Research Talk by Anita Rao, Georgetown University (US)

As part of the Marketing Research Seminar Series, Anita Rao presented joint work with Ziwei Cong on a question that platforms are quietly wrestling with every day: can ranking algorithms be used as a tool for content moderation, and if so, what gets lost along the way? The paper leverages a natural experiment on a large online knowledge-sharing platform that shifted from a simple upvotes-minus-downvotes ranking to an algorithm that penalizes downvotes more heavily. The change had exactly the intended effect on content creators, who started producing more neutral, less polarizing answers. But viewers noticed: "Unhelpful" clicks rose by 1.2% and "Thank" clicks dropped by 6.3%, suggesting that the toned-down content was also perceived as less useful. Further analysis shows that the mechanism runs through visibility — creators respond to what the ranking rewards — and the paper closes with a broader discussion of what all of this means for platform design.

What I liked about this talk was that it puts numbers on a trade-off that usually stays pretty abstract in public debates about content moderation. The idea that more "neutral" content might be less polarizing but also less valuable to the people actually looking for answers is not a new one, but seeing it identified so cleanly in a real-world setting made it feel much more concrete. It also led to a really good discussion afterwards about what platforms should even be optimizing for, and about how much of our online experience is quietly shaped by ranking decisions that most users never see.