as Twitter - makes it one of the more useful to examine, not least because the practice of relying on Grok as an all-in-one information resource, fact-checker, and opponent-debunker has quickly integrated itself into many of the platform's subcultures.
In the early stages of the virality of the arrest video on October 11th, Grok was called upon frequently. It's understandable, too, considering that other than the video, there wasn't much detail. No police reports, no news coverage. Nothing. People wanted to know whether all of this was real. Unfortunately, one of the problems with AI models like Grok is that they are, by and large, engineered to please the user. They want to give an answer. So Grok did exactly that, hallucinating from related ICE coverage in the Chicago area from the first time it was summoned to the viral post at 3:30 PM ET until just after 6:00 PM ET. Its responses to users asking for more information came in one of three flavors: it asserted that the detained young woman was arrested for being an illegal immigrant, that she was actually one of two DACA adults who had harassed, obstructed, and harmed Border Patrol agents, or that she was a gang-affiliated illegal immigrant. You can see examples of each below. Note that the model might give any of the three of these very distinct responses to an inquiry - and was doing so quite happily and arbitrarily for about three hours that afternoon.
Starting at around 6:00 PM, Grok apparently had access to enough of the viral tweets highlighting the Aunt's video to begin incorporating a new class of response, this time consistent with social media reports of a 15-year old being detained briefly if rather aggressively by ICE agents. These four competing narratives continued to compete for oxygen until 8:11 PM. That's when Grok's responses to inquiries in this and other viral video threads shifted on a dime. At this point, just about anyone who asked Twitter's AI model what was being portrayed in the video received responses like the below, promoting the narrative that ICE were neither present nor involved and that the video was a recording of the 2024 arrest from Doug's accidental find. In many it would go even further to accuse users of themselves promoting irresponsible descriptions of the video themselves, feeding tens of thousands of social media posts not only triumphantly celebrating the idea that the young woman being dragged from the car had committed various violent crimes in Chicago in 2024, but chiding the authors of the viral video posts for attempting to "twist the narrative" to serve "anti-police" agendas.
If you thought it would end here, well, then you just haven't been paying attention. But if you don't know what happened next, you would never guess. As it happens, the combination of Grok's chiding and the increasingly aggressive posturing of influencers confident they had uncovered an attempt to "control the narrative" found one more receptive target: the Assistant Secretary for Public Affairs at the Department of Homeland Security, Tricia McLaughlin. This is the principal communications coordinator for the administration's immigration enforcement efforts. I am sorry for saying the same thing with different words, but I am going to do it anyway. This is the person who owns the external messaging for the federal agency that yanked that young woman from the car and detained her.
I don't mean to pick on Doug - I'm sure someone else would have decided to perform a similar search at some point. Again, the whole reason we're able to walk through the timeline with such ease is because the age, location, and date of the actual event gave us a magical setup where it coincided almost perfectly with another event precisely one year in the past. And yet we must still grapple with the fact that a random person who accidentally identified a sufficiently useful narrative was able to create a cascade of events that changed the very public position and messaging of the federal agency responsible for the events in question. Oh, sorry, are you still contemplating that Tricia McLaughlin was not lying? Tell you what, I'll concede that there's some small chance she wasn't, but if so she's gonna need to craft a very sternly worded internal memorandum to staff about maintaining control over department equipment and resources. Somehow that gentleman kneeling on the young lady's neck in the video from 2024 got his hands on both a time machine and an ICE Enforcement and Removal Operations vest. OpSec, people, OpSec.
In all that gives us a timeline of events that looks a little bit like a snowflake triggering an avalanche.
I think there are a lot of things worth paying attention to in this episode.
First, the notion of social networks as marketplaces of ideas, arenas where we work out what is true and what is not, is not just a stretch but almost diametrically opposed to what they really are in practice. Social networks are not some lofty lyceum. They are a petri dish for the spread of narrative viruses, an almost comically perfect environment to emphasize the usefulness of stories over their accuracy. Do you truly think the tens of thousands of people who have drawn conclusions about the events that took place here, or the rightness of immigration policy, or the malevolence of their political counterparts in the crafting of manipulative narratives are going to change their minds if they read this piece? If they discovered that the video was, in fact, legitimate and current? That it was, in fact, an ICE operation? If so, you're dreaming. They've all moved on. The conclusions are fixed because they were never about the truth but the usefulness of the underlying stories. There is no revelation of fact which could or will change their usefulness at this point.
I think we've also learned that the critical mass of truth or influence necessary for an idea to become globally influential is vanishingly small. Infinitely smaller than I think any of us would have imagined - and the interaction of social media algorithms and the current state of AI models sits at the center of the explanation for it. You can tell me that the Doug Origin Story is also a just-so story and sure, fine. But the viral spread of this narrative was unequivocally driven by randoms finding a useful narrative with no special argument, insight, expertise, or reach, which was then amplified by the combination of reply-guy mechanics and AI reinforcement to reach the center of the political zeitgeist for a day or two.
There is another cautionary tale to be told about AI's reinforcement of emerging narratives, too. When most people talk about LLM sycophancy, they are usually concerned about mental health, bad advice, and garbage outputs. But the combination of arbitrariness and confidence creates a special problem in social media in our present political moment. The survival advantage conferred to a politically useful narrative shows that the risk of flippant confidence is not garbage output - it is supremely confident and bifurcated output which entrenches political division and separate realities. These two completely oppositional responses from Grok were posted within minutes of each other this morning, October 15th.
And don't get caught up on Grok. All of this is happening every day with OpenAI, Anthropic, and Google models, too. As we look forward, the risk of model collapse - models training themselves into oblivion on their own slop - isn't that models become less useful but that they become even more fundamentally prone to support bifurcated constructed realities. All the while, for most of us, all of this is happening in a paralyzing field of uncertainty. We are faced with the choice between lacking all conviction and being full of passionate intensity. When I see a video or an image, I doubt it immediately. I feel completely ill-equipped to speak to its veracity and terrified to stake my credibility on its authenticity. More often than ever I find it necessary to ask myself what I need to be true and to challenge it. More than ever I find it hard to justify how I spend my time online against what I know that social networks are doing to the way that I think.
And I know that I'm not alone.
