How Spotify Recommends New Music: 2026 Algorithm Guide

Spotify recommends new music through a hybrid system it calls "algotorial," where human editors and machine learning algorithms work together to surface tracks you are likely to love. Your every action shapes this process. When you listen to a full song, skip at the 15-second mark, save a track to your library, or search for a specific mood, Spotify logs that signal and updates what it calls your taste profile. Playlists like Discover Weekly and Release Radar are the most visible outputs of this system, refreshing weekly with picks calibrated to where your taste sits right now, not where it was six months ago.

The system does not just feed you more of what you already know. Spotify's personalization approach aims for long-term satisfaction rather than the immediate click, which means it actively pushes you toward artists you have never heard while keeping enough familiar ground to hold your attention.

Here is what drives every recommendation you receive:

• Taste profile: Built from your listening history, skips, saves, searches, and followed artists
• Collaborative filtering: Matching your profile against millions of listeners with overlapping but not identical tastes
• Audio analysis: Acoustic features like tempo, key, and energy extracted from each track
• Natural language processing (NLP): Scanning song metadata, reviews, and editorial text to understand genre and mood
• Editorial curation: Human editors building candidate pools that algorithms then personalize
• Real-time feedback: Hundreds of billions of events processed daily to keep recommendations current

***

How human editors shape what Spotify's algorithm works with

Spotify's editorial teams do not just run flagship playlists. They act as the first filter in the entire recommendation pipeline, and that role is more technical than most listeners realize.

Editors build what Spotify calls a "pool," a large collection of candidate tracks for a given playlist theme, mood, or genre. Rather than picking the exact final order of songs, they select which tracks are even eligible for the algorithm to work with. That distinction matters because it lets editors bring cultural expertise and trend awareness into a system that would otherwise rely entirely on engagement data. A playlist like RapCaviar reflects editorial judgment about what is culturally relevant in hip-hop right now, not just what got the most streams last week.

Editor-curated pools give the algorithm room to work with a wider range of tastes, including emerging artists who lack the listening history to surface organically. Editors also monitor performance metrics within playlists, tracking which tracks are connecting and which are getting skipped, and adjust the pool accordingly. The algorithm then takes that pool and personalizes the final selection and order for each individual listener.

Benefits and limitations of editorial curation in Spotify's system:

• Cultural depth: Editors catch regional trends and cultural moments that pure engagement data misses
• Emerging artist access: Human judgment can include newer artists before they have enough streams to register algorithmically
• Quality control: Editors filter out tracks that technically fit a genre but would feel out of place
• Scale limitation: No editorial team can manually curate for over 600 million listeners individually, which is exactly why the algorithm handles the final personalization step
• Potential bias: Editorial taste reflects specific cultural perspectives, which can underrepresent certain regional or niche genres

For artists, understanding this layer is useful. Getting into an editorial pool is a separate goal from getting algorithmically recommended, and both paths matter. Spotify's own editorial playlist strategies are worth studying if you are trying to reach either gate.

***

How Spotify builds your taste profile and turns data into recommendations

Your taste profile is the core of how Spotify suggests songs. It is not a static list of your favorite artists. It is a high-dimensional representation of your listening behavior, updated continuously as you interact with the app.

The data signals Spotify collects

Spotify's recommendation algorithms draw on four main categories of input. Each one captures a different dimension of who you are as a listener.

| Data type | Examples | Role in recommendations | |---|---|---| | Listening behavior | Play completions, skips, saves, repeat listens | Builds and refines your taste profile in real time | | User context | Location, language, device, age, followed artists | Signals genre preferences and regional relevance | | Content metadata | Genre, release date, audio features, podcast category | Identifies similar tracks and content clusters | | Collaborative signals | Actions of listeners with overlapping taste profiles | Surfaces tracks you haven't heard but similar users love |

Collaborative filtering: the "near-match" engine

Collaborative filtering is the technique behind Discover Weekly's most surprising picks. Spotify's Vice President of Personalization, Oskar Stål, described it this way: imagine you and another listener share four of the same top artists, but your fifth artist is different. Spotify takes that near-match and suggests each person's fifth artist to the other. Now scale that to millions of simultaneous comparisons, and you have a system that can find genuinely new music for you based on the collective behavior of listeners who think like you do.

Audio analysis and NLP

Spotify analyzes the acoustic features of every track, things like tempo, key, energy, and danceability, to find sonic similarities across songs. This is how it can recommend a track you have never heard that still feels right for a particular mood or activity. NLP adds another layer by processing song titles, artist bios, playlist descriptions, and music press coverage to understand how a track is talked about and categorized in the real world. Together, these two methods let the algorithm understand music content independently of how many people have streamed it, which is critical for surfacing new releases.

Key algorithmic methods

• Collaborative filtering: Matches your profile to listeners with similar but not identical tastes
• Audio feature analysis: Compares acoustic properties across tracks to find sonic neighbors
• NLP on metadata: Extracts genre, mood, and cultural context from text associated with tracks
• High-dimensional user embeddings: Maps each listener into a vector space for scalable, fast retrieval
• Real-time event processing: Updates taste profiles from hundreds of billions of daily events across the platform

***

How you can actively shape your Spotify recommendations

You are not a passive recipient of Spotify's algorithm. Every action you take is an input, and some actions carry more weight than others.

Finishing a song is a strong positive signal. Skipping before the 30-second mark is one of the clearest negative signals the system receives. Saving a track to your library tells the algorithm this is something you want to return to, which pushes similar music higher in future recommendations. Searching for a specific mood or genre, like "late night jazz" or "aggressive workout," feeds intent data directly into your taste profile.

When you first sign up, Spotify asks you to select favorite artists during onboarding. This solves what engineers call the cold-start problem: without any listening history, the algorithm needs a starting point. Those initial picks prime the system immediately, so your first Discover Weekly is already reasonably personalized rather than generic.

Practical ways to influence your recommendations:

• Play songs to completion when you enjoy them; partial listens send mixed signals
• Skip quickly on tracks you dislike so the algorithm registers the negative feedback
• Save tracks you love to your library, not just to playlists
• Follow artists you want to stay updated on, since followed artists feed Release Radar directly
• Use the Enhanced playlist toggle on your own playlists to let Spotify insert suggestions that fit the vibe
• Search with descriptive terms rather than just artist names to signal mood and context preferences
• Select favorite artists carefully during onboarding, since those choices shape your early recommendations significantly

Non-listening signals also matter. Your device, language setting, and general location all give the algorithm context about what is culturally relevant to you. If you set German as your language, Spotify may surface German-language podcasts and artists. If you primarily listen on a desktop, it may weight that context differently than mobile listening.

***

How commercial factors influence what Spotify recommends

Spotify's recommendations are not purely driven by what you love. Commercial considerations play a role, and the platform is transparent about it.

Discovery Mode is the clearest example. Artists and labels can flag specific songs as priorities, and Spotify's system adds that signal to the algorithms that determine personalized listening sessions. When Discovery Mode is active for a song, Spotify charges a commission on streams of that song in those areas of the platform. The result is that the flagged track has a higher probability of appearing in your recommendations than it would organically.

The key constraint is that Discovery Mode does not override the engagement filter. Spotify only recommends songs it believes you are likely to enjoy. If a promoted track gets skipped repeatedly, the algorithm pulls back on recommending it, even with the commercial signal active. This keeps the system from degrading into pure advertising.

Commercial factors that affect recommendation probabilities:

• Discovery Mode activation: Increases a song's likelihood of appearing in personalized sessions
• Commission-based streaming: Spotify earns a cut of streams generated through Discovery Mode placements
• Engagement monitoring: Tracks promoted via Discovery Mode are still subject to skip-based filtering
• Editorial playlist exclusion: Discovery Mode signals do not apply inside editorially curated playlists
• Content cost and monetization: Whether Spotify can monetize a piece of content may influence how prominently it surfaces

The balance Spotify strikes here is worth understanding, especially for artists. Commercial promotion can get a song in front of new listeners, but only consistent engagement keeps it there. The algorithm's long-term satisfaction goal acts as a natural check on purely commercial recommendations.

***

The algotorial system and what 2026 updates actually changed

Spotify's 2026 architecture represents a meaningful step forward from the system that powered Discover Weekly in its early years. The changes are not cosmetic. They reflect a fundamental rethinking of how user identity is represented at scale.

Large-scale user embeddings

The 2026 framework maps over 600 million monthly listeners into high-dimensional embeddings within a stable vector space. Each listener becomes a compact numerical representation that captures both long-term preferences and short-term shifts, like a mood change or a new genre phase. Downstream systems, whether they handle playlist generation, search ranking, or home page curation, all draw from this shared representation rather than building separate user models from scratch.

The framework uses an autoencoder architecture: one component compresses multi-signal user data into a compact embedding, and another reconstructs the original inputs to verify that the embedding preserves enough information to be useful across different tasks. Near-real-time updates mean that if you spend a weekend listening to nothing but ambient electronic music, your embedding shifts within minutes, not days.

The Mostra multi-objective framework

Most recommendation systems optimize for a single goal: maximize the probability that you engage with the next track. Mostra, Spotify's multi-objective recommendation framework, optimizes for several goals simultaneously.

| Mostra objective | What it means in practice | |---|---| | User satisfaction | Recommending tracks the listener is likely to play fully | | Discovery | Surfacing songs from artists the listener has never heard | | Emerging artist exposure | Prioritizing tracks from artists early in their career | | Boosting | Elevating songs tied to cultural moments or platform priorities |

Mostra uses counterfactual reasoning to decide when to deviate from pure satisfaction optimization. It asks: if I swap the top-ranked track for one that serves a creator-centric goal, how much does predicted satisfaction drop? If the drop falls within a set threshold, the creator-centric track wins the slot. This lets Spotify promote emerging artists and diverse content without noticeably degrading your listening experience.

The multi-armed bandit framework

Spotify's multi-armed bandit approach addresses one of the hardest problems in recommendation: how do you keep showing people what they love while still introducing genuinely new music? The framework balances exploitation, recommending tracks based on confirmed preferences, with exploration, testing uncertain content to learn more about your evolving taste. Without this balance, the algorithm would eventually trap you in a loop of the same 50 songs.

The algotorial system ties all of this together. Human editors set the cultural context and candidate pool. The embedding framework captures who you are as a listener. Mostra decides which objectives to weight in a given session. The bandit framework decides how much to push into unfamiliar territory. Playlists like Discover Weekly and Release Radar are the consumer-facing outputs of this entire stack, refreshed weekly with picks that reflect both your current taste and Spotify's broader goals for the platform.

Understanding the Spotify algorithm in 2026 also matters for artists who want to reach new listeners through these systems. The same signals that shape your recommendations as a listener are the signals that determine whether a new track gets surfaced to someone who has never heard of the artist. Music storytelling and cross-platform presence also feed into how Spotify's NLP layer categorizes and contextualizes a track, as explored in this 2026 guide to music storytelling.

***

Key Takeaways

Spotify's recommendation system works because it combines human editorial judgment, real-time behavioral data, and multi-objective machine learning to balance what you love with what you haven't discovered yet.

| Point | Details | |---|---| | Algotorial hybrid system | Human editors build candidate pools; algorithms personalize the final selection for each listener. | | Taste profile is central | Every listen, skip, save, and search updates your profile, which drives all personalized recommendations. | | Hundreds of billions daily events | Spotify processes hundreds of billions of events daily to keep taste profiles current and responsive. | | Mostra balances multiple goals | The Mostra framework optimizes for user satisfaction, discovery, and emerging artist exposure simultaneously. | | Commercial signals exist but are filtered | Discovery Mode increases a song's recommendation probability, but engagement data overrides it if listeners skip. |

***

If you're an artist, this system is your stage

Understanding how Spotify recommends new music is one thing. Getting your music into that pipeline is another. Spotify's algorithm rewards tracks that generate genuine engagement, which means the quality of your pitch to playlist curators directly affects your streaming trajectory.

Playlist Pilot analyzes your track's audio characteristics, genre, and mood, then matches it to playlists curated by real humans who are actively looking for new music. The average curator response rate through the platform is 47%, and every pitch is personalized to show exactly why your song fits a specific playlist. No per-pitch fees, and you keep direct contact with curators for future submissions.

If you want your music to reach the listeners Spotify's algorithm is already primed to send it to, AI-powered playlist matching is the most direct path from release to discovery.