Using the Meta Ad Library
Abridged from the forthcoming paper in First Monday
The Meta Ad library in the UK contains what is currently the most comprehensive library of adverts delivered and funded by governmental organisations in the UK. Unlike other countries, the Meta Ad Library for the UK is well populated with public sector ads, due to much higher rates of adoption. For other countries , the library mostly contains ads related to elections and political candidates.
Advertisers and ads are being added to the database all the time: while some organisations with registered ‘Pages’ have advertising histories dating from 2020 (the current database dates from 2020), there are many that have listings that start in Jan or Feb 2023. We suspect this is an artefact of the registration process, rather than that these organisations having just started advertising.
The Ad Library itself presents methodological challenges to researchers. The broader issues of working with datasets provided by platforms (or indeed any organisation) are increasingly well-documented, with ‘platform epistemologies’ mediated by the partial datasets provided through APIs, platform logics, and other biases or lenses (Soares, 2018).
The Meta Ad Library consists of two databases with entries for each ad linked by a unique identifier, one database containing information about the ad and its reach, the other about the targeting . Meta provides two public interfaces – the “Ad Library”, which can be searched by country and keywords used in ads, including by organisation name (registered Facebook ‘Page’ name), and returns all the ads, copy, spend, potential reach and actual impression of active and inactive ads related to that keyword or organisation, and the “Ad Library Report” offers a way to range top spending advertisers, search by location, and monitor particular advertisers. The Ad Library API, open to vetted researchers, provides a Jupyter Notebook interface, allowing direct searching of the two databases directly using SQL and processing and export of results. This is accessed through a VPN and run on a server owned by Meta; we created researcher accounts with Meta in order to access these tools, which included making an application describing our intended research.
In analysing these data, we first searched a range of UK public sector advertisers. These were conducted in monthly sweeps on the 20th of the month from January to June 2023. Some generic searches, like NHS (National Health Service) generate a range of local health trusts, national online services, and specialist services. We ran queries based on advertisers paying in GBP, with names including ‘Police’, ‘Constabulary’,’NHS’,’Council’, and ‘Department’, then curated this dataset to remove irrelevant advertisers (such as political campaigns for local council seats). We additionally directly searched the name of every government department, local council, and police force in the UK. This resulted in a dataset of 15,325 adverts from 363 advertisers. We restrict ourselves to a qualitative analysis of adverts in order to avoid making misleading quantitative claims about the dataset, which we acknowledge is only a partial picture. Relative numbers of ads in the dataset may not reflect higher use by different bodies (instead simply showing ads that have been captured by Meta) and these only reflect advertising on Meta, which is only a single platform among many.
The Meta Ad library is the latest iteration of a number of policy changes that have resulted from years of political pressure and legislation concerned with the potential political impact of advertising deliberately intended to destabilise democracies. The revelations that psychographically targeted ads had been distributed according to personal data collected on Facebook by Cambridge Analytica, and the purchasing of polarising Facebook ads by the Russian Internet Research agency, led to intense government and public pressure on Internet platforms. In 2018/9 Facebook, Twitter and Google introduced requirements for advertisers to declare the identity and funding of ‘political’ ads, and started to provide reports and ‘‘libraries’ of declared political ads that had been bought on their platforms (Leerssen et al 2021). This also coincided with closure of APIs and Facebook’s controversial attacks on researchers using scraping techniques, obliging researchers to register and use specific research APIs (Beraldo et al 2021). Twitter and TikTok declared they would not take election-related advertising at all, but have struggled to enforce this. For Meta, all advertisers are currently required to register their ‘Page’ with valid credentials. After the first wave of political advertisers, government agencies in the UK (including security and law enforcement) are being validated and registered. This is an ongoing process, and over the first months of 2023 increasing numbers of government agencies were included. Occasionally the library interface shows a public sector ad being deleted either because it does not include a disclaimer or the advertising Page is not registered in the correct way. What also emerges from exploration of Meta policies on government advertising is the degree to which they have become invested in Government as a major client for their services. While much of this has been focused on the messaging tools such as Whatsapp or Pages, it also includes marketing and support for paid advertising as a means to engage citizens and attain policy goals.
After a number of iterations and criticisms of the design (e.g. Edelson, 2020; Pochat et al, 2022; Mehta and Erickson, 2022; Scott 2022), the scope and rigour in detecting ‘political’ ads and enforcing transparency has slowly improved, and has also been extended by Meta in particular to include ‘state controlled media’, and a broader range of organisations advertising on ‘social’ issues (Rosen 2019). Nonetheless, it should not be treated as ‘accurate’ or ‘comprehensive’ and the researcher can only access data that the platform has chosen to make available, or that is generated as part of their business processes. Thus it provides a useful minimum, non-exhaustive sample of campaigns run over the past two years by different bodies. The vast computer-based systems such as Facebook are also continually manipulated by other users, and automatic filtering and control is limited in its ability to populate the database correctly. Nonetheless we would caution against treating the Ad Library as an accurate dataset of purely quantitative analysis.
All advertising is regulated, and the major platforms have a complete set of policies governing advertising behaviour . However, by the end of 2022, the requirements and Ad library of Meta are far ahead of the other major platforms, requiring registration, permitting centralised platform-level auditing, and providing research access to ad data, targeting, and impressions via the “FORT “secure interface. Google is also beginning to provide complete libraries of all ads being run, and in Mid-2023 Tiktok released its open database of all commercial ads, and extended a pilot research portal providing more detailed access. Alongside the requirements for transparency, there are a number of restrictions that apply, as a result of sectorial regulations at advertisers in “special ad categories’ the electoral, political and ‘social’ ads particularly those relating to credit, housing and employment. The platforms reserve the right to take down ads in these categories, and promise to publish the targeting, spending and impression data for each. Ads registered in these categories are restricted in the use of some of the more advanced targeting techniques – such as uploading customer data and finding ‘lookalike’ audiences (Meta, 2022).
Links and References
Meta Business Help Centre (2022) Targeting transparency information for ads about social issues, elections or politics. Available at: https://en-gb.facebook.com/business/help/736091520909332 (accessed 9 March 2023).
Advertising Policy page, Meta https://transparency.fb.com/en-gb/policies/ad-standards/
Meta Ad Library https://www.facebook.com/ads/library/?active_status=all&ad_type=political_and_issue_ads&country=GB&media_type=all (Accessed 8 MArch 2023)
Meta Business Help Centre 2022. About ads about social issues, elections or politics. Available at: https://en-gb.facebook.com/business/help/167836590566506 (accessed 9 March 2023).
Twitter Political Content Policy https://business.twitter.com/en/help/ads-policies/ads-content-policies/political-content.html (Accessed 20 July 2023)
Guy Rosen, Katie Harbath, Nathaniel Gleicher, and Rob Leathern.Helping to Protect the 2020 US Elections. Facebook. Oct. 21, 2019. URL: https://about.fb.com/news/2019/10/update-on-
election-integrity-efforts/
Facebook/Meta Ads Library https://www.facebook.com/ads/library/
Tiktok Ads Library https://library.tiktok.com/ads?region=GB
L. Edelson 2020. Publishing Facebook ad data (redux). In: Online Political Transparency Project. Available at: https://medium.com/online-political-transparency-project/publishing-facebook-ad-data-redux-ff071c41c12e (accessed 8 March 2023).
D. Beraldo, S. Milan, J. de Vos, et al. 2021. Political advertising exposed: tracking Facebook ads in the 2021 Dutch elections. Berlin Internet Policy Review.
S.C. Boerman, S. Kruikemeier, F.J. Zuiderveen Borgesius 2017. Online Behavioral Advertising: A Literature Review and Research Agenda. Journal of Advertising 46(3). Routledge: 363–376. DOI: 10.1080/00913367.2017.1339368.
I. Goodwin 2022. Programmatic alcohol advertising, social media and public health: Algorithms, automated challenges to regulation, and the failure of public oversight. International Journal of Drug Policy 109: 103826. DOI: 10.1016/j.drugpo.2022.103826.
P. Leerssen, J. Ausloos, B. Zarouali, et al. 2019. Platform ad archives: promises and pitfalls. Internet Policy Review 8(4).
S. Mehta, K. Erickson 2022. Can online political targeting be rendered transparent? Prospects for campaign oversight using the Facebook Ad Library. Internet Policy Review 11(1).
V. L. Pochat, L. Edelson and T. V. Goethem 2022. An Audit of Facebook’s Political Ad Policy Enforcement. In: Proceedings of the 31st USENIX Security Symposium, Boston, MA, 2022. Available at: https://www.usenix.org/conference/usenixsecurity22/presentation/lepochat.
A. T. N. Soares 2018. Epistemology, methods and theories of communication in the Big Data Era: a critical panorama of social media research. Comunicação e sociedade, (33), 167-181.
K. Sender 2018. The Gay Market is Dead, Long Live the Gay Market: From Identity to Algorithm in Predicting Consumer Behavior. Advertising & Society Quarterly 18(4). Advertising Educational Foundation. DOI: 10.1353/asr.2018.0001.
