Audit of Facebook ad transparency finds missed political ads

Filling in the gaps of Facebook’s political ad library

After the attempts at election interference made via Facebook’s ad platform in 2016, Facebook fought off regulation by promising to bring transparency to political ads. Facebook said it would make public who is paying for political ads, along with basic facts such as how much they cost and dates they run. But our cybersecurity analysis shows vulnerability in Facebook’s transparency algorithms that reveals they routinely miss including political ads in its public archive.

The idea behind transparency: researchers, journalists, and the public have a way to understand the ads and hold political advertisers accountable. This is important, particularly for ads that mislead voters, misrepresent who is behind the ad, or suppress votes. However, it’s also crucial, even in the case of legitimate political advertising, that the public have information about who is trying to influence them and why, just as they do, for example, for political advertisements that run on TV.

[O]ur cybersecurity analysis shows vulnerability in Facebook’s transparency algorithms that reveals they routinely miss including political ads in its public archive.

And there are some things that we — a cybersecurity lab, the Online Political Transparency Project, housed at NYU’s Tandon School of Engineering–commend about Facebook’s approach to transparency. For example, we appreciate that Facebook’s criteria for what it includes in the political ad library are broad and includes ads focusing on social issues–which Google does not do.

But our 2019 analysis shows showed that Facebook routinely fails to include political ads as soon as they launch in the Facebook Ad Library, only identifying them as “political” after they’ve been running for some time. That observation left us wondering, though– how many political ads does Facebook never catch? Are there specific political issues or types of ads that are harder to identify? How consistent is enforcement generally?

In order to understand how well Facebook was doing at enforcing its disclosure policies, we decided to build a browser extension called Ad Observer, which allows Facebook users to share with us ads that are displayed on their Facebook feeds. This isn’t a new idea — ProPublica developed the Facebook Political Ad Collector in 2017, which several organizations promoted and supported. Other researchers have also developed similar tools, such as the UK-based Who Targets Me?

Why build something new? Well, first the Facebook Political Ad Collector had to battle through Facebook’s attempt to block it via technical means. But the Facebook Political Ad Collector project needed a way to collect ads from Facebook’s new site design, and from Youtube, and Who Targets Me? doesn’t publicly release the political ads collected by their users.

Ad Observer has been running for several months now, and we’re starting to have meaningful data to explore. Since our focus is on political advertising, we trained a classifier on Facebook’s corpus of ads included in their Ad Library under the labels “Social Issues,” “Elections,” and “Politics.” This classifier identifies content that is similar to content that’s already in the library; in other words, it’s a political ad classifier, for Facebook’s definition of “political.” We use this classifier to identify ads that weren’t disclosed as political to our browser extension users, but should have been. Then, we provide transparency about these as ones that Facebook missed, here.

We’re still in the early stages of this analysis, but it’s interesting to explore what we’re starting to see. Here’s one example of an ad Facebook missed:

As I’m writing this, this ad is actually still running, so I can see it via the web portal for non-political ads, although we can’t see how much was spent on the ad, or the age, gender or state breakdown of the people who were shown it:

But what’s really interesting about this ad is that it’s not just an ad, it’s also a ‘paid partnership’:

“Paid Partnership” is Facebook “ad-speak” for ads where the advertiser (in this case, Joe Biden) is paying an influencer (in this case, “We are mitú”) for the ad, rather than Facebook. Facebook does not generally provide information about these types of ads. The reason it’s visible through the Facebook Ad Library, and that it was caught by Ad Observer, is that in addition to running this as a Paid Partnership ad, someone also paid to run it as a Facebook ad. As to how Facebook wasn’t able to identify this ad with Joe Biden’s name and picture in it as political, we can’t say.

You can see eleven more examples of political ads Facebook missed from the past few weeks that we are highlighting we chose here. We’re going to keep updating Ad Observatory this page every few days with some of the more interesting examples, so check back soon as well!

We will continue to write about technical problems with Facebook political ad disclosure. Check out Ad Observatory, a searchable site revealing trends in Facebook political advertising in the 2020 elections, and download the Ad Observer plug-in tool to safely volunteer information for researchers and journalists on what ads you are seeing on Facebook.