The persuasive power of platforms like YouTube has long been apparent. It’s why the Trump campaign, for instance, bought out the masthead ad space at the top of YouTube 20 times during the 2020 election cycle, including an audacious buyout on Election Day.
But the platform’s algorithm can also be politically persuasive. A new study published in the Cornell University repository arXiv suggests YouTube’s recommendation system actively directs male and female users into vastly different political information environments, even when their initial political interests are identical.
Researchers deployed 160 automated social bots: 80 programmed with what the researchers called male-coded viewing habits like sports and gaming, and 80 with female-coded habits like style and vlogs. Both groups were given the exact same baseline interest in YouTube’s News & Politics category. (The authors did not respond to an interview request; among the questions we would have asked was how reliable it is to stereotype viewing habits this way.) The bots then completed 150 consecutive interaction steps so that researchers could track where the recommendation algorithm led them.
While female-coded accounts actually encountered a higher overall volume of political videos, the kinds of issues recommended diverged sharply depending on whether the account displayed male- or female-coded habits.
Male-coded profiles were disproportionately funneled toward a narrow set of confrontational domestic issues, including law, crime, and defense. They were also pushed heavily toward state-power entities like Immigration and Customs Enforcement and the Department of Justice. In contrast, female-coded accounts were presented with a broader, more moderate mix of macroeconomic and lifestyle-related public policy topics, including international affairs, culture, and the arts. Female-coded profiles also received significantly more neutral political content, while male-coded profiles were shown more polarizing videos.
“YouTube is one of the most widely used platforms on the planet, yet its algorithms remain opaque and poorly understood,” says Jonathan Gray, codirector of the Center for Digital Culture at King’s College London. Gray was not involved in the study but reviewed its findings.
The recommendation system also trapped male-coded profiles in a highly concentrated network of overlapping videos, creating a cohesive echo chamber in which they repeatedly encountered the same content. Female-coded profiles, meanwhile, experienced a far more diffuse and differentiated information network.
“For many it is a primary source for news, advice, and guidance,” Gray says. “In a moment where platforms are promoting increasingly misogynistic and extremist content, this study contributes to a growing body of work investigating the role that their algorithms play in shaping society, culture, and politics, highlighting an urgent need for greater public scrutiny and oversight.”