Within Algorithms
Why new artists are hard to recommend
New artists need listener data to be recommended, but often need recommendations before they can collect that data.
On this page
- What the cold start problem means for music
- Why early signals matter for algorithmic confidence
- How exploration systems can give new tracks a chance
Page outline Jump by section
Introduction
Recommendation systems help listeners navigate enormous music catalogues, but they face a fundamental challenge when dealing with brand-new artists. Most recommendation engines learn from behaviour signals such as saves, skips, completion rates, repeat plays and playlist additions. A new artist, however, has little or none of this data. The result is a classic “cold-start” problem: an artist needs listener engagement to earn recommendations, yet often needs recommendations to generate that engagement in the first place. This creates one of the most important bottlenecks in modern music discovery, because recommendation quality depends on evidence, while new releases arrive with almost no evidence attached. [Sander Dieleman]sander.aispotify cnnsWe want to be able to recommend new music right when it is released, and we want to tell listeners…Read more…
What the cold-start problem means for music
In recommendation research, a cold start occurs when a system lacks sufficient interaction data to make confident predictions. For music platforms, this affects both new songs and new artists. Collaborative filtering—the technique that finds patterns among listeners with similar tastes—works best when many people have already interacted with a track. If nobody has listened yet, there is little behavioural information to compare. Sander Dieleman LinkedIn This creates a structural disadvantage for emerging artists. Established acts arrive with followers [linkedin.com]linkedin.comHow Spotify recommender system worksIf it finds similarities in songs that they listen to, the system will recommend songs from one perso…, historical listening patterns and extensive engagement histories. Their new releases can immediately be connected to known audiences. A debut artist, by contrast, may upload a track that the system understands only through limited metadata or audio characteristics. The platform can estimate what the music sounds like, but it cannot yet know who will actually enjoy it. [arXiv]arxiv.orgarXiv A Deep Multimodal Approach for Cold-start Music RecommendationarXiv A Deep Multimodal Approach for Cold-start Music Recommendation
The challenge is intensified by scale. Tens of thousands of new tracks can arrive on major streaming services each day, making it impossible to expose every release broadly enough to gather robust behavioural data. Platforms must decide which tracks receive valuable early opportunities for listener feedback. [Hypebot]hypebot.comBut what does it mean for artists and labels hoping to standHypebotThe Cold Start Problem and What Spotify Algorithms Mean…April 25, 2023 — 25 Apr 2023 — Spotify recently shared how its recommen…
Why early signals matter for algorithmic confidence
Recommendation systems become more confident when they observe consistent patterns of listener behaviour. A track that attracts saves, repeat listens and playlist additions sends a stronger positive signal than one that is frequently skipped. These interactions help the system estimate whether the music should be shown to more people with similar tastes. [Vohnic Music]vohnicmusic.comhow spotify algorithm works 2026Vohnic MusicHow Spotify's Algorithm Really Works in 20261 Mar 2026 — Spotify tracks save rate, completion rate, skip rate, and repeat lis… [Spotify]spotify.comSpotifyUnderstanding recommendations on SpotifyAs you engage with Spotify, actions such as searching, listening, skipping, or saving to Y… [Loop Solitaire]loopsolitaire.co.ukspotify algorithmic playlistsSave Rate · 2. Repeat Listens · 3. Playlist Adds · 4. Monthly Followers · 5. Completion Rate / Skip Rate · 6. Web and…Read more…
For a new artist, the first wave of listeners can therefore have an outsized influence. Early engagement effectively serves as training data for the recommendation model. If the system observes strong positive responses from a small audience segment, it gains confidence that the track may appeal to a wider group. Conversely, weak early signals leave the model uncertain about where the music belongs. [Spotify Research]research.atspotify.comaccelerating creator audience building through centralized explorationSpotify ResearchAccelerating Creator Audience Building through Centralized…23 Feb 2024 — All Spotify recommendation systems have been…
Several types of behaviour are particularly informative:
- Saves and library additions suggest a listener wants future access to the track.
- Repeat plays indicate sustained interest rather than curiosity alone.
- Playlist additions imply the listener considers the track valuable enough to keep.
- Completion rates show whether listeners stay engaged throughout the song.
- Skip behaviour, especially very early skips, can signal a poor match between recommendation and listener expectations. [iMusician]imusician.proiMusicianHow Spotify Changes Affect the Music Industry10 Nov 2025 — Key signals include track completion rates, repeat listens, saves, pl… [Spotify]spotify.comSpotifyUnderstanding recommendations on SpotifyAs you engage with Spotify, actions such as searching, listening, skipping, or saving to Y… [Loop Solitaire]loopsolitaire.co.ukspotify algorithmic playlistsSave Rate · 2. Repeat Listens · 3. Playlist Adds · 4. Monthly Followers · 5. Completion Rate / Skip Rate · 6. Web and…Read more…
Because these signals arrive gradually, recommendation systems typically become more accurate over time. The difficulty is that new artists need enough exposure to generate those signals in the first place.
Why popularity can reinforce the barrier
Cold-start challenges often interact with popularity bias. Recommendation engines are generally designed to maximise the likelihood that users enjoy what they hear next. Content with strong historical performance provides safer predictions than content with little history. As a result, established artists can receive additional exposure simply because the system has more evidence about them. [Ones To Watch]resources.onestowatch.comOnes To Watch How Music Discovery Algorithms Recommend New ArtistsOnes To WatchHow Music Discovery Algorithms Recommend New ArtistsApril 1, 2026 — 1 Apr 2026 — New artists lack this historical data, whic…
This does not necessarily mean platforms deliberately suppress newcomers. Rather, uncertainty itself becomes a disadvantage. A recommendation model may know with high confidence that an established artist will satisfy a listener, while having only a rough estimate for an unknown artist. When engagement optimisation is a primary objective, the safer choice often wins. [Spotify Research]research.atspotify.comaccelerating creator audience building through centralized explorationSpotify ResearchAccelerating Creator Audience Building through Centralized…23 Feb 2024 — All Spotify recommendation systems have been…
The outcome can resemble a feedback loop:
- Popular artists generate more engagement data.
Amazon book picks
Further Reading
Books and field guides related to Why new artists are hard to recommend. Use these as the next step if you want deeper reading beyond the article.
The Lean Startup
Rating: 3.0/5 from 33 Google Books ratings
Useful for understanding experimentation and early signal gathering.
eBay marketplace picks
Marketplace Samples
Example marketplace items related to this page. Use the search link to explore similar finds on eBay.
- More data improves recommendation confidence. [medium.com]medium.com23, with a dedicated team of over 100 data scientists and…Read more…
- Higher confidence leads to more recommendations.
- More recommendations generate even more data.
New artists must somehow enter this cycle without already possessing the evidence that drives it. [Sander Dieleman]sander.aispotify cnnsWe want to be able to recommend new music right when it is released, and we want to tell listeners…Read more…
How exploration systems can give new tracks a chance
Recommendation designers have long recognised that relying only on proven favourites limits discovery. Modern systems therefore balance two competing goals: exploitation, which recommends content expected to perform well, and exploration, which deliberately tests less-certain content to gather information. [Spotify Research]research.atspotify.comaccelerating creator audience building through centralized explorationSpotify ResearchAccelerating Creator Audience Building through Centralized…23 Feb 2024 — All Spotify recommendation systems have been…
Exploration acts as a mechanism for overcoming cold starts. Instead of showing only tracks with extensive engagement histories, the platform occasionally introduces songs whose potential is uncertain. The resulting listener responses become new data points that help the system learn. Research on music recommendation frequently frames this as an exploration–exploitation trade-off, where some recommendation opportunities are effectively used as experiments. [arXiv]arxiv.orgarXiv A Deep Multimodal Approach for Cold-start Music RecommendationarXiv A Deep Multimodal Approach for Cold-start Music Recommendation ResearchGate Streaming services employ several strategies to make these experiments more informed: [researchgate.net]researchgate.netTo learn…Read more…
Using content instead of behaviour
When behavioural data is scarce, platforms can analyse the music itself. Audio features, genre indicators, textual descriptions and artist metadata provide clues about similarity to existing music. This allows a track to be placed near comparable artists even before substantial listener feedback exists. [arXiv]arxiv.orgarXiv A Deep Multimodal Approach for Cold-start Music RecommendationarXiv A Deep Multimodal Approach for Cold-start Music Recommendation
Testing with small audiences
Rather than exposing a new song to millions of users immediately, platforms can show it to smaller groups whose listening histories suggest potential interest. Positive engagement from these listeners gives the system greater confidence to expand distribution. [Spotify Research]research.atspotify.comaccelerating creator audience building through centralized explorationSpotify ResearchAccelerating Creator Audience Building through Centralized…23 Feb 2024 — All Spotify recommendation systems have been…
Centralised discovery mechanisms
Spotify researchers have publicly described efforts to accelerate audience building through exploration-focused systems designed to help new creators overcome cold-start challenges. These approaches attempt to identify promising matches between creators and listeners earlier than traditional recommendation pipelines would allow. [Spotify Research]research.atspotify.comaccelerating creator audience building through centralized explorationSpotify ResearchAccelerating Creator Audience Building through Centralized…23 Feb 2024 — All Spotify recommendation systems have been…
The tension at the centre of music discovery
The cold-start problem reveals a deeper tension within streaming recommendations. Listeners generally want accurate suggestions, but accurate suggestions depend on historical evidence. New artists, meanwhile, need visibility before such evidence exists.
Too little exploration can make discovery feel repetitive and favour already successful artists. Too much exploration can reduce recommendation quality and frustrate listeners with irrelevant suggestions. Recommendation design therefore becomes an exercise in managing uncertainty: deciding how much risk to take on unfamiliar music in order to discover future favourites. [Spotify Research]research.atspotify.comaccelerating creator audience building through centralized explorationSpotify ResearchAccelerating Creator Audience Building through Centralized…23 Feb 2024 — All Spotify recommendation systems have been…
For new artists, success on streaming platforms often depends on crossing this initial information gap. Once enough listeners generate meaningful engagement signals, recommendation systems can begin to understand where the music fits and whom it should reach. Until then, the artist exists in the cold-start zone—a period where the platform knows the music exists but has not yet learned who is most likely to love it. Sander Dieleman [Spotify Research]research.atspotify.comaccelerating creator audience building through centralized explorationSpotify ResearchAccelerating Creator Audience Building through Centralized…23 Feb 2024 — All Spotify recommendation systems have been…
Endnotes
-
Source: sander.ai
Title: spotify cnns
Link: https://sander.ai/2014/08/05/spotify-cnns.htmlSource snippet
We want to be able to recommend new music right when it is released, and we want to tell listeners...Read more...
-
Source: arxiv.org
Title: arXiv A Deep Multimodal Approach for Cold-start Music Recommendation
Link: https://arxiv.org/abs/1706.09739 -
Source: linkedin.com
Link: https://www.linkedin.com/pulse/how-spotify-recommender-system-works-daniel-roy-cfaSource snippet
How Spotify recommender system worksIf it finds similarities in songs that they listen to, the system will recommend songs from one perso...
-
Source: hypebot.com
Title: But what does it mean for artists and labels hoping to stand
Link: https://www.hypebot.com/the-cold-start-problem-and-what-spotify-algorithms-mean-for-musicians/Source snippet
HypebotThe Cold Start Problem and What Spotify Algorithms Mean...April 25, 2023 — 25 Apr 2023 — Spotify recently shared how its recommen...
Published: April 25, 2023
-
Source: spotify.com
Link: https://www.spotify.com/safetyandprivacy/understanding-recommendationsSource snippet
SpotifyUnderstanding recommendations on SpotifyAs you engage with Spotify, actions such as searching, listening, skipping, or saving to Y...
-
Source: imusician.pro
Link: https://imusician.pro/en/resources/blog/how-spotify-changes-affect-the-music-industrySource snippet
iMusicianHow Spotify Changes Affect the Music Industry10 Nov 2025 — Key signals include track completion rates, repeat listens, saves, pl...
-
Source: arxiv.org
Link: https://arxiv.org/abs/1311.6355Source snippet
arXiv[1311.6355] Exploration in Interactive Personalized Music...by X Wang · 2013 · Cited by 215 — This paper presents a new approach to...
-
Source: researchgate.net
Link: https://www.researchgate.net/publication/274478765_Exploration_in_Interactive_Personalized_Music_RecommendationSource snippet
To learn...Read more...
-
Source: arxiv.org
Link: https://arxiv.org/abs/1812.03226 -
Source: arxiv.org
Title: arXiv Machine Learning Approaches to Hybrid Music Recommender Systems
Link: https://arxiv.org/abs/1807.05858 -
Source: researchgate.net
Link: https://www.researchgate.net/publication/319285554_A_Deep_Multimodal_Approach_for_Cold-start_Music_RecommendationSource snippet
audio information with user feedback data using deep network...Read more...
-
Source: researchgate.net
Title: 381853790 SPOTIFY RECOMMENDATION SYSTEM
Link: https://www.researchgate.net/publication/381853790_SPOTIFY_RECOMMENDATION_SYSTEMSource snippet
(PDF) SPOTIFY RECOMMENDATION SYSTEM1 Jul 2024 — This research explores the augmentation of user-artist interaction on Spotify by developi...
-
Source: artist.tools
Title: spotify playlist recommendations
Link: https://www.artist.tools/post/spotify-playlist-recommendationsSource snippet
They respond to track context, skip behavior, saves, playlist co-...Read more...
-
Source: research.atspotify.com
Title: accelerating creator audience building through centralized exploration
Link: https://research.atspotify.com/2024/02/accelerating-creator-audience-building-through-centralized-explorationSource snippet
Spotify ResearchAccelerating Creator Audience Building through Centralized...23 Feb 2024 — All Spotify recommendation systems have been...
-
Source: loopsolitaire.co.uk
Title: spotify algorithmic playlists
Link: https://loopsolitaire.co.uk/blog/spotify-algorithmic-playlists/Source snippet
Save Rate · 2. Repeat Listens · 3. Playlist Adds · 4. Monthly Followers · 5. Completion Rate / Skip Rate · 6. Web and...Read more...
-
Source: vohnicmusic.com
Title: how spotify algorithm works 2026
Link: https://vohnicmusic.com/blog/how-spotify-algorithm-works-2026Source snippet
Vohnic MusicHow Spotify's Algorithm Really Works in 20261 Mar 2026 — Spotify tracks save rate, completion rate, skip rate, and repeat lis...
-
Source: resources.onestowatch.com
Title: Ones To Watch How Music Discovery Algorithms Recommend New Artists
Link: https://resources.onestowatch.com/music-discovery-algorithms-recommend-artists/Source snippet
Ones To WatchHow Music Discovery Algorithms Recommend New ArtistsApril 1, 2026 — 1 Apr 2026 — New artists lack this historical data, whic...
Published: April 1, 2026
-
Source: research.atspotify.com
Link: https://research.atspotify.com/publications/explore-exploit-explain-personalizing-explainable-recommendations-with-banditsSource snippet
Spotify ResearchPersonalizing Explainable Recommendations with BanditsThe multi-armed bandit is an important framework for balancing expl...
-
Source: andrmusic.co
Title: Spotify Metrics That Trigger [Discover Weekly]({{ ‘discover-weekly/’ | relative_url }})
Link: https://andrmusic.co/behind-the-music/spotify-metrics-trigger-discovery/Source snippet
AndRSpotify metrics that trigger algorithmic recommendations see 300-1000% growth in 90 days. The exact completion rates and skip pattern...
-
Source: themetalverse.net
Title: spotify algorithm 2026
Link: https://www.themetalverse.net/spotify-algorithm-2026/Source snippet
Look at your save rate, completion rate, and where your streams are coming from.Read more...
-
Source: pyimagesearch.com
Title: spotify music recommendation systems
Link: https://pyimagesearch.com/2023/10/30/spotify-music-recommendation-systems/Source snippet
playlists that balance exploration and exploitation (i.e., finding new songs that the user might like and playing songs that the user alr...
Additional References
-
Source: research.atspotify.com
Link: https://research.atspotify.com/2025/9/personalizing-agentic-ai-to-users-musical-tastes-with-scalable-preference-optimizationSource snippet
Agentic AI to Users' Musical Tastes with...23 Sept 2025 — At Spotify, our goal is to build systems that learn listening preferences dire...
-
Source: beatstorapon.com
Link: https://beatstorapon.com/blog/ultimate-guide-to-spotify-music-algorithm/Source snippet
Beats To Rap OnSpotify's Music Recommendation Algorithm: The Complete...1 Mar 2025 — Discover our ultimate guide to Spotify's music algo...
-
Source: medium.com
Link: https://medium.com/%40briansrebrenik/introduction-to-music-recommendation-and-machine-learning-310c4841b01dSource snippet
Introduction to Music Recommendation and Machine...These music recommendation systems are part of a broader class of recommender systems...
-
Source: music-tomorrow.com
Link: https://www.music-tomorrow.com/blog/how-spotify-recommendation-system-works-complete-guideSource snippet
Inside Spotify's Recommendation System: A Complete...1 Sept 2025 — Discover how Spotify's recommendation algorithms work...
-
Source: github.com
Link: https://github.com/anthonyli358/spotify-recommender-systemsSource snippet
However, as we will see later it is a good method to mix in for variety and to avoid the cold-...Read more...
-
Source: orphiq.com
Link: https://orphiq.com/resources/spotify-algorithm-independent-artistsSource snippet
skip it (especially before 30 seconds), how often they add it to their...Read more...
-
Source: medium.com
Link: https://medium.com/%40myliemudaliyar/behind-the-playlist-analyzing-spotifys-recommendation-system-5044a13f5ccfSource snippet
23, with a dedicated team of over 100 data scientists and...Read more...
-
Source: artistrack.com
Title: Optimize your music to unlock Discover
Link: https://artistrack.com/spotify-algorithm-skip-rate-save-rate/Source snippet
Decoding the Spotify Algorithm: Skip Rate, Save Rate, &...11 Dec 2025 — Learn the 3 crucial signals for the Spotify Algorithm: Skip Rate...
-
Source: hal.science
Link: https://hal.science/tel-04865002/file/Modeling%20and%20Influencing%20Music%20Preferences%20on%20Streaming%20Platforms.pdfSource snippet
Modeling and Influencing Music Preferences on Streaming...by K Matrosova · 2024 · Cited by 3 — • Cold start problem: CF struggles with b...
-
Source: youtube.com
Title: How Streaming, Algorithms and AI Have Changed the Way We Listen to Music
Link: https://www.youtube.com/watch?v=CTkyDu7aIGkSource snippet
The ISRC "Loophole" for Millions of Streams (Legally)...
Topic Tree