Evidence · Digital & platforms

Algorithmic curation: how recommenders and playlists steer demand

On streaming platforms, recommendations, rankings and playlists decide much of what people encounter, and studies show this exposure measurably changes what they consume. Algorithm-driven listening goes with narrower individual taste, a randomized field experiment found personalised recommendations raised engagement while reducing individual diversity, and being featured by the platform lifts demand for the featured content. Effects differ by user and by type of curation, and most of the evidence comes from music streaming.

Definition

Algorithmic curation is the selection and ordering of content by software rather than by shelf space or human schedulers: recommendation lists, personalised home feeds, search rankings and generated playlists. Because it decides what a user sees first, it controls exposure, and exposure shapes demand. Editorial curation by platform staff, such as flagship playlists, sits alongside it and is often studied together with it.

What the research finds

Algorithmic listening goes with narrower consumption

In a large observational study of a music streaming service, listening that came through recommendations was associated with less diverse consumption. Users who broadened their taste over time did so mainly by moving toward music they found themselves, and more diverse users showed better conversion and retention.^[1]

Personalisation trades engagement for individual diversity

In a randomized field experiment on podcast recommendations, personalised suggestions raised podcast streams by 28.90% but cut the average individual diversity of what people streamed by 11.51%, while aggregate diversity rose by 5.96%. Users became more alike within themselves and more different from each other, and the effect spilled over into listening outside the tested feature.^[2]

Being featured by the platform drives demand

Across more than 30,000 popular playlists, being featured on the app's search page raised daily followers by 0.95%, about twice the effect of adding a song by an exceptionally popular major-label artist (0.45%). Placement on first-party playlists is itself studied as a possible source of platform bias, because it carries measurable promotional power.^[3]^[4]

Algorithmic and editorial curation can pull in opposite directions

A year of listening histories from about 9,000 users showed no single answer: algorithmic and editorial recommendations moved diversity in opposite directions depending on the type of user, leading the authors to speak of filter niches rather than filter bubbles. In another sample of 4,000 users, use of recommendations was modest and uneven; discovery features did lead listeners to less popular music, but they were a small part of listening, which stayed concentrated on the most popular artists.^[5]^[6]

Feedback loops can make behaviour more uniform

When recommenders are trained on data from users who were already shaped by earlier recommendations, simulations show that user behaviour becomes more homogeneous without any gain in the value users get. This is a modelling result, not an observed platform effect.^[7]

A cultural force, not only a technical one

Consumer culture research argues that the opacity, authority, non-neutrality and self-reinforcing nature of algorithms shape consumption at the individual, group and market level. This framework treats algorithms as neither simply empowering consumers nor simply controlling them, but as a space of both marketer control and consumer resistance.^[8]

How strong is the evidence?

Most of the evidence comes from music streaming, and much of it from a single platform, often using data and researchers from that platform. Several key studies are conference papers rather than journal articles, and most are observational; the podcast field experiment is the strongest causal design, but it tested one recommendation goal in one product area. Diversity is measured differently across studies (distance between songs, repetition, popularity of what is consumed), so apparent disagreements partly reflect measurement. The feedback-loop result comes from simulation, and the consumer culture account is conceptual. How far these findings carry over to retail, news or advertising platforms is not established by the studies reviewed here.

References

Ashton Anderson, Lucas Maystre, Ian Anderson, Rishabh Mehrotra, Mounia Lalmas (2020). Algorithmic Effects on the Diversity of Consumption on Spotify. ACM.
David Holtz, Benjamin Carterette, Praveen Chandar, Zahra Nazari, Henriette Cramer, Sinan Aral (2020). The Engagement-Diversity Connection: Evidence from a Field Experiment on Spotify. Proceedings of the 21st ACM Conference on Economics and Computation.
Max J. Pachali, Hannes Datta (2024). What Drives Demand for Playlists on Spotify?. Marketing Science.
Luis Aguiar, Joel Waldfogel, Sarah Waldfogel (2021). Playlisting Favorites: Measuring Platform Bias in the Music Industry. International Journal of Industrial Organization.
Quentin Villermet, J\\'er\\'emie Poiroux, Manuel Moussallam, Thomas Louail, Camille Roth (2021). Follow the Guides: Disentangling Human and Algorithmic Curation in Online Music Consumption. Proceedings of the 15th ACM Conference on Recommender Systems.
Jean-Samuel Beuscart, Samuel Coavoux, Sisley Maillard (2019). Les algorithmes de recommandation musicale et l'autonomie de l'auditeur. R\\'eseaux.
Allison J. B. Chaney, Brandon M. Stewart, Barbara E. Engelhardt (2018). How Algorithmic Confounding in Recommendation Systems Increases Homogeneity and Decreases Utility. Proceedings of the 12th ACM Conference on Recommender Systems.
Massimo Airoldi, Joonas Rokka (2022). Algorithmic Consumer Culture. Consumption Markets & Culture.

Bibliographic details only. Follow the DOI to read each study at its publisher.

Building on this evidence?

ScholarSignal checks a new research idea against the closest existing work and an editor rubric, and tells you whether the contribution is new enough for the journals you target.

Evaluate my idea for free