A Wired article by a former Facebook advertising manager over the weekend has touched off an intense debate in media, technology, and money-in-politics circles. The central question, among many, is this: Did Facebook effectively subsidize Donald Trump’s campaign by charging him less to advertise on the social network than it charged Hillary Clinton’s team?
If the answer to that sounds like it should be a simple “yes” or “no,” well, get ready to be confused. The back-and-forth between Wired contributor Antonio García Martínez, Facebook Vice President Andrew Bosworth, and people who worked on both the Trump and Clinton campaigns has shed more heat than light thus far. It’s become wonky enough, and hard enough to follow, that a pair of BuzzFeed reporters decided to give it the full reaction-GIF treatment, covering it as they would a juicy online spat between celebrities. (For those who want to catch up on the basic contours of the debate, the BuzzFeed post is actually not a bad place to start.)
At issue is the very nature of Facebook’s advertising system and whether it skews the playing field for political ads in a way that gives Trump-like candidates an (arguably unfair) advantage. For all of Facebook’s protestations, there remain good reasons to believe it does. But as to what extent, and exactly how that played out in the 2016 U.S. presidential election, it’s now clear that Facebook hasn’t yet released enough data for anyone to confidently say.
The case laid out in García Martínez’s original Wired piece was an explosive one. He explained that while ads on Facebook are sold via an automated auction, the advertiser who bid the highest dollar amount doesn’t always win. Instead, he wrote, the winner is chosen by an algorithm that combines the dollar amount of the bid with a prediction of how likely users are to interact with the ad in some way, such as clicking a link or hitting the like button. (Facebook calls this the ad’s “relevance score.”) That could have had a significant impact on the effective ad rates the two campaigns faced, García Martínez wrote. From his op-ed:
During the run-up to the election, the Trump and Clinton campaigns bid ruthlessly for the same online real estate in front of the same swing-state voters. But because Trump used provocative content to stoke social media buzz, and he was better able to drive likes, comments, and shares than Clinton, his bids received a boost from Facebook’s click model, effectively winning him more media for less money. In essence, Clinton was paying Manhattan prices for the square footage on your smartphone’s screen, while Trump was paying Detroit prices.
That insight, coming from a man who helped to create Facebook’s ad-targeting tools, provoked outrage from commentators who saw Facebook as essentially rigging the system in Trump’s favor, albeit presumably unintentionally. The flames were fanned when Brad Parscale, who directed Trump’s digital media efforts in 2016 (and was just named his campaign manager for 2020), not only endorsed the Wired story but added an incredible detail. He suggested that the Clinton campaign was in some cases paying 100–200 times as much as the Trump campaign to reach the same number of people on Facebook.
If that weren’t enough, Clinton’s former communications director, Jennifer Palmieri, appeared to tweet her agreement with Parscale’s estimate. It seemed, briefly, as though that might close the case. Twitter swelled with calls for Facebook to be regulated or boycotted, or—as the journalist Tom Scocca suggested, perhaps half jokingly—for CEO Mark Zuckerberg to be sent to prison.
Then came a curveball from Facebook, in the form of a series of tweets by Andrew Bosworth. Here’s the money chart (figuratively and literally):
CPM is ad-industry jargon for “cost per mille”—that is, the cost an advertiser pays for 1,000 people to view a given ad. Higher CPM means you’re paying more to reach the same number of people. So whereas García Martínez implied that Trump paid less for ads, and Parscale claimed Trump paid 100 times less, Facebook’s data seemed to indicate that he actually paid a little more than Clinton, on a per-view basis.
Bosworth went on to note that the two campaigns pursued different strategies in their Facebook advertising, and that “[p]rices depend on factors like size of audience and campaign objective.” He added that Facebook wants to share more campaign data with researchers and will reach out to the campaigns for permission to share information about their ad campaigns, which Facebook would otherwise keep confidential. In the meantime, he said the company was putting out that price chart “to clear up any confusion.”
Clearing up confusion, however, does not seem to have been its chief effect. People struggled to reconcile Bosworth’s chart with everything they had thought they just learned from García Martínez about how Facebook worked. Could it be that everyone was wrong, and Trump wasn’t the perfect Facebook candidate after all?
Not necessarily. Let’s consider two seemingly conflicting claims:
1. Trump got more bang for his buck on Facebook ads than Clinton.
2. Trump paid higher ad rates, on average, than Clinton, to reach the same number of people.
Those couldn’t both be true, could they? In fact, insiders say, they very well could.
The key to understanding this is that the Trump and Clinton campaigns, by all accounts, used Facebook in fundamentally different ways—as the Washington Post’s Philip Bump explained and Bosworth himself acknowledged. From Bump’s illuminating article:
Clinton ran a lot of video ads, aiming to boost turnout or make the case for her candidacy. Trump’s ads were often single images with an enticement to click. The ads were designed to do different things.
In other words, many of Clinton’s ads were simply meant to be viewed, whereas many of Trump’s ads were meant to spur a specific action from users—whether that was signing up for a mailing list, buying a MAGA hat, or even making a donation. In ad parlance, Clinton was running more brand advertising whereas Trump was running more direct-response advertising.
That’s crucial, if true, because the two strategies lend themselves to different pricing schemes and different ways of measuring success. If all you care about is putting your ad in front of swing-state voters’ eyeballs, you might as well buy ads on a CPM basis. But if you’re trying to sell merch, there might be more relevant metrics, such as CPC (cost per click) or CPA (cost per action)—and many advertisers, including Facebook, allow you to pay by the click or action rather than the impression.
Direct-response ads tend to be more expensive on a per-view basis, as Bosworth acknowledged in a statement. But they can also be much more potent, depending on your goals. As a political candidate, if you get 1,000 people to click on your campaign’s donation page with the intent to give money, that could be worth far more than getting 1,000 people to spend a few seconds viewing your autoplay video in their Facebook feed. If Trump’s campaign was focused on the former, that could help to explain how Trump ended up paying higher ad rates on a CPM basis, as Bosworth’s chart showed—even if he was paying far lower rates on a CPC basis.
As one Republican operative told Wired’s Issie Lapowsky: “A CPM price isn’t the metric we’re measuring success against.” Indeed, Lapowsky reports, the Trump campaign experimented with a staggering variety of different ad types, including at least one ad that reportedly just asked voters to “text Trump to 88022.” The cost of that ad, in CPM, according to Lapowsky’s source: a paltry 20 cents.
Some observers have also wondered whether a difference in organic reach between Trump’s ads and Clinton’s could have caused disparities that weren’t captured in the CPM chart. In other words, if Facebook users were more likely to share Trump ads with their friends on Facebook, then more people would have seen them in their feeds. Two people with knowledge of how Facebook’s ad system works told me they doubted that effect was major, simply because people don’t tend to intentionally share ads on Facebook very often. Still, it seems possible that some of Trump’s campaign ads were an exception, given the often inflammatory tone of his messaging and the level of enthusiasm among some of his core supporters. Many of his ads were essentially memes.
In short, nothing Bosworth has said disproves the speculation by García Martínez and others that Facebook’s advertising system played to Trump’s advantage in various ways. We know that ads that generate interaction from users travel farther on Facebook for the same amount of money, so if Trump’s ads did that, then it stands to reason that his campaign benefitted accordingly.
That said, Facebook’s response to the controversy served as an important reminder that without more data, any claims about which campaign paid how much for what result should be treated with caution. Wired posted an update to García Martínez’s story on Wednesday acknowledging Bosworth’s tweets and explaining why the data he shared “do not confirm or deny the hypotheses contained in this piece.”
More data from Facebook and the campaigns would certainly be welcome, at this point. For what it’s worth, Facebook notes that its forthcoming ad-transparency tools should help to clarify some of these sorts of questions in the future. Until then, it’s worth remembering that campaign ads on Facebook represent just one way in which the platform shaped the election. Claims of biases in Facebook’s ad rates draw attention, because political advertising has traditionally been regulated. But ads make up just a fraction of people’s news feeds. When it came to amplifying pro-Trump and anti-Clinton messages, the biases in Facebook’s news feed algorithm—which govern who sees what posts from their friends and the pages they follow—likely had a far greater effect.