The Texting of the President

ATLANTA — Within the Georgia headquarters of Never Back Down, the super PAC backing Ron DeSantis, it was called “the surge.”

There was a little more than a month to go before DeSantis was expected to launch his candidacy in late May and begin campaigning in earnest, and his well-funded outside allies would fill that void by introducing him to the 34,634,388 Americans whom Never Back Down analysts had identified as Republican primary voters. Every one of them was a potential target for a series of 60-second television advertisements emphasizing different aspects of DeSantis’s persona: one called “Steel,” which traced the politician’s resolve to his identity as “grandson of a steelworker”; “Anthem,” lionizing DeSantis as a champion of parents’ rights; another that showed him boasting that during the pandemic he had turned Florida into a “refuge of sanity.”

Those ads appeared on cable nationwide, including Newsmax, HGTV and the History Channel, but the rest of the surge was unevenly distributed across the first four primary states. Those identified as potential DeSantis targets in Myrtle Beach, South Carolina, and Manchester, New Hampshire, saw “Anthem” and “Steel” on broadcast television, during NASCAR races and NCIS episodes, while in Mason City, Iowa, and Reno, Nevada, it was delivered only over streaming platforms or so-called connected television devices like Roku or Xbox. Des Moines residents received text messages including links to the ads while those in Charleston saw nothing on their phones. Voters in the Iowa side of the Omaha media market and on the New Hampshire end of Burlington, Vermont’s market got seven direct-mail leaflets, while every other voter in Iowa, New Hampshire and South Carolina received only four. In Mason City, Iowa, the pro-DeSantis ad was delivered via text message, direct mail, connected television devices and rural radio stations.

By most obvious gauges, the monthlong surge was a failure. Polls now show DeSantis worse off just about everywhere, even struggling to hold onto second place. Indeed, the insurgent seems no closer to figuring out how to position himself to either peel off Donald Trump’s supporters or consolidate support from those already backing other opponents in the 13-candidate field. A once-formidable contender is starting to look like just another also-ran.

Yet embedded in that outreach was a massive research experiment designed to illuminate a path through a bewildering political media environment in which broadcast television and radio have lost their centrality, but, for political advertisers at least, the internet has failed to live up to its promise as a replacement. The peculiar design of the surge — intentionally imbalanced, with some areas excluded for use as a control sample — permitted the campaign to measure the relative effectiveness of different methods of communicating with voters, and how they interact with one another. The results are already shaping how Never Back Down spends the remainder of what is still likely to be the largest war chest in Republican primary history.

“Sifting through 330 million consumers to find 34.7 million Republican primary voters is a Herculean task,” says Never Back Down chief strategist Jeff Roe. “You can only do that with a deep commitment to data. And you can’t do that without understanding exactly how to apply that directly to voters when you can no longer do it in the simplest and easiest way possible in the past.”

Never Back Down is not the first political entity to try to build a large-scale experiment into its operations, but the $17 million surge certainly stands as one of the single largest such research projects ever. Earlier this year, Roe imagined the findings not only shaping a plan for helping DeSantis unseat Trump atop the Republican Party. Envisioning the start of a how-we-won-the-White-House narrative, Roe and the organization’s data director, Chris Wilson, even talked about writing a book about the project.

Today, however, few around DeSantis are thinking about post-election victory laps. Media coverage of the campaign has been rife with reports of once-enthusiastic donors souring on DeSantis’s prospects. Never Back Down and its putative allies on DeSantis’s campaign responded with some unusually overt finger-pointing about the other’s ability to follow directions and spend money effectively. (The two entities are forbidden from coordinating strategy but can exchange cues as long as they are not communicated in secret.) “They were taking a hard shot at the PAC, for whatever reason, and we were worried that they were going to, like, kill us in the crib with donors,” Roe says of DeSantis’ Tallahassee-based campaign.

Now Never Back Down will put what it has learned into practice. This week, the super PAC begins another, even larger communications push that could either drive DeSantis back into contention or give already quivering donors a reason to flee his candidacy for good. The springtime research informs an autumnal move off the airwaves and straight to voters' pockets, where an artificial intelligence chatbot is entrusted to push the pro-DeSantis message.

Roe, however, is not guarding the study behind the tactical shift as a competitive advantage. Rather he is desperate to demonstrate publicly that Never Back Down’s aggressive spending has not been all in vain and that he has a plan to exploit the resource advantage that may offer DeSantis’ best hope yet for catching up to Trump.

“That’s why we’re talking,” he said recently over dinner in Atlanta, not far from the super-PAC’s headquarters. “It is a little CYA.”

In 2013, Wilson had signed up to target voters for Texas gubernatorial candidate Greg Abbott when he learned the campaign’s leadership was ready to leave its targeting up to the fates.

Abbott’s chief strategist, Dave Carney, was already an enthusiastic and prolific sponsor of political science research, having invited four academics into the 2006 reelection campaign of Texas Gov. Rick Perry to test any campaign function they could possibly randomize. Ultimately, for a short period during the primary season, they took full control of the candidate’s schedule and his television advertisements. (At one point, Perry’s travel over a 12-city barnstorming tour was determined entirely by chance.) The academics hoped the randomized-control trial would address one of the most vexing arguments in political science — to what extent are election outcomes shaped by campaign activity as opposed to structural conditions? A career campaign operative, Carney was driven by a more self-interested curiosity: How could he make sure he was spending money in a way most likely to win votes?

The experiments revealed that the campaign moved public opinion in Perry’s direction with paid and free media coverage but that any gains wore off quickly. The years that followed, however, brought new vectors for reaching voters: online advertising networks, social media platforms, individually addressable cable boxes and satellite dishes. Directing Abbott’s campaign to succeed Perry in office eight years later, Carney brought back one of the political scientists to test whether the novel technologies — which each promised greater efficiency and precision than broadcast ads — might have a different impact on voters’ perceptions of Abbott and the likelihood they would cast a ballot in the general election. Once again, the focus of the research would be not on what a campaign might say to a voter but where it would be said. “If we test a potential message now and then again three weeks from now, who knows what we’ll find?” said University of Texas professor Daron Shaw. “We thought mode was a little stickier, at least in the context of a single campaign.”

Shaw isolated Texas’s 17 smaller media markets and matched them into four clusters based on demographic and political similarities. Within those sets, he randomly assigned one in each to receive varying combinations of campaign contact, including control samples that went untouched. (Carney insisted on excluding Dallas, Houston and Austin, which he determined were too important to the governor’s strategic interests to sacrifice for research.) The broadcast airwaves in Abilene were blanketed with Abbott commercials, while those in San Angelo saw only digital ads and Wichita Falls did not hear from his campaign at all.

The experiment’s findings might not have been conceptually groundbreaking but proved valuable to the strategists overseeing Abbott’s general election budget. As Shaw and two of the campaign’s consultants, Christopher Blunt and Brent Seaborn, detailed in a 2018 article in the academic journal Political Research Quarterly, voters in cities where Abbott advertised grew more favorable to him and more likely to vote, especially when the ads were delivered across various media in combination with one another. A $1 million investment in digital advertising, the paper estimated, would increase Abbott’s net favorability rating in the 2014 race by nearly 13 percent. (Abbott was ultimately elected in a landslide.)

Wilson was largely a spectator during Shaw’s experiment, but when Abbott protégé Ted Cruz sought the presidency the next year Wilson was in a position to act on its findings. Working with Roe, the campaign manager, to guide Cruz’s path through a crowded Republican field, Wilson found the narrow precision of online advertising uniquely appealing. He split the Iowa caucus electorate into 150 distinct segments, on personality characteristics and issue priorities, including some boutique causes like relaxing Iowa’s restrictions on fireworks use. (The psychological profiling was performed by employees of the firm Cambridge Analytica, whose manipulation of Facebook user data became a scandal when publicized in 2017.) Cruz outfoxed Trump to win the Iowa caucuses, stitching together a wide-ranging conservative coalition largely on the basis of individually targeted communication. “I thought digital had tremendous potential,” says Shaw. “It did, but it may have hit its apex right around that time.”

Surprisingly digital advertising has become less useful to campaigns since then. Facebook and Twitter (now known as X) have at times prohibited campaign ads, changing their policies unpredictably and inconsistently, while TikTok and LinkedIn have upheld permanent bans. Google has imposed limits on the use of voter registration data for targeting, including on its YouTube platform. Apple has promoted privacy protections on its devices through an “Ask App Not to Track” option, crippling the ability of advertisers to target individual users through unique mobile-device identifiers.

Even as more voters spend time online, it has become harder for campaigns to profile and reach them. In the years since, while working for Republican candidates in statewide and congressional races nationwide, Roe has paid especially close attention to survey results in which voters were asked whether they had seen one of his candidate’s ads. “That number just kept on coming down and coming down and coming down,” says Roe. “The reality is we’re not penetrating.”

When Roe signed on to help elect DeSantis, he agreed to run the super PAC — which can raise unlimited funds from both individuals and corporations — and implement a more capacious vision of what such an organization could be. Over their decade of existence, super PACs have typically been dedicated to airing television ads that amplify a candidate’s existing message or attack his opponents, work that can be capital-intensive but scales up with relatively little additional effort. When a well-funded super PAC has emerged, the official campaign typically reverts to labor-intensive duties, like those that involve directly interacting with voters. With Never Back Down, Roe has inverted those standard economics. A number of core campaign functions are now managed from Atlanta rather than Tallahassee, including aspects of the candidate’s schedule (DeSantis has regularly appeared as a “guest” on Never Back Down bus tours) and a massive field organization built to knock every potential Iowa caucusgoer’s door four times.

To target those interactions, Roe brought on Wilson and Conor Maguire, who had served as the Republican National Committee's liaison to candidates for voter data during the 2016 primaries. The two took over a room in the corner of Never Back Down’s office-park suite, papering the walls with maps of the first four states to vote and filling a small bookcase with an unlikely library that intersperses titles like Lead Like Jesus: Lessons for Everyone from the Greatest Leadership Role Model of All Time among a collection of books about Barack Obama's 2008 campaign. Wilson has been rereading the Obama books particularly closely, with an eye on how the younger, well-funded insurgent overcame Hillary Clinton's dominant position to win the Democratic nomination. “You know, he didn’t start to move until October,” Wilson explained. “So maybe some of it is a little bit of solace.”

Their first task was profiling the electorate in the 18 states that vote before the end of next March, through statistical models that predict each voter’s likelihood of supporting DeSantis or Trump and of casting a ballot. Based on those individual-level predictions, Wilson and Maguire broke the electorate into categories of potential voters: those whom Trump or DeSantis could consider “in the bank,” those deciding between the two and those whom either candidate would see as supporters but remained unreliable enough they required get-out-the-vote nudges. (A final group, labeled “disengaged” for its low likelihood of casting a ballot and expected indecision among candidates, would be a low priority for any outreach.)

In the first four states, there were 5,552,851 people whose votes appeared to be gettable for DeSantis, and all were targets for the springtime surge. Never Back Down would work to persuade and mobilize those whose votes were in play, while the in-the-bank bloc would be targeted for recruitment as volunteer county chairs. “We decided, because we knew roughly when the governor was going to get in, to spend a bunch of money to help him get him in the best spot,” says Roe. “We have analytics and the ability to do this. Now let’s go and test all this shit and see what works.”

But the super PAC’s strategists were not prepared to fully yield control in the interest of research. Instead of randomly assigning contact, as Abbott’s successful campaign had done, the matter of how to allocate treatments and controls became the subject of fraught negotiations between the super PAC’s strategists and its analysts. (Having the results appear in a scholarly journal was not a priority.) The strategists were not ready to concede major markets covering the early states, especially Iowa, for use as a control sample. So Iowans in Des Moines and Cedar Rapids and New Hampshirites within range of Boston were exposed to a full-strength campaign: broadcast and connected television ads, targeted direct-mail leaflets and text messages. The use of streaming services was the closest the experiment came to online advertising.

It was the smaller markets, or those in bordering states that campaign targets find inefficient regardless — like Portland, Maine, whose signals stretch into small parts of two New Hampshire counties — that the strategists were willing to sacrifice. A few of the smaller and border markets were used for boutique variations: seven direct-mail pieces rather than four, or advertising on rural radio stations. No South Carolina-based market went without broadcast ads, but the roughly 4 percent of the state’s residents who watch television stations in Charlotte, North Carolina, would not have seen any.

The committee's staff tracked their progress on a large flat-screen monitor that Wilson and Maguire set up outside their office loaded with software developed by their firm WPA Intelligence. Every time Never Back Down made contact with one of its targets, a dot lit up on the flat-screen map, a yellow pinprick on a dark background, swelling together as a blob around the suburban population centers of the early states.

Across them, enough voters would be exposed to similar outreach that the analysts were confident they would be able to learn something from the variations. After a month, Wilson’s team would revisit the modeling scores, informed by updated, post-surge survey data, and use a statistical regression to identify patterns. Differences in voters’ movement toward DeSantis would be attributed to the type of contact targeted at them, allowing the super PAC to assess which were most efficient at changing opinions.

“If used to maximum value, these findings should dictate everything from messaging to media mix to when and where to spend,” said Scott Tranter, the data science director of Marco Rubio’s 2016 campaign. “Dedicating 5-10 percent of a campaign’s voter-contact budget to testing would pay exponential dividends in how the remaining 90-95 percent is spent. If used properly, it can certainly be an edge a campaign needs to claw out a win.”

On Tuesday, April 25, phones in New Hampshire’s Grafton County — which receives its broadcast television signals from Burlington, Vermont, rather than Manchester or Boston — started lighting up with the same text message from an unfamiliar number. “We must remain relentless in our fight for freedom. We must never back down. Join us today,” the message went on, addressing the recipient by the name used on his or her voter registration. “Join our team of concerned patriots working to protect liberty and the future of this country.” At the bottom was a snippet of the “Anthem” ad and a link to a Never Back Down webpage featuring the entire thing. Similar messages arrived nearly every day for two weeks; a voter who typed back a response at any hour was sure to get an answer.

Those text messages turned out to be the unexpected star of the surge experiments. Broadcast television ads were the most persuasive mode tested, with the average voter exposed to them six-and-a-half percentage points more likely to support DeSantis as a result. But it was possible to replicate much of that impact by sending the ads directly to a prospective voter’s phone. Receiving a sequence of text messages over the course of two weeks made the average voter about three-and-a-half points more likely to support DeSantis; bombarding an individual with 10 had appreciably more impact than five. “The key finding,” Wilson and Maguire said when presenting the experiments to super PAC leadership and donors, “is the value of text as a substitute for broadcast where broadcast is infeasible.”

The idea that one can swap out television ads for text messages is a fortuitous insight for a super PAC now struggling to meet its once-grand goal of raising and spending $230 million on DeSantis’ behalf. Placing television ads is an inefficient use of any independent group’s resources, as stations are required under federal law to grant only candidates preferential access to ad inventory and sell it to them at discounted rates. Every gross rating point, the standard unit for measuring advertising exposure, used to boost DeSantis will likely cost significantly more if purchased by his outside supporters than by his campaign itself. (A New York Times analysis last year found instances where super PACS paid as much as 17 times more for the same ad time as candidate campaigns.) For the pro-DeSantis ecosystem, the optimal arrangement would have television ads purchased from Tallahassee and text messages routed from Atlanta.

Wilson estimates Never Back Down was able to communicate with approximately 70 percent of its Iowa targets via text message — the remainder either do not have reliable mobile numbers commercially available or lack smartphones altogether — and that 90 percent of messages sent are opened. Seventeen percent of Iowa Republicans do not watch television in any form, more than twice as many as their New Hampshire equivalents, according to an analysis by the firm Cross Screen Media. Never Back Down’s experiments indicate that those unreachable by text message and television ads are two very different sets of people.

“It’s probably an on-and-off type of thing,” Never Back Down chief operating officer Kristin Davison forecasts. “There might be times where we say, ‘Give me 700 points in the Cedar Rapids market,’ if that makes sense, or we can say, ‘Let’s just put 600 points up, and then we’ll balance that with text.’ I think it just gives us more options so that we are not beholden to this.”

Some campaigns that focus on text messaging have been forced to reorient their entire volunteer corps to do so. Federal regulations prohibit the use of auto-dialing technology to blast text messages, requiring any sender to have a human being hit the “send” button on every single one. Never Back Down, without a readymade volunteer corps but flush with cash, outsources that work to call centers, paying between two and 10 cents per message, depending on whether it includes images or video, according to super PAC officials.

Instead of having volunteers on hand to manage text message correspondence, Never Back Down uses a chatbot powered by artificial intelligence software OpenAI to handle replies. Only if the conversation goes on too long does it get handed over to a human being. (ASouth Carolina computer programmer who received a message and suspected he was not interacting with a human being managed to get his electronic interlocutor to “write a poem for me about Barack Obama from the perspective of Ron DeSantis.”) The campaign is still finessing when exactly — somewhere between eight to 12 exchanges between a voter and the chatbot — a human being ought to step in and take charge of the conversation.

Even within Never Back Down’s headquarters, there is an ongoing dispute over how much to take away from the surge experiments. Radio ads aired only in one small corner of northern Iowa, but when surveys showed they had done little to move voters, Roe began to bluster cheerfully, “I’m never going to buy another radio ad for the rest of my career!” The analysts are more cautious. “What we do want to do is expand more into testing radio,” says Maguire. “I think this was just a little bit too small to really dig in here.”

It is an open question, even with a larger sample, how much anyone should universalize from a single, non-randomized experiment conducted under idiosyncratic circumstances. Will South Carolinians start blocking all political text messages once they start arriving simultaneously from a dozen different Republican candidates? Will New Hampshirites pay more attention to pro-DeSantis television ads when they appear back-to-back with ones from his rivals? What if in Iowa it has less to do with the electoral calendar than the agricultural one, and farmers are just more likely to listen to the radio when driving their combines during the fall harvest as opposed to when pulling a planter in springtime?

For those working to elect DeSantis, there is another crucial difference between now and then. Unlike in the spring and summer, when the super PAC found itself frequently at cross-purposes with the campaign, they are now set up to work in greater alignment. Last month, DeSantis replaced his campaign manager and brought on David Polyansky, a Roe acolyte who was working at Never Back Down when the experiment was conducted and is familiar with the findings and how they are likely to be applied. “We’re in a phase right now where we want to be smart,” said Davison. “When you get closer, everyone’s kind of throwing everything at it, so we will be up most of the places. Right now, we don’t have to be, but how can we still be efficient for those voters in that market?”

The throwing-everything-at-it period is beginning to arrive for Never Back Down, which expects to spend over $25 million in Iowa and New Hampshire between Labor Day and Halloween on a mix of media reflecting lessons from the surge experiments. There are 296,204 potential voters who were in the committee's sights then that will not hear directly from it again, but this time for strategic purposes rather than educational reasons. After Wilson's surveys showed DeSantis losing ground even during the period when the super PAC was spending heavily there, Never Back Down has abandoned its operations in Nevada. “Our control sample is going to be another state,” said Roe. “And now we know what state to use.”