Can a platform's recommendation feed quietly favor some creators over others, even when no one meant it to?

This explores whether recommender systems can tilt visibility toward some creators and away from others, whether a platform does it on purpose or the system's design produces it without anyone deciding to. The short answer is yes. The less obvious part is that the most widespread favoritism in this collection isn't deliberate. It comes out of ordinary engineering choices that look neutral. The collection has little on platforms deliberately boosting particular creators, such as self-preferencing or paid promotion. It is much richer on how favoritism gets built into the machinery.

The deliberate route does show up. Research on recommendation feeds describes them as persuasion infrastructure. The weights a platform assigns in its feed don't just decide what users see. They also change what creators produce, because creators adapt to whatever gets rewarded How do recommendation feeds shape what people see and believe?. So a platform that adjusts its feed is choosing which creators win, and it is also shaping which kinds of creators exist in the first place.

The surprising route is accidental. One study finds that when a recommender's embeddings are too small (the compact numerical profiles it builds for each user and item), the system learns to fall back on popular items because that's the easiest way to score well on ranking accuracy. Niche items get too little exposure, so they never gather the engagement data that might have lifted them. The effect compounds over time and can't be cleaned up afterward. The researchers argue that embedding size should be treated as a fairness setting, not just a technical one Does embedding dimensionality secretly drive popularity bias in recommenders?. A related finding: systems tuned purely for accuracy over-serve each user's dominant interests and crowd out their smaller ones. One fix is to re-rank results afterward so they reflect a user's full mix of tastes Why do accuracy-optimized recommenders crowd out minority interests?. Another is to model each user as several distinct personas rather than one averaged profile, which brings diversity back naturally Can modeling multiple user personas improve recommendation accuracy?. From a creator's point of view, both problems mean the same thing: if you serve a minority taste, the default system is quietly working against you.

The recommender's structure also matters. Recommendation networks like "frequently bought together" and "also viewed" draw different audiences, and those audiences rate the same products differently Do different recommender types shape opinion convergence differently?. Which widget puts you next to which neighbor affects how your work gets judged. A newer twist: AI-generated posts can win engagement while building no human creator's reputation. They push human influencers aside even as the platform keeps making money Does AI content displace human influencers on social media?. If ranking rewards engagement alone, the system can end up favoring content that has no creator behind it.

The takeaway that may be new to you: asking whether platforms can favor certain creators assumes the neutral option is a level playing field. The research suggests there isn't one. Every recommender favors someone, whether through embedding size, its accuracy target, or which content it puts side by side. The real question is whether anyone measures that favoritism and treats it as a choice.