YouTube recommendations – A Cornell University study using “male-coded” and “female-coded” viewing bots found YouTube’s recommendation system can steer political content in noticeably different directions, even when both groups start with the same interest in News & Politics.
For 150 straight interaction sessions, the bots watched YouTube the same way a viewer might: click, watch, let the next video appear. But the results didn’t look the same for everyone.
A new study published in Cornell University’s arXiv repository suggests YouTube’s recommendation algorithm may be shaping political perspectives differently for men and women—despite both groups starting with identical interest in the YouTube News & Politics category. The researchers didn’t change the politics they asked the bots to like. They changed the viewing habits they fed into the system.
To run the experiment. researchers created 160 automated social bots and split them into two groups: 80 accounts coded “male” and 80 coded “female.” Both sets of accounts showed identical interest in YouTube’s News & Politics category at the start. Then the study watched how the recommendations evolved over time as each account completed 150 consecutive interaction sessions.
The bot setup was designed to mimic different lifestyle viewing patterns. The “male-coded” accounts were programmed with viewing habits associated with traditionally male-oriented content—gaming and sports. The “female-coded” accounts were assigned habits tied to traditionally female-oriented content, including fashion, lifestyle, and vlog videos.
Over repeated sessions, the study reports the recommendations drifted in sharply different directions. Male-coded accounts were reportedly more frequently directed toward confrontational and politically charged topics such as crime. law enforcement. immigration. and defense-related issues. The same accounts were also reportedly shown more content linked to powerful state institutions like Immigration and Customs Enforcement (ICE) and the Department of Justice.
Female-coded accounts, by contrast, reportedly encountered a broader mix of political content. Their recommendations leaned toward international affairs, culture, arts, and lifestyle-related policy discussions. Researchers also found these accounts received more politically neutral recommendations overall.
The tightest difference may be about how the algorithm keeps users inside certain lanes. The study claims that male-coded profiles became trapped inside tighter recommendation loops, repeatedly encountering overlapping videos that reinforced similar viewpoints. Female-coded accounts experienced a more varied and less concentrated information ecosystem.
That matters because YouTube isn’t a niche site for political exploration—it’s one of the world’s largest content platforms and an increasingly influential source of news and political information. During the 2020 US election cycle, political campaigns invested heavily in YouTube advertising to influence voters and shape narratives online.
This study shifts attention away from paid promotions and toward the recommendation engine itself: the system that decides what plays next.
Jonathan Gray. codirector of the Center for Digital Culture at King’s College London. said the findings contribute to growing concerns about algorithm-driven political influence and online radicalization. Gray also argued that recommendation systems remain largely opaque despite their enormous societal impact.
The research lands in the middle of an intensifying global debate about whether large tech platforms unintentionally amplify polarization by creating personalized echo chambers around users. With scrutiny on AI-driven recommendation systems rising worldwide. studies like this are adding weight to calls for platforms such as YouTube to provide greater transparency about how their algorithms shape public discourse and political behavior.
And for viewers, the takeaway is unsettling: even when the starting point is the same interest in political content, the path the algorithm builds can diverge—subtly at first, then steadily—until it starts to feel like a different internet entirely.
YouTube recommendations arXiv study Cornell University political polarization algorithmic influence echo chambers online radicalization social bots gendered viewing behavior ICE Department of Justice