Welcome to part two of our exploration of Facebook’s Ad Library and Ad Library API – an application programming interface, or set of computer protocols, designed to let researchers and journalists perform more robust and in-depth analysis on the ads in the library. The Facebook Ad Library (and API) are publicly accessible record of paid election, political and social issues ads that have been posted to Facebook in certain countries where Facebook have a major presence. In Part 1, we gave a brief overview of what the Ad Library and Ad Library API are and how to access them. This post provides an overview of the capabilities of these tools (as available in mid-July 2019), compares them to the US-version of Twitter’s Ad Transparency tool (the Canadian version of Twitter’s Ad Transparency tool is not yet available, ETA August 2019). The post concludes with some general recommendations for how the Facebook Ad Library and its API can be improved.
What did we find using the Ad Library and its API?
In looking through the existing Ad Library and the recentlyreleased Ad Library API for Canada, there are a few interesting things that jumped out at us. First, regardless whether you are searching using the Ad Library or the Ad Library API (specifically the search_terms parameter), it appears that you cannot query within individual search fields. For example, if you search for “New Democratic Party,” it will return anything and everything with “New Democratic Party” in it, whether it is the body of text in the ad, the name of the funding entity, the name of the page administering the ad, etc. Using the Ad Library, it is possible to search just for page names, but for everything else, searching within certain fields or aspects of an ad requires visually reviewing the search results or sorting through the data manually if it is collected via the API. On the surface, this limitation seems inconsequential, in practice this design flaw (the lack of bulk data access) introduces barriers that makes it more difficult for researchers and journalists who are working with these tools to conduct more fine-grained inquiry in a timely manner. As a result, this tool is better suited for historical analysis than for real-time or even near real-time election ad transparency and accountability tool for fast moving and dynamic events such as elections.
Second, we noticed many inactive ads are missing dates for when they stopped running (See Fig. 1). This is true whether the ad is accessed via the Ad Library’s web interface or the API. It is unclear why, especially given that some other inactive ads do report end dates (e.g. See Fig. 2).
With respect to political ads specifically, classification of whether ads are political or issue-oriented seems to not always be accurate. There appear to be several seemingly apolitical ads that were taken down because they were subsequently deemed by Facebook to be political or issue-oriented and no disclaimer about who is paying for them was provided.
Ads that are taken down for this reason tend to be annotated with the following message: “This ad ran without a disclaimer. After the ad started running, we determined that the ad was about social issues, elections or politics and required the label. The ad was taken down.” Take for example the following ad (Fig. 2) promoting a local food co-op:
This is a seemingly broad application of what constitutes political advertising, resulting in removal of many ads lacking disclaimers (presumably because the advertisers didn’t believe that they were political ads). Conversely, some have found examples of ads that do not appear in the Ad Library, yet are more unambiguously political. These may be outliers, but any possibility of false positive or false negative determinations of political content should be investigated, as they may represent blind spots or loopholes that could be exploited.
What we couldn’t do with Facebook’s Ad Transparency Tools for Canada
As noted earlier, there is no way to retrieve the entire contents (bulk data access) of an Ad Library at once. Both the Ad Library and the Ad Library API require an input search term to function, thus requiring users to know the terms or issue that they want to examine and unnecessarily limiting the scope of the ads returned. Without the ability to review all the ads on the platform, research on Facebook ads might be subject to sampling bias and provide an incomplete picture of the ad landscape in Canada. Those conducting such research would have to rely on using many search terms to try and minimize the number of ads that are not captured through the searches and attempt to predict what keywords would cover the most ground. These various searches would then have to be aggregated and checked for duplicates.
Additionally, as part of the Ad Library, Facebook also released an Ad Library Report (see Fig. 3) which provides some statistics about ads from a specific country, including: the total number of ads in the Ad Library, total amount spent country-wide, by location, and by specific advertisers, as well as the ability to filter all of the ads by pages that have the most ads or spent the most on them. Ad Library Reports also offer the option to download the full report containing a list of all of the advertisers who have run ads in a given time period, with some high-level information such as page IDs, page names, total amount spent, number of ads in the library, and who they disclaim is funding them.
Perhaps the most notable omission from the Ad Library and Ad API is information about the microtargeting of ads. That is, the targeting of ads towards specific demographics of individuals – something that society became collectively aware of after the recent Cambridge Analytica scandal. Facebook’s Head of Public Policy, Kevin Chan, has stated that there are no plans for disclosing microtargeting information about ads, arguing that Facebook microtargets using numerous inferences from user profiles and that this is “consistent with how the marketing industry works generally.” But this is a curious omission and position to take considering that when an ad is served to a Facebook user on the Facebook platform, users can already click the ellipsis button on it to reveal information about why they are seeing the ad, including targeting information (see Fig. 4). Facebook’s position on this contrasts with Twitter’s, which does provide a detailed breakdown of demographics that advertisers intended to target on a per-ad basis as part of their equivalent of the ad library.
Ad Transparency Tools for Facebook vs. Twitter
While Twitter is not accepting political ads on its platform in Canada until the general election period, it is worth comparing how the two tech giants stack up in terms of their general efforts to provide researchers, journalists, and users with transparent information about ads being run on their platforms. Twitter’s equivalent of Facebook’s Ad Library is their Ads Transparency Centre. Since the Canadian version of Twitter’s Ad Transparency tool is not yet available (ETA August 2019), for this comparison we used the US version of the Twitter’s Ads Transparency Centre. (We expect that the Canadian Version of the Twitter’s Ads Transparency Centre will share much of its DNA with the US version.)
Both Facebook and Twitter’s web-based offerings are available freely to all, and offer similar features. Both show when ads started and finished running, show disclaimers, show total spent and money spent on a per-ad basis, and permit searching through both accounts and the ads that those accounts are running. Interestingly, both also exempt news organizations from disclaiming their ads as political or issue-related to avoid conflating journalism with political action or propaganda.
Beyond the base features discussed above, the two offerings diverge considerably, with differences summarized in Table 1 below. It is important to note that the affordances of both of these ad transparency tools are still very new and iterative, so changes may have already been made by the time you read this.
Note: An ad campaign refers to a collection of ads that are run with the same objective or targets.
| Feature | ||
| Archive Length | 7 years | 7 days |
| Countries | 61 nations | US, EU, Australia, India, Canada (later this year) |
| Archive Content | All active and inactive ads, political or issue-oriented or otherwise | No inactive or deleted ads, or ads from deleted/suspended accounts |
| API Availability | Yes | No |
| Aggregate Advertiser Spend | Only in US | Always shown |
| Impression Statistics | Per ad: Ranges of total impressions and percentages of viewership broken down by region and by gender per age range. | Per ad: Rounded, specific values of total impressions Per ad campaign: Rounded total impressions, impressions broken down across age ranges, regions (including metropolitan areas), genders, and languages |
| Spend Information Format | Per ad: Ranges | Per ad and per ad campaign: Rounded, specific values |
| Targeting Information | None | Disclosed per ad campaign |
| Advertiser Transparency Information | Page name changes, page manager location(s), page merge(s) | Billing information, including payment method and address |
| Served Ads Link to Corresponding Ad Library (Facebook) or Ad Transparency Centre (Twitter) pages | No | Yes, via clicking “Learn More” on the Tweet |
| Advertiser Certification Process for Political and Issue Ads | Two-factor authentication, confirmed mailing address via mailed verification code, valid identification, and disclaimer | Profile photo, header photo, associated website with handle’s online presence and valid contact info, Political Advertiser Certification form (includes mailing address and government-issued photo ID for individuals or Employee Identification Number/Individual Taxpayer Identification Number for organizations |
Table 1: Differences between Facebook Ad Library & Twitter Ads Transparency Centre (as of mid July 2019)
As a note, certification processes (see the last item in Table 1 above) vary by country and are governed by legislation and regulation in that country. Facebook Canada’s minimum requirements are compared with Twitter’s requirements for US-based advertisers, and thus are not a direct comparison.
You can see the differences in how Facebook and Twitter report spent and impression metrics in Figures 5A to 5C below, which shows metrics for a similar ad run on both platforms by the American Civil Liberties Union (ACLU).
Conclusions
Facebook and Twitter’s Ad Transparency tools are good first efforts, and any progress to make political advertising more transparent should be encouraged and commended. However, if the intent is to make political ads more transparent and to hold advertisers to account, there is much still that can be done to make these tools more useful.
As a start, Facebook should allow for bulk data access of the content of the ad library. In addition, Facebook should start providing some of the additional information that Twitter currently discloses about ads in its Ads Transparency Centre. These include replacing the ranges of advertiser spend and impressions with specific values, more detailed breakdowns of impressions within different groups of demographics, linking to corresponding Ad Library pages for served ads, and providing page names, payment method, and payment addresses. Information about the pages associated with an ad that are currently only accessible through the Ad Library’s web interface, such as advertiser contact information, should also be searchable through the Ad Library API as well.
According to Facebook, the company uses a combination of artificial intelligence and human reviewers to classify ads in Canada, so further refinement of one or both of those methods may be warranted to produce more accurate classifications. Based on our preliminary review, the classification of whether an ad is political or issue-related needs more work, as we and others have seen examples of mis-classification of ads (e.g. false positives and false negative classifications).
Similarly to other organizations like the Electronic Frontier Foundation, Mozilla, and ProPublica, we too are calling for greater disclosure of microtargeting information. As mentioned above, users may be able to see some information about why they were targeted when they have ads served to them, but this information can sometimes be opaque (e.g., belonging to “lookalike audiences” of people who have liked similar Facebook pages as you), and the types of data collected about users that enable those targeting options can be similarly opaque. At a minimum, the “Why Am I Seeing This Ad?” targeting information shown to users should be reflected in the Ad Library, no matter how many separate demographics that an ad may target.
The information currently accessible through the Facebook Ad Library and Ad Library API details who ultimately ends up seeing the ads and represents only half of the puzzle of political advertising on the Internet. The other half of the puzzle, and the one that many are most interested in a post-Cambridge Analytica world, is who was intended to see those ads. For Canadians, there is still time between now and the start of the general election period for Facebook to fill in this part of the puzzle and help give Canadians a more complete picture of how advertising on Facebook may be affecting and shaping their politics.
Note: For inquiries, comments, or corrections, please contact us at info[at]socialmedialab.ca
*Written By Arvin Jagayat, Anatoliy Gruzd and Philip Mai