Original title: Recommendation Quality and the Concentration of Consumption: Experimental Evidence from Netflix
Guy Aridor, Winston Chou, Nathan Kallus, Antoine Scheid +2 · August 2026
Do recommendation algorithms make everyone watch the exact same blockbusters, or do they push people toward obscure titles? Researchers tested this on Netflix with an experiment. When the recommendation system improved, people watched more overall and relied more on its suggestions. Crucially, the better system spread out what people watched. It shifted viewing away from the most popular hits, known as superstarsThe most popular titles that get the most viewership, like blockbuster movies.. Instead, it guided users toward moderately popular titles, called the middle-tailModerately popular titles that get a decent amount of views, but aren't the biggest hits.. At the same time, it had minimal effects on the most niche titles, known as the long-tailVery niche titles that only a small number of people watch.. This shows that better recommendations do not split viewing between only giant hits and obscure picks. Instead, they boost medium-sized titles as algorithms get better.
Imagine a town where everyone used to buy their books from a single mega-bookstore that only stocked bestsellers. Then a librarian starts suggesting books based on what each person likes. At first, people stick to the bestsellers. But as the librarian gets better at matching people to books, more readers start borrowing moderately popular titles that weren't on everyone's radar. The mega-bookstore's sales don't collapse, but they spread out to many more titles. In this paper, better Netflix recommendations shift viewing from the biggest hits to a wider range of moderately popular shows, much like that improved librarian spreads out borrowing.
If you've worried that algorithms only push you toward what everyone else watches, this study shows the opposite: better recommendations actually spread attention away from mega-hit 'superstars' and toward moderately popular titles. For Netflix and similar platforms, that means investing in a wider range of shows becomes more valuable as their algorithms improve—potentially changing what types of content get made.
The abstract does not specify how the recommendation improvement was implemented (e.g., which algorithm change), the exact magnitude of effects, or whether results generalize beyond Netflix. It also does not test causal mechanisms explaining why consumption diffuses to the middle-tail specifically.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Recommendation Quality and the Concentration of Consumption: Experimental Evidence from Netflix — Guy Aridor, Winston Chou, Nathan Kallus et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.