How Facebook rewards polarizing political ads

As the debate intensifies around Russian ad buys in the US election, a fundamental aspect of Facebook’s platform has gone mostly overlooked. Facebook’s auction-based system rewards ads that draw engagement from users by making them cheaper, serving them to more users for less money. But the mechanics that apply to commercial ads apply to political ones as well. Facebook has created a powerful system that dynamically, and unpredictably, changes the prices of political ads. The system also encourages polarization by incentivizing ads that users are predisposed to agree with.

Unless Facebook makes its internal data public, it’s impossible to say which ads reach which audiences, or how much candidates spend to reach them. After the 2016 presidential election, a senior Facebook employee said that Trump’s cost of reaching voters was substantially lower than Clinton’s, according to communications reviewed by The Verge. Trump was able to reach a larger audience than Clinton for less money, the employee said, illustrating the power of mastering Facebook’s ad platform. At a time when the company’s advertising business is under increasing scrutiny, Facebook’s platform dynamics could represent a new avenue for regulators to investigate.

Most media outlets do not routinely set prices based on the content of the advertising, or how much their audiences are expected to enjoy it. But Facebook does: the company adjusts the number of people an ad will reach after it is posted and users began interacting with it. This approach has upended traditional approaches to political advertising, in ways that campaigns are only beginning to reckon with.

The Trump campaign’s mastery of Facebook advertising is well established. Speaking to Bloomberg Businessweek in the run-up to Election Day, the campaign’s digital agency, Giles-Parscale, said it generated more than 100,000 ad variations. Trump paid Giles-Parscale about $90 million to run ads, the majority of which ran on Facebook, where they were micro-targeted to an unknown number of disparate audiences. Clinton’s digital marketing efforts were led by Bully Pulpit Interactive, which the campaign paid $100 million.

A Facebook spokesperson said ad costs fluctuated for both candidates during the campaign, and that at different times each candidate had the advantage. The spokesperson argued that Facebook’s ad platform usually limits the spread of polarizing messages — messages that reflect the views of a minority of the population — by making them more expensive to distribute. This occurs naturally, because News Feed algorithms resist sharing items that are likely to drive people away.

But even as Facebook makes unpopular messages more expensive to distribute, it can still promote polarization on a grander scale: thanks to Facebook’s microtargeting capabilities, voters can see fewer viewpoints that they might disagree with, because Facebook algorithms treat disagreement as bad for the user experience. On one hand, the actual impact of these dynamics is unknown; public data simply isn’t available. But on its face, the ad engine would seem to promote the creation and exploitation of filter bubbles.

In the pre-digital era, political candidates seeking to buy advertising competed on a relatively even playing field. Candidates buying airtime on television shows, ad space in newspapers, or postage for direct mail, are charged roughly equivalent prices to reach the same amount of people. Since 1927, the equal-time rule has required radio and television stations to offer equivalent time to political candidates if they ask for it, and at the same price. It originated out of fears that broadcasters would refuse to sell time to political candidates — or sell it to them at unfair rates.

The fact that one of the country’s dominant political ad platforms throttles messaging based on its shareability is a fact political campaigns must reckon with, and adapt to. “It’s the same dynamic that BuzzFeed and Jonah Peretti really mastered,” says Tim Wu, author of The Attention Merchants, an examination and critique of ad technology on Facebook and other platforms. “Vitality depends on strong emotional responses, hence polarized messages, whether advertising or news or whatever, travel further.”

The Trump campaign did not respond to a request for comment. It has previously said the company’s advertising tools were essential to its strategy. “Without Facebook, we wouldn’t have won,” Trump digital campaign official Theresa Wong said in a BBC documentary last month. Representatives of the Clinton campaign declined to comment.

To place an ad on Facebook, a political campaign has to win an automated auction. At any given time, millions of advertisers are competing to place ads in front of Facebook’s 2 billion-plus daily users. Advertisers can price their ads by the number of people who see it, the number of people who click on a link, or the number of people who engage with the ad, such as by watching a video or installing an app. Facebook averages out the cost of these various ads into a figure it calls an “eCPM” — the effective cost per 1,000 impressions.

The CPM is a standard measurement in the advertising industry. But Facebook’s ads differ from traditional ads in an important way: the company offers advertisers a monetary incentive to create more engaging ads. As users begin to click, share, and engage with an ad, Facebook begins showing it to more people. That lowers the eCPM, often allowing advertisers to reach a larger audience for the same amount of money. In some cases, Facebook’s automated systems will choose to display ads that had lower bids, if it believes the content of the ad will draw more engagement from users. The monetary goal of this system is to keep users scrolling through the News Feed, maximizing the number of ads that they encounter.

It’s the same approach that Facebook takes to the News Feed generally. Your friends’ status updates, links from publishers, and longform videos all compete for slots in the News Feed. Facebook shows each one to a subset of the account’s friends or followers, and expands that number as people read, watch, comment, and share. It’s the system that has turned the News Feed into one of the most engaging and lucrative products of all time — one that generated $17 billion in ad revenue in the first half of this year.

But the incentive structure that works so well for commercial advertising has unintended consequences when applied to political advertising. Over the past year, problems caused by the spread of more “engaging” content have repeatedly put Facebook on the defensive. Hoaxes and fictional news stories about the election were among the most engaging stories on Facebook in 2016, and the company’s algorithms generated a fire hose of readers for them as a result, generating as much as $10,000 a month in ad revenue for some solo publishers pushing fake news.

The form and content of political ads are largely unregulated — unlike commercial ads, where advertisers can be held accountable for the claims they make. Treating commercial and political speech as equals by rewarding viral posts can foster polarization.

“If you think of this as a game of incentives, that would incentivize campaigns to not only target their messages, but to target them in ways that would further inflame and polarize opinions,” said Michael Franz, co-executive director of the Wesleyan Media Project, which archives and analyzes political messages, “not only as a mechanism to increase support among your base, but also as a mechanism to make it cost-efficient.”

“Facebook’s profits depend on people coming back, clicking and sharing things,” said Alex Howard, deputy director of the Sunlight Foundation, which advocates for transparency in political advertising. “It’s not based on, ‘did we arrive at a resonated discourse on this policy proposal?’ or ‘did the best questions get asked at this town hall?’ or ‘did this politician get fact-checked on the lies that he or she was propagating?’”

Recently, Mark Zuckerberg laid out a nine-point plan primarily aimed at reducing the influence of foreign governments and other bad actors — requirements like labeling ads with disclosures about who paid for them and collecting them on pages where the public can inspect them. The steps, he said, would work “to protect election integrity and make sure that Facebook is a force for good in democracy.”

Despite the changes, the ads themselves will still be sold and distributed in the same way, offering the broadest and least expensive reach to the most-clicked and shared ads that candidates create.

Whether or not promoting built-to-share political ads swayed the 2016 election, it appears that, at the very least, Facebook’s algorithms granted them a wider reach. The tactic perfected by the Trump campaign — creating countless variations of ads, and then targeting them narrowly to as many niche groups as possible — will likely be studied and adopted by candidates around the world.

In the meantime, public interest groups and observers have called on the FEC to initiate a new rulemaking proceeding to create new regulations for political advertising on social media platforms. “They need to reexamine all the federal campaign finance laws and how they apply to new technology,” said Paul S. Ryan, vice president for policy and litigation at Common Cause. Ryan noted that the last such proceeding took place a decade ago. Former FEC commissioner Ann Ravel called for such a rulemaking proceeding three years ago, but was rejected by the commission’s Republican members.

Wu recently told New York magazine, “Facebook should be required to screen and disclose what their advertising practices are, how much people are paying, whether people get the same rates.”

So far, Facebook’s dynamic ad pricing has attracted little attention. But as congressional scrutiny of Facebook intensifies, that could change. And what happens — or doesn’t happen — could shape elections for years to come.