Facebook’s Election Disinformation Moderation: Progress in the US, Gaps Remain Globally
The 2021 attack on the US Capitol served as a stark reminder of the potential consequences of misinformation spreading unchecked on social media platforms. In response, Facebook, now under the umbrella of its parent company Meta, pledged to curb the proliferation of political content on its platform. This included reducing the visibility of such content in users’ feeds and shutting down Crowdtangle, a tool utilized by researchers to track viral trends. As the 2023 US elections approached, Meta CEO Mark Zuckerberg appeared to further distance himself from political matters, delegating key responsibilities and reportedly downsizing the election integrity team. This raised concerns about the platform’s commitment to combating election disinformation.
To assess the effectiveness of Facebook’s content moderation efforts, a recent investigation tested the platform’s ability to identify and reject false election information. Researchers submitted eight sample ads containing clear disinformation, bypassing the standard ad authorization process specifically designed for political advertising. These ads were crafted to directly violate Facebook’s stated policies, which prohibit the spread of misinformation about voting procedures, eligibility, and candidate participation. The results revealed a marked improvement compared to previous tests conducted in 2022.
While Facebook rejected seven out of the eight ads submitted, one misleading ad slipped through the cracks. This problematic ad falsely claimed that a valid driver’s license is required to vote, a blatant misrepresentation of US voting laws. Although only a minority of states require photo identification, and even those do not mandate a driver’s license specifically, the ad was approved, exposing a vulnerability in Facebook’s moderation systems. While the rejection of the majority of the test ads suggests improvement since 2022, where acceptance rates for similar disinformation ads ranged between 20% and 30%, the fact that any such ad was approved remains a significant concern.
The simplicity of the disinformation used in the test ads raises further questions about Facebook’s ability to detect more nuanced and sophisticated forms of manipulation. The blatant nature of the test ads should have made them easy to identify and reject. The fact that one still slipped through highlights the limitations of automated moderation systems and the need for human oversight. The investigation’s findings underscore the ongoing challenge of effectively combating election disinformation, particularly as bad actors employ increasingly sophisticated tactics. The ease with which blatantly false information can still bypass Facebook’s defenses raises serious doubts about the platform’s preparedness for more subtle forms of manipulation.
Furthermore, while Facebook appears to have made strides in addressing election disinformation within the US, its global track record paints a less encouraging picture. A previous investigation in Brazil revealed a stark contrast, with Facebook approving 100% of ads containing election disinformation. Even after being informed of these findings, a retest showed that while Facebook’s detection processes had improved, they still approved half of the resubmitted disinformation ads. This discrepancy in performance raises concerns about the equitable application of content moderation policies across different regions and elections. The stark difference between the US and Brazil results suggests that resource allocation and enforcement efforts may be disproportionately focused on certain regions, leaving others vulnerable to manipulation.
The inability of Facebook to consistently detect and remove even blatant election disinformation across its global platform underlines the urgent need for comprehensive and consistent enforcement of its policies. This concerning trend suggests that the resources dedicated to combating disinformation, including human moderators and advanced detection algorithms, may be unevenly distributed. The lack of response from Meta to inquiries about these findings further compounds concerns about transparency and accountability. Addressing this disparity requires a commitment to investing in robust moderation systems and ensuring their effective deployment across all languages and regions, not just those deemed politically significant. The failure to address these global vulnerabilities undermines trust in the platform and risks further erosion of democratic processes worldwide.