Amid the current uproar surrounding social media platforms’ policies towards political ads, Facebook founder and CEO Mark Zuckerberg continues to insist that “Ads on Facebook are already more transparent than anywhere else.” He claims that this is due to Facebook’s ad library, created following the 2016 election. “We have a political ads archive so anyone can scrutinize every ad that’s run — you can see every message, who saw it, how much was spent — something that no TV or print media does.” Unfortunately, Facebook advertising works very differently than TV or print media advertising. More unfortunately, the ad library is not the paragon of transparency Zuckerberg describes. The limited information made available in the library, on top of its lack of basic tools for sorting and reviewing, invalidate any claims of transparency.
In a nutshell, the library, launched in 2018, increases accountability by storing and making available for public view — for seven years — all advertisements about political or social issues. With the ad library, as their story goes, if a campaign is delivering one message to a certain group and a contradictory message to another, this can come to light. It does not store all advertisements, as one might think based on Zuckerberg’s above comment. Ads deemed non-political are not stored in the library, but now must be visible on the Page running them for as long as they are active. To give credit where credit is due, this is still a major change from advertising in 2016, and it is a big step forward. In 2016, ads came and went from select Facebook News Feeds with no trace. Much of the controversy Facebook is embroiled in over its decision not to fact-check advertisements from (some) political figures emerged because it makes these ads available for review. The ad library certainly helps. But as long as Facebook can target ads at the level at which it does, it is no model of transparency and accountability.
Before continuing, I should make clear that I cannot write code or interact with Facebook’s tools for developers. I spent an afternoon trying to see if I could use the Facebook API by opening YouTube videos for beginners and deciding they were all too advanced for me. It is possible (although, based on other articles that use research from the ad library, pretty unlikely) that more information is available if it is accessed this way. If this is the case, I consider this another flaw. The ad library is supposed to provide information and transparency, neither of which should be limited to dedicated researchers, developers, or whoever can hire one of those. To live up to Zuckerberg’s bold claims, the ad library should be accessible for any Facebook user being targeted by all these advertisements.
Most of the inherent danger of Facebook’s advertising lies in its ability to target ads to the right segments of the population where they will be most effective — the exact right segment of the population. Facebook has so many data points about everyone on the site that it can, beyond the information willingly shared with the platform, predict political views, sexual orientation, ethnicity, religion, substance abuse, and whether someone’s parents stayed together until they were 21 years old. A 2014 study found that a computer model could predict personality based on Likes — ten likes, and it was more accurate than a coworker; seventy and it was more accurate than a friend; 150 likes, and the system was more accurate than a parent or sibling. Facebook has bragged in marketing pitches that it can determine when teenagers are feeling insecure, anxious, or depressed. In 2017, Facebook was caught offering options to target audience categories such as “Jew hater”. In October 2019, Cambridge Analytica whistleblower Christopher Wylie said that Steve Bannon directed Trump’s 2016 campaign advertising at ‘incels’, who he described as “unmarried straight white dudes who can’t get laid”, in swing states. Campaigns can upload voter rolls through the Custom Audiences targeting tool and use these to direct their ads. Facebook allows advertisers to place what they call a “pixel” on their own website, away from Facebook, to match their visitors to their Facebook profiles, and target those individuals with Facebook advertisements. Once advertisers have identified users to target, they can ask Facebook to send the advertisement to a “Lookalike audience” — people that Facebook has identified as similar to the audience they have chosen. Facebook delivers the right ad to the right person, at the right time, wherever and whenever it will have the most effect.
It is with this in mind that the ad library results seem insultingly vague. The library allows searches by keyword, topic, or organization, or by page. If you search a popular enough keyword to have lots of results from different pages, they are returned in a sort-of-chronological jumble. The option to sort results by, say, date, is hardly rare on today’s internet, but Facebook’s ad library offers no sorting capability at all. So perhaps you found the ad you are after. You ask for the details. Zuckerberg says we can see how much was spent. The library shows how much was spent within a range: less than $100; between $100 and $199; between $100 and $499; between $200 and $299; between $500 and $999… They are broad ranges, and they often seem to overlap.
Facebook also shows how many impressions the ad has — how many times it was shown. Again, it does this in vague and overlapping ranges: less than one thousand; 1,000–2,000; 1,000–5,000; 10,000–15,000; 40,000–45,000… The brackets sometimes get larger as the numbers go up, but there is little consistency I can find. An ad promising the truth about the “MMR Vaccine’s poison pill” reached somewhere between 10,000 and 50,000 individuals. Another posted by anti-vaccine nonprofit that promoted medical conspiracy theories and fear about vaccines reportedly reached “50K — 100K” people, and there is no way to find more detailed numbers. I’m not sure the reason for such a lack of clarity, but it hardly seems transparent.
Zuckerberg also says we can scrutinize who saw every ad. Facebook does offer broad categories about who advertisements are delivered to, but there is not room for much in the way of scrutiny. A bar graph shows how many people of each age bracket (18–24, 25–34, 35–44, 45–54, 55–64, and 65+) saw the ad, divided by gender (men, women, and unknown). Below, on a map of the U.S. (I have not looked into other countries), Facebook breaks down by state where the advertisement was shown; for example, 10% in Texas, 9% in Michigan, 1% in Wyoming, <1% in Ohio, and so on. Marking the full extent of the ad details provided, Facebook offers a handy little link to the other ads being run by that advertiser. Now, Facebook targets advertisements by postal code for a number of countries, the United States being one of them, and then layer other filters on top: people in a certain zip code who read Breitbart or The New York Times. Facebook can target down to an individual level, knowing exactly which person will be most receptive to which ad, but we are left to content ourselves with knowing that 9% of the 50K — 100K people who saw a certain advertisement were in Michigan.
Looking at the ad library returns, I can’t help but wonder: Why is there no information about the specific zip codes targeted by an ad? Why does Facebook not share exactly how many people were exposed to an ad? Why not provide more information about who was targeted — Democrats? Republicans? People who like cats? People who own guns? Young single men when they are feeling depressed? Facebook has, in fact, already blocked a project that sought to shed light on this targeting data. At a basic level, information about whether a candidate is targeting their own supporters or those of their opponent would provide valuable information about campaigns and voter suppression. It would also help reveal instances of bias or discriminatory practices.
After facing five discrimination lawsuits accusing Facebook of using its targeting capabilities to exclude certain groups (among them African Americans, Jews, older individuals, and people whose interests include “guide dogs for the blind” and “wheelchair ramps”) from housing, employment, and credit offer advertisements, Facebook reached a settlement in 2019. The company agreed to separate these types of ads, limiting their targeting to only a few hundred interest categories and preventing targeting of a geographic area smaller than a 15-mile radius. Political ads currently remain exempt from such limitations, though Facebook’s own employees have called for similar restrictions to be placed on political advertising. Other calls to ban microtargeting of political advertisements have been raised, including by the Chair of the US Federal Election Commission (FEC) Ellen Weintraub and Facebook’s own former chief security officer Alex Stamos, but Facebook has indicated no intention to place limitations on itself.
As with housing, employment, and credit offer ads, knowing who is targeted will bring divisive or exclusionary campaigns to light. By keeping this information hidden, Facebook only enables further polarization. We know what messages are being shared, but we are unable to tell who is seeing them, and the audience can be key to understanding the full impact of an ad. One Russian-sponsored ad focused on Black Lives Matter, targeting individuals in Baltimore, MD and Ferguson, MO. It was crafted to seem supportive of Black Lives Matter, while at the same time portraying the group as threatening to some residents of those cities. In a study of Facebook advertising leading up to the 2016 election, researchers analyzed targeting of demographic groups. Users of lower incomes were specifically targeted with ads focusing on immigration and racial conflict. Ads prioritizing nationalism were almost exclusively directed to white Facebook users. This information had to be collected by recruiting volunteer participants, who installed a browser extension that captured ad information, and then working backwards from there to see which groups received which ad topics. Facebook makes none of this available in its ad library, meaning it is content to not only allow, but also obscure, the delivery of different political information to different zip codes, income levels, ethnicities, and personalities. As the FEC Chair wrote, “The microtargeting of political ads may be undermining the united character of our United States.” Considering the vast amounts of data Facebook holds about everyone, even those without Facebook profiles, the ad library represents the bare minimum of information Facebook could condescend to share. As long as Facebook enables microtargeting of political advertisements, we need at the very least to know who is seeing which message.
