Nieman Foundation at Harvard
HOME
          
LATEST STORY
From shrimp Jesus to fake self-portraits, AI-generated images have become the latest form of social media spam
ABOUT                    SUBSCRIBE
Aug. 30, 2023, 2:06 p.m.
Audience & Social

With five old phones and some Pew data, the BBC’s Marianna Spring monitors social media from the inside

“The reason I have to have undercover voters is because social media sites won’t — and to some extent, can’t — tell you exactly what they’re recommending every single voter.”

Marianna Spring realizes it’s a little bizarre to carry around five old cell phones.

Spring, the BBC’s misinformation and social media correspondent, keeps them in a tote bag, and uses stickers to distinguish them. It’s challenging, among other things, to keep them all charged, she tells me in an interview. But for nearly a year, that clunky, antiquated tech has served a cutting-edge reporting purpose for the 27-year-old London-based reporter: helping her get inside the black box of social media algorithms, and get a firsthand sense of what they feed to American voters with different political and personal profiles.

Since September 2022, just before the U.S. midterm elections, Spring has maintained social media accounts that correspond to five different “voter profiles” she developed using Pew Research Center data:

  • Larry, a “faith and flag” conservative, is a 71-year-old white retired insurance broker living in Oneonta, Alabama.
  • Britney, a 50-year-old white school secretary living in Texas, is a Populist Right voter.
  • Gabriela, a 44-year-old Hispanic nanny living in Florida, is a “stressed sideliner” who is not that interested in politics.
  • Michael, a 61-year-old Black protestant and a teacher in Milwaukee, is a “Democratic mainstay.”
  • Emma, a 25-year-old graphic designer who lives in New York City with her girlfriend, is a Progressive Left voter.

Each character’s accounts are confined to a single phone to avoid contaminating Spring’s findings. The reporter maintains accounts with computer-generated profile photos on a range of platforms — Facebook, Instagram, Twitter (now X), TikTok, and YouTube — and, essentially, lurks. She likes posts, but the accounts are all private and are not “messaging people or commenting on stuff,” she said. “They’re very much passive social media users to an extent — so all I have to do is feed the algorithm: watch content, like content, follow content. But they’re not deceiving people in any way.”

Crafting the accounts

Spring emphasized that she has carefully considered the ethical implications of creating five social media accounts that appear real but are not. She consulted the BBC’s editorial guidelines extensively to weigh the question of “how do we ensure that the level of deception is justified by the public interest?” That’s why the accounts stick to passive engagement, she noted; she never contacts people from the account or disguises her identity as a reporter. (When Spring was first interviewed about the project by the Associated Press, Donald Trump Jr. seized on the interview and tweeted about it as an example of media deception by the BBC. Spring said she would still like to sit down with him and explain the project.)

In maintaining the accounts, Spring tries to tailor her social media activity to the typical habits of her characters — Emma, for instance, is more likely to scroll TikTok than Larry, an older and more conservative voter who spends more time on Facebook. Spring monitors these accounts approximately every other day, and tries to devote the same amount of time and activity to each voter’s feed — about 20 minutes to half an hour each, she said. She screen-records everything and takes notes, and especially watches for posts that appear on multiple voters’ feeds.

This isn’t the first time Spring has run “undercover” social media accounts; she had experimented with this kind of reporting tool for other investigations, including for BBC Panorama, to see “what a troll could be recommended, or what a teenager could be recommended.” The seed for the Undercover Voters investigation was planted when the BBC was preparing to relaunch its Americast podcast, and Spring’s editor asked her what might be a good way to investigate social media around the next election. Spring reached out to Pew for insight on how she might create realistic voter characters, and began researching relevant data sets, with the goal of developing characters with “a range of different perspectives, but also that…didn’t feel too stereotypical or too much like a bit of a joke.”

“Once I’d done that, I had to find all these old phones,” she added, “which actually isn’t as easy as you’d think.” Spring set up each on a VPN with its own SIM card. (She recently “gave a couple of my undercover voters a phone upgrade” as the phones were getting too slow.)

Spring’s analysis is far more limited than the sweeping academic studies that delve into social media algorithms and their effects on polarization and the proliferation of misinformation. Though Spring realizes “these voters can’t offer me an exhaustive view of what every single person is being recommended…they do give you that taste of what different people could be seeing,” and give her a much better sense than she would have otherwise — especially as a reporter across the pond.

The Undercover Voters project won the Royal Television Society’s Innovation of the Year this March, which Spring said was “very exciting, and also slightly hilarious” given the old-school, multi-phone setup.

Inside the feeds

Spring has been monitoring the social media algorithms in action across these accounts for about a year. “As you can imagine, their feeds have been, really, quite contrasted,” she said.

She hasn’t been surprised to see her right-leaning voters recommended content that supports Trump in the face of the indictments, and condemns them as a witch hunt, while left-leaning voters are shown content condemning Trump. Content about Joe Biden’s age has generally cut through the feeds of voters on both the left and right — but while on the right, content tends to portray him as “old and incompetent,” on the left she sees some praise of policies and (of course) the Dark Brandon meme (which has been co-opted by Biden’s team).

Spring is also paying attention to how newer entrants, like Republican presidential candidates Ron DeSantis and Vivek Ramaswamy, “are courting the social media world…and where it works and where it doesn’t work.” She’s observed the DeSantis campaign’s “attempts to create content that feels like it’s organic content,” but has seen the same story playing out in the polls in action across her cell phones: “On their feeds, I tend to find of all of the people who could be president next year, it’s still Donald Trump who seems to have the most committed base of grassroot activists and the most realistic ‘fan-driven content’… that really cuts through and dominates feeds.”

Beyond trends relating to specific candidates, Spring has seen that coverage of issues like immigration is more likely to appear in right-leaning voters’ feeds, while left-leaning voters see more anti-gun content and discussion of education — and Gabriela, the apolitical voter, tends to see content focused on “the cost of living, and money, and the economy.”

Across platforms, content has been “quite similar” generally, Spring said. But she has noted the biggest change over time on X — she sees a lot more hateful and polarized content now than she did a year ago (an observation that recent research backs up), especially on the more apolitical account she maintains for Gabriela. The change in blue-check policy under Elon Musk appears to have played a role in this shift: “There are accounts that have blue ticks that didn’t have blue ticks before — their content’s being actively promoted on someone like apolitical Gabriela’s feed,” Spring said. “Her feed was starting to become a bit more political at the end of last year, but it’s really become a hell of a lot more about politics than it was before.” While Spring has seen an increase in blue-check accounts across the political spectrum, she’s especially noticed the increase in the prominence of right-leaning and pro-Republican accounts.

The undercover voter setup allows Spring to see how disinformation, and notably, conspiracies about the 2020 election being stolen, “can be actively recommended to the voters — [not] things that they’ve gone and sought out, but accounts that are readily recommended to them.”

For instance, accounts promoting election denial content and “more extreme conspiracies and hate” have been recommended to populist right Britney on Instagram, Spring said. She noted that it’s not necessarily misleading and false posts themselves that are being promoted — it’s the accounts that share some of these posts. These accounts often “aren’t very big, but are actually doing a pretty good job of continuing to pump out these narratives long after the 2020 [election].”

On TikTok, Spring is interested in the unpredictability of what goes viral. Since the Republican debate last week, she’s been paying close attention to which snippets pick up steam. Ramaswamy managed to “cut through” with comments he made that “inspired a lot of reaction,” and Trump’s separate interview cropped up (while his mug shot release the following day took over TikTok along with the rest of the internet). More intriguing to Spring was that “Nikki Haley has been popping up on the TikTok accounts of a couple of the undercover voters” — including clips of her quoting the Margaret Thatcher line, “if you want something said, ask a man; if you want something done, ask a woman.” To Spring, these have appeared to be organic clips shared by inspired supporters — not content propagated by Haley’s campaign (Spring noted that TikTok has fairly strict restrictions regulating political campaigning.). But Spring acknowledged that it’s difficult to discern how much campaigns are involved in encouraging, capitalizing on, and feeding this “organic” content, and plans to investigate this more in the leadup to 2024.

Meeting real voters

Last week, for the first time, Spring took her observation beyond the screen by meeting with a focus group of five voters whose political views approximately matched those of her “undercover voters.”

Spring hoped this focus group could give her a better sense of the impact the content voters see online has on their views. Along with the Americast team, she scoured Facebook groups (including local groups and different political groups), Telegram channels, and other online channels to find five people with a range of political views willing to meet her in Milwaukee the same day as the first Republican debate. She found Ken, a professor, radio talk show host and retired Milwaukee police lieutenant; James, a shift manager at Walgreens, a father, and a self-described “average Black guy”; Andrew, a government employee from California on vacation in Milwaukee; Mary, a retiree living in Florida originally from Wisconsin; and Amanda, a Milwaukee native who recently quit working at Walgreens to pursue film. (These voters don’t perfectly match the backgrounds of the undercover voters, since Spring’s personas are modeled on voters all around the country and the in-person conversation took place in Milwaukee.)

Despite being baffled by her many phones, the focus group was very engaged in the conversation, and was far more “positive and constructive” than what Spring typically sees online. (She reflected that her standing as a Brit, and being perceived as an outsider to U.S. politics, is “a huge asset” for gaining trust with American voters across the political spectrum.) The voters described how they struggle to distinguish between what is and isn’t true, and how social media was a natural source to turn to for people who have lost trust in traditional media.

Spring’s standout takeaways from the focus group: All five voters are grappling with severe social media and political fatigue. “People are exhausted; they’re tired of the polarization on their feeds,” she said. These voters were also eager for new leadership and a change from the unrelenting drumbeat of Trump and Biden content. Spring suggested this challenges the truism that all publicity is good publicity, and that the omnipresence of these figures online has worn voters out.

The Undercover Voters project has given Spring a deeper sense of the power social media companies have over perceptions of reality. “The reason I have to have undercover voters,” she said, “is because social media sites won’t — and to some extent, can’t — tell you exactly what they’re recommending every single voter.”

Investigating misinformation beyond social media

In addition to her Undercover Voters work focused on social media, Spring recently wrapped up a podcast tracing the following, funding, and distribution of a newspaper called The Light in a small English town, Totnes. The newspaper, despite its “slick” and professional-looking production, is packed with conspiracy theories about climate change, COVID-19 vaccines, and other issues. In the podcast, which is one of the BBC’s most popular storytelling podcasts, especially with an under-35 audience, Spring traced the newspaper’s reach across the U.K. and beyond, to Germany and Ireland. Since reporting that story, Spring has received some tips about comparable American misinformation schemes, she said (and welcomes more, if you have any), and may investigate one of these next.

Tracing the spread of misinformation via a print newspaper alongside her work looking at social media, Spring said, has shown her you don’t necessarily need the most complex tech to distribute conspiracy theories effectively. She is often asked about AI, and while it’s something she’s monitoring, she feels that “the unsophisticated disinformation tactics, whether it’s a meme or a newspaper, are doing just as effective job at sharing disinformation and hate, normalizing it, and actually resulting, to some extent, in people becoming effectively radicalized into some of these movements.”

The podcast is called “Marianna in Conspiracyland”; its cover features Spring peering down a rabbit hole. In it, Spring is the listener’s personal guide to the conspiracies infiltrating the U.K. But as the face, and the voice, of a mainstream news behemoth interrogating misinformation, Spring is also a prime target for abuse and hate from those who buy into conspiracies; the Sunday Times Magazine recently reported that she was the target of an astounding 80% of all online abuse directed at the BBC for the first six months of this year. “The more of a hit a podcast I do is,” Spring told me matter-of-factly, “the more extreme the hate that I will be subject to.”

Despite the staggering scale of the vitriol she faces, a compassionate desire to understand how people are taken in by conspiracies and misinformation drives Spring’s reporting. “I hate the attitude of ‘conspiracy theories are silly and bonkers and stupid and inconsequential,’” she said. “I also hate the ivory tower attitude of ‘I’m just going to sit up here and tell you what’s right and what’s wrong.’” She tries to approach people taken in by misinformation with empathy, while not being afraid to ask uncomfortable questions and challenge falsehoods — and believes this approach has helped both the podcast and her Undercover Voters investigation connect with audiences.

“I think increasingly, people don’t want broadcasters that say, ‘this is correct, and this is incorrect,’ and sort of broadcast down to them,” Spring said. “They want to feel like they’re a part of it, and they can interrogate this work with you.”

Photo of a stack of phones by Eirik Solheim on Unsplash.

Sophie Culpepper is a staff writer at Nieman Lab. You can reach her via email (sophie@niemanlab.org) or Twitter DM (@s_peppered).
POSTED     Aug. 30, 2023, 2:06 p.m.
SEE MORE ON Audience & Social
Show tags
 
Join the 60,000 who get the freshest future-of-journalism news in our daily email.
From shrimp Jesus to fake self-portraits, AI-generated images have become the latest form of social media spam
Within days of visiting the pages — and without commenting on, liking, or following any of the material — Facebook’s algorithm recommended reams of other AI-generated content.
What journalists and independent creators can learn from each other
“The question is not about the topics but how you approach the topics.”
Deepfake detection improves when using algorithms that are more aware of demographic diversity
“Our research addresses deepfake detection algorithms’ fairness, rather than just attempting to balance the data. It offers a new approach to algorithm design that considers demographic fairness as a core aspect.”