The "bot farm" that wasn't

In which The Daily Beast attempts to report on social media manipulation, but winds up generating disinformation of its own instead

In early 2023, The Daily Beast published two articles alleging the use of various underhanded social media tactics by the DeSantis presidential campaign (“Ron DeSantis’ Secret Twitter Army of Far-Right Influencers” and “How Ron DeSantis Is Taking a Page Out of Nixon’s Playbook”). Peppered with phrases such as “bot farm” and “online army”, these articles by campaign reporter Jake Lahut contain bold claims about various social media activity, including an allegation that pro-DeSantis online operatives posted sexual imagery involving children. If true, this would likely be a serious crime, but this claim along with several others falls apart under scrutiny. Although social media manipulation is common in politics today and it would not be surprising if the DeSantis campaign were indeed engaged in shady behavior, these two articles fail to document anything of the sort, and most of the key claims would never have seen publication had The Daily Beast done a modicum of due diligence.

as it turns out, these accounts do not in fact have a 24/7 pace of tweeting

The saga began with a January 25th, 2023 article with the title “Ron DeSantis’ Secret Twitter Army of Far-Right Influencers”. The term “bot farm” is repeatedly dropped over the course of forty meandering paragraphs, but actual evidence of a “bot farm” is virtually absent (regardless of whether the term “bot” is being correctly used to describe automated activity or more colloquially to describe any sort of inauthentic account). All but two of the social media accounts specifically mentioned are the X (formerly Twitter) accounts of reasonably well-known right wing influencers posting manually under their own names, however, and therefore not “bots” in any sense of the word.

These schedule plots were generated at various points in time by users of makeadverbsgreatagain.org. Although the tool is now defunct, these charts were easy to track down due to the magic of log files.

The only two exceptions are @MaxNordau and @zuoying19, both of which are alleged in the article to have a “nearly 24/7 pace of tweeting”. This is false, however; schedule plots generated at various times over the last year indicate that the operators of both accounts take breaks for sleep and the @zuoying19 account isn’t even a particularly prolific poster. The degree to which the accounts “relentlessly parrot the same phrases” has also been exaggerated, and in the absence of other indicators of inauthenticity isn’t particularly meaningful. Since these are the only two accounts described in the January 25th article that even potentially qualify as “bots”, and the evidence of their inauthenticity does not hold up when examined, the general allegations of a “bot farm” presented in this article are merely rumors and the specific allegations are untrue.

a few minutes of fact-checking could have saved The Daily Beast from falsely portraying an AI-generated face as a real child

The tale of alleged pro-DeSantis social media shenanigans drifted further from reality in an April 12th article (“How Ron DeSantis Is Taking a Page Out of Nixon’s Playbook”), which cites the January article as background material. The first four paragraphs are dedicated to a bizarre anecdote wherein a Texas schoolteacher was supposedly attacked online by pro-DeSantis operatives who produced a “sexually graphic meme” of one of the teacher’s daughters. All of this is false, however: the image was a GAN-generated image of a male child from thispersondoesnotexist.com, not an image of a specific person, and nothing about it is “sexually graphic”. The most obvious sign of the image’s synthetic origins is the implausible teeth; the positioning of the facial features is also the same as one finds in all unmodified StyleGAN-generated faces. (The phrase “Mr. Golden Balls”, the alleged “porn-style message”, is actually the name of a stage magic trick.)

This August 2022 post on Steven Jarvis’s blog confirms that the image that The Daily Beast represented as Jarvis’s child is this GAN-generated face. Contrary to Jarvis’s assertions, faces generated by StyleGAN are not based on any specific human being.

More disturbingly, the source of this lie and other key claims in the article, Steven Jarvis, is a hoaxer and serial harasser who, far from engaging in disinformation research as he proclaims, is himself a high-volume source of online disinformation and abuse. By the time The Daily Beast published his falsehoods, he had written over 1300 lengthy Substack posts laden with bizarre conspiracy theories and myriad allegations of imaginary crimes. A quick review of this material should have prevented The Daily Beast from treating Jarvis as a credible source, but astonishingly Jake Lahut published Jarvis’s claims without verifying anything, and specifically agreed to a request not be fact-checked by then-Daily Beast contributor Will Sommer (now at the Washington Post). Sommer and several other journalists who cover disinformation are among the regular targets of Jarvis’s lies and harassment, which should have been yet another red flag about running this particular article.

a real live Daily Beast journalist received this wall of text and decided “I’m going to use this person as a source without verifying ANYTHING!”

The bogus claim that a GAN-generated face was an image of a real child is not the only one of Jarvis’s lies that The Daily Beast published in the April article, which in its original form falsely accused two left-wing X/Twitter users (@KassandraSeven and myself, although my handle was misspelled) of harassing people on behalf of Ron DeSantis. Rather than reaching out to either user before publishing these lies, The Daily Beast’s Jake Lahut made the inexplicable decision to seek comment from the Ron DeSantis campaign instead. (In yet another bizarre twist, my handle and only my handle was removed immediately after I pointed out the issues with the article, but the article remains online and none of the other false claims have been corrected as of the time of this writing, over 8 months later.)

Although debunking all of the falsehoods in Steven Jarvis’s massive corpus of Substack posts would be impractical due to the sheer volume, here’s a brief look at a few of the issues with his alleged “research”. One of the most pervasive problems is that Jarvis continually misidentifies the real people behind many of the social media accounts he writes about and accuses them of imaginary felonies. These assertions are presented without supporting evidence, and since the burden of proof lies with the person making the claim, they can be reasonably dismissed without additional debunking. (The additional debunking is not difficult, but is omitted from this article to avoid exposing the individuals Jarvis has targeted to further harassment.)

this is why reverse image search can be so important

In a December 11th, 2022 blog post, Steven Jarvis confidently assessed that the black-and-white photograph above was generated with AI. A quick reverse image search reveals that this is a very real photograph of a very real realtor, however. This is the inverse of the false claim he convinced The Daily Beast to publish about an AI-generated face supposedly being an image of a real child, and is part of a recurring pattern of Jarvis falsely claiming that real photos are fake.

calendars, how do they work?

On February 27th, 2023, Jarvis wrote a post alleging that a random discussion on Twitter was some kind of preview of the 15th installment of Matt Taibbi’s “Twitter Files” (published January 27th, 2023). This conspiracy theory hinged on the claim that the discussion in question was three days prior to this “Twitter Files” release, but it actually took place on January 24th, 2022, placing it more than a year earlier. (Trouble with chronology is a recurring theme; Jarvis confuses the years 2012 and 2021 in multiple posts containing bizarre theories about John McAfee, and he seems to believe that ISIS, which was founded in 2013, carried out the 9/11 attacks in 2001.)

All of this is only the tip of the iceberg; other bits of Jarvis’s fabulism include assertions that taking a screenshot of a post when it has 666 likes is evidence of Satanism, major social media platforms have no employees who work weekends, basic reverse image searches require access to secret military technology, and monkey pox is a bioweapon clandestinely dispersed by Russia at LGBTQ social events.

What are we to conclude from all of this? Whether intentionally or otherwise, The Daily Beast wound up creating disinformation rather than reporting on it. Both articles in this series appear to have been written by starting with the conclusion that the rumored DeSantis “bot farm” was real, and little to no effort was put into ensuring that any of the alleged evidence presented in the articles was accurate or supported the desired conclusion. This is an inversion of how coverage of disinformation and adjacent topics ought to be handled: evidence-based investigation should come first, bold claims from sources should be treated with skepticism and vetted thoroughly, and specific allegations and conclusions should only be published if and when a solid case can be made.

Footnote: It is important to keep in mind that the lack of valid evidence in this pair of articles doesn’t prove that there is no DeSantis “bot farm”; it simply means The Daily Beast didn’t find it or report on it in this pair of articles.