The Facebook Defectors Turning Trump’s Strategy Against Him

The platform was key to the president’s upset victory in 2016. Can a group of former employees use it to boost Joe Biden?

Three months before Election Day, James Barnes teleconferenced into a strategy meeting about how, exactly, to persuade people not to vote for Donald Trump. The 32-year-old wore his hair loosely gathered in a man bun and had the sober expression of someone who, as a hazard of his occupation, thinks about the president nearly all the time. Other faces popped up on his monitor, offering glimpses into millennial apartments, until someone started a screen share. They were here to review a series of video testimonials by conservatives who had decided to oppose Trump. Barnes and his colleagues wanted to know which ones resonated most with prospective voters on Facebook. Were testimonials from men or women more effective? Midwesterners or Southerners? How many viewers made it past the first 15 seconds?

Barnes, who works at the political nonprofit Acronym, attends these meetings every week. His team has two goals—to nudge voters away from Trump and to close what he politely calls the “enthusiasm gap” for Joe Biden. Using a custom-built tool dubbed Barometer, they micro-­target “movable” voters on Facebook, run randomized tests to see what kind of ads work best, and then adjust them to taste.

Barnes, who spent the early part of his career at Facebook, leads his colleagues in two-week-long “sprints,” following Mark Zuckerberg’s adage about moving fast. In the past year, they’ve completed hundreds of tests, refining their own strategy and sharing insights with other Democratic groups. “We’ve got a larger corpus of data than anyone else about what’s moving ­people,” he says. (One takeaway: Voters love the Midwestern men.)

In 2016, Barnes was the Facebook staffer assigned to get Trump’s digital team comfortable on the platform. Raised in Tennessee, he had been a conservative all his life. As a student at George Washington University, he had chaired the DC Federation of College Republicans, then spent several years as a GOP political consultant. But “working with Trump specifically was not something that I wanted to do,” he says. “There is nothing to like about that man.”

Barnes loved Facebook, though, and he believed in the vision of building a new space for political engagement. He also saw the campaign as a chance to move up in the company’s intensely competitive ranks. And so he pushed through what he describes as “an enormous amount of internal conflict,” reassuring himself that the work was interesting and that he was doing a good job. He showed the campaign, led by an operative named Brad Parscale, how to measure the impact of its ads and fine-tune its messaging strategy, how to expand its reach with the Lookalike Audiences tool, how to use engagement data to hook first-time donors. The result, Facebook executive Andrew Bosworth would later say, was “the single best digital ad campaign I’ve ever seen.”

On the night of the election, before the returns came in, Barnes recalls “thinking that part of my life would be over, because the mission would be accomplished”—his mission as an engineer, that is, not the mission of claiming the White House. Nobody on the team, least of all him, believed Trump would win. Barnes, in fact, had voted for Clinton. The outcome left him rattled. “Now, here it is four years later, and I’m still at it,” he says. “A different chapter of the same book.”

Barnes didn’t leave Facebook until 2019, by which point he’d registered as a Democrat, moved from DC to San Francisco, and cycled through several teams at the company. Then, during his “recharge”—a 30-day vacation perk that Facebook employees receive every five years—he traveled to Peru, drank ayahuasca with a shaman, and found himself on the road to Damascus. When he returned to the States, he quit Facebook and started intermittent fasting. He set to work on a project that would repurpose the strategies he’d learned in 2016 to oppose Trump in 2020.

Last fall, Barnes met Tara Mc­Gowan, Acronym’s founder and CEO, who loved his idea so much that she hired him. He began recruiting former colleagues to help. Since then, according to Damon McCoy, a researcher at New York University’s Online Political Ads Transparency Project, Acronym has developed “the most sophisticated digital advertising campaign on the Democratic side.”

Still, mainstream liberals have been slow to welcome the organization into the fold. Over the past year, Acronym has developed a problematic reputation. It is perhaps best known for its association with Shadow, the smartphone app that spectacularly failed to tally caucus results in Iowa, triggering rounds of recriminations and conspiracy theories. (Pete Buttigieg, who initially won the most delegates, had previously engaged Shadow’s services, and McGowan’s husband worked for his campaign.) Acronym is also the majority owner of a digital news ecosystem called Courier. Modeled after the right-wing blogosphere, it promotes partisan ideology disguised as local news. NewsGuard, which rates news websites for their adherence to journalistic standards, gave Courier a 57 out of 100, placing it just above RedState and Blaze Media.

Acronym’s cozy attitude toward Facebook is another major point of contention. Since the 2016 election, the platform has become something of a bogeyman in Democratic circles, owing to the role it played in Trump’s upset victory. His campaign used Facebook’s marketing tools not only to galvanize his supporters but also, as an unnamed senior campaign official told Bloomberg barely two weeks before Election Day, to engage in “major voter suppression operations” against Clinton. (Parscale has denied that the campaign’s “memes and things” constituted suppression.) As a result, according to Neera Tanden, the president of the Center for American Progress, many progressives now see Facebook as “antithetical to a well-functioning democracy.”

But McGowan rolls her eyes at the notion that Trump conjured some kind of digital black magic in 2016. She takes no issue with copying his playbook. In fact, she admires the way that conservative power brokers, including the Koch brothers, have used data to their advantage. She just wants to do it on the left. Some Democratic megadonors seem to buy her argument: Acronym and its affiliated political action committee have received millions in funding from LinkedIn cofounder Reid Hoffman, film director Steven Spielberg, and venture capitalist Michael Moritz.

Nearly three-quarters of adults in the US use Facebook, most of them every day. “This is where the game is played,” Barnes says. “As long as the field is there, that’s the field we’re going to play on.”

Image may contain Plant Human and Person

In 2016, Barnes was a Facebook employee embedded with the Trump campaign. Now he’s using his skills to promote Joe Biden.

Photograph: GABRIELA HASBUN

There are two types of voters: Those who know whom they’re voting for and those who don’t. Most fall into the first category, immutable long before Election Day, so candidates must fight not only to get their people to the polls but also to sway the very small pool of undecided, ambivalent, or otherwise out-of-touch voters that remains.

Since at least the George W. Bush era, they’ve done this with targeted messaging. Back then, the work took place offline. A campaign might identify conservative voters with religious leanings by, say, mining the public records of hunting licenses, purchasing membership lists from mega-churches, and looking at home ownership in specific zip codes. The winnowed-down group might then receive a glossy campaign pamphlet about the candidate’s views on abortion. Barack Obama, in his 2008 presidential bid, used everything from demographic data to television preferences to target voters. (One discovery: Those who watched TV Land, the basic cable channel best known for airing reruns, were less likely to have a presidential preference.)

Modern tech platforms have only sharpened that precision. Rather than guessing where the religious voters live, today’s political advertisers can use geofencing technology to, for instance, locate Catholics who have been to mass at least three times in the past 90 days. On Google they can match their message to a specific search query, such as “impeachment.” On Facebook, the level of control is even more granular. “It’s very easy for these political advertisers to partition and very narrowly message and tell different people different things,” says NYU’s McCoy. “It definitely has an element of manipulation.”

In 2016, with Barnes’ help, the Trump campaign made particularly good use of a Facebook tool called Brand Lift. It was originally designed to help advertisers run randomized, controlled tests on their audiences: What proportion of people could recall seeing an ad for that $2,000 Gucci tote? How did they feel about Gucci generally? How likely were they to recommend it to their friends? In Trump’s case, his staff would develop specialized ads for different slices of the electorate, push them out in huge numbers—reportedly as many as 60,000 a day—and gauge how people responded. Then, crucially, they would change the message as often as needed to keep voters’ attention. Using Facebook, Parscale said in 2017, “I can find, you know, 15 people in the Florida Panhandle that I would never buy a TV commercial for.” He added, “we took opportunities that I think the other side didn’t.”

Barnes coached Parscale through some of these strategies during meetings in San Antonio, Texas, where the operation was based. At the time, none of it seemed especially repugnant. Facebook wasn’t the problem; the candidate was. But in the years afterward, as Barnes watched the company reckon with misinformation, foreign interference, and other abuses, his belief began to waver. In 2018 it emerged that Facebook’s policies had allowed a Trump-­affiliated consulting firm to harvest millions of users’ personal data without their consent. Met with blowback from furious consumers and threats of government regulation from trustbusters in Congress, the platform changed its rules for political advertisers. It revoked access to some of the tools that had been crucial to Trump’s efforts, including Brand Lift. At Acronym, Barnes essentially rebuilt them.

The work of the Barometer team is to constantly be in motion, to push “the messaging strategy for as long as it works,” says McGowan. “And it won’t work for very long.” They begin by creating an audience of ­people whom they consider movable, based on official voter files and political preference data provided by Facebook. For most campaigns and PACs, the ideal target is someone who may not have strong partisan leanings but nevertheless casts a ballot year after year—who is likely, in other words, to provide a good return on investment.

McGowan believes that’s a flawed strategy, one that cost Democrats in 2016. In the Trump-Clinton race, there was an unexpected surge in turnout among low-education voters with ­little political knowledge. McGowan predicts an even bigger one this year. To isolate that key demographic, Acronym blankets Facebook with surveys consisting of relatively simple multiple­-choice questions, like “Who controls the House of Representatives?” Those who score lowest make up the target audience.

Much of the thinking behind Acronym’s strategy comes from Solomon Messing, its chief scientist, whom Barnes poached from Facebook last fall. Academic literature supports Messing’s views: Research from the Kaiser Family Foundation found that while swing voters come from many different demographics, a consistent similarity is their tendency not to pay attention to the news. Messing says he wasn’t sure this would still hold true in 2020, “because everyone has an opinion about Trump,” but, remarkably, it has.

Once the Barometer team has identified its low-information audience, it splits people into experimental groups and shows them different sets of messages. In July, Barnes and his colleagues tested ads criticizing Trump’s response to the pandemic. One highlighted his “refusal to promote clear public-health guidelines.” Another said that his administration “spent billions bailing out big corporations while Americans struggled.” Most political groups survey the audience again immediately after they’ve seen an ad; Barnes and his colleagues wait a week, to assess how well the message stuck. For the set of pandemic ads, middle-aged voters were most responsive to economic arguments, while older voters responded best to ads that focused on Trump’s threat to their health.

Some of the findings have been substantial. It turns out that posting a link to a story in The New York Times or other established media outlets has a much bigger effect than running straight ad copy. “For folks who don’t watch the news, who don’t see how badly Trump is handling things, we just show them the facts and they respond pretty well,” Messing says. McCoy calls this a “trust shortcut.” People are naturally wary of political ads, he says, but they’re more receptive to a message when it comes from a news organization that doesn’t seem to have an agenda—even if it’s being promoted by an advertiser that does. (Facebook includes a label marking the post as a political ad and gives users the option to see why they were targeted.) Another genre that works quite well is critical testimony from conservative commentators. In January, Acronym promoted a clip of Fox News host Tucker Carlson blasting the president for the killing of Iranian major general Qasem Soleimani. It dented Trump’s approval rating by 4 percent among the low-information group.

At other times, though, Acronym’s experiments have had exactly the wrong effect. After the group pushed out a series of ads attacking Trump on impeachment, Barnes and his team gathered for their standing meeting to review the data. It didn’t look right. “We were like, is everything broken, or did we just move a bunch of people in the wrong direction?” he recalls. They found that all of the ads were received badly, and some of them actually made voters more sympathetic to Trump by reinforcing the idea of a Democratic witch hunt. “The more partisan attacks and content there are, the more backlash we see,” says McGowan.

Later in the spring, Barnes proposed a way of getting a jump on that kind of backlash. Before he left Facebook, he had helped set up the elections integrity team, an internal effort to stop abuse and election interference on the platform. As part of its work, that group used various forms of user engagement (likes, comments, emoji reactions) as an indicator of outbreaks of misinformation.

Acronym had been treating the engagement data as an irrelevant byproduct, what Barnes calls “digital exhaust.” But when he and his colleagues parsed it, they found “really strong correlations.” Ads that flopped, for example, had a much higher number of “haha” emoji reactions; it seemed as though people were laughing at Acronym. When an ad had more comments than shares—a classic case of getting ratioed—that also foretold a backlash. Barnes’ team incorporated these findings into a predictive machine-learning model. He calls it Dorothy, for Dorothy Thompson, a wartime journalist who got kicked out of Germany for calling attention to the rise of Nazism, and for the instrument used in the film Twister to spot a tornado before the funnel forms.

When I spoke with Barnes in August, he had just watched Joe Biden take the faux stage at the Democratic National Convention, cheered on by people Zooming in from all over the country. It stirred something inside of him—hope for a more presidential president, and maybe a morsel of relief.

Earlier that month, in a series of tweets about Stoicism and fasting, Barnes had quoted a few lines from Seneca, the Roman philosopher who served as an adviser to Nero, then was accused of conspiring to have the ruthless emperor assassinated. In his “Moral Epistles,” Seneca had extolled the benefits of a monastic existence, of wearing coarse clothing and eating “hard and grimy” bread. Do this for long enough, he wrote, “and you will understand that a man’s peace of mind does not depend upon Fortune.”

Barnes thought of “Seneca’s wisdom” every day. “I’ve done a lot of work coming to terms with the last four years of my life, so I’m not investing too much of my ego in a victory,” he says. But if Donald Trump wins again? “It is an absolute nightmare scenario for the country.”

As Election Day draws closer and the pool of movable voters dries up, the job of persuading them gets even harder. Fewer people pay attention to Acronym’s ads now than they did in the spring; almost no one watches the video testimonials for more than 15 seconds. Back in March, when the “persuasion window” was wider, the Barometer team found that people who had seen its ads had a 3.6 percent lower approval rating of Trump, compared to a control group. By August, barely 1 percent of people would budge. Still, if these numbers seem trivial, consider a recent academic study of the persuasiveness of political ads on TV, which found “an average effect of zero.”

Of course, none of that matters unless people vote. “The problem is, with all the money going into persuasion—and we’ve done the best work in the entire field—it’s really fucking hard to connect the work you do to the actual votes converted at the end of the day,” McGowan says. Barometer’s second-round surveys now ask respondents to name the date of the general election, on the supposition that anyone who can likely plans to cast a ballot. But this is at best a proxy. Even for a team devoted to measurement, genuine enthusiasm is a hard thing to measure.

At Facebook, Barnes once led a group whose mission was to show that ads could make people do things, like visit a Macy’s store after seeing the retailer’s ads online. Although Barnes and his colleagues had pipelines of receipts from the Macy’s data warehouse to track exactly who made a purchase, they couldn’t prove much. Measuring the tangible impact of the political ads is even more complicated. “My basic perspective,” he says, “is that we don’t know anything.” And as in 2016, they won’t know what they don’t know until the ballots are tallied.


ARIELLE PARDES (@pardesoteric) is a senior writer at WIRED, where she works on stories about our relationship to technology. She profiled the actor Chris Evans in issue 28.02.

This article appears in the October issue. Subscribe now.

Let us know what you think about this article. Submit a letter to the editor at [email protected]


Image may contain Logo Symbol and Trademark
Special Series: A More Perfect Election

Fraud-proof. Hacker-proof. Doubt-proof. Across the country, people are working hard to reboot the American voting system.

  • A Texas County Clerk’s Bold Crusade to Transform How We Vote
  • The International Playbook for Foiling Russian Interference
  • The Facebook Defectors Turning Trump’s Strategy Against Him
  • Stacey Abrams on How We’ll Beat Back Voter Suppression
  • How We’ll Know the Election Wasn’t Rigged
  • One Data Scientist’s Quest to Quash Misinformation

Latest posts