How Spotify's Algorithm Recommends Music: Surfaces, Signals, and What Artists Control

A matching system, not a reward system

The most useful reframe for understanding Spotify's recommendation engine is this: it is a matching system, not a reward for popularity. Its job is to find the right listener for a track at the right moment. It is not designed to promote music because it is popular, though popularity can result from good matches accumulating at scale.

Spotify for Artists describes personalized recommendations as drawing on "a multitude of signals" to connect the right song to the right ears at the right time. Those signals come from listener behavior, not from an artist's biographical facts, follower counts, or label affiliation.

Spotify does not publish a precise ranking formula or signal weights. What it does document, through Spotify for Artists, Spotify Newsroom, and its engineering output, is the direction of the inputs. This article covers what is publicly attributed.

The main personalized surfaces

Spotify's recommendation system operates across several surfaces, each with a different function.

Discover Weekly updates every Monday with tracks a listener has not heard before, chosen by the algorithm based on their listening history and taste profile. It uses collaborative filtering at scale: if two listeners share patterns and one engages with a track the other has not heard, the system treats that as evidence of potential fit. As Spotify Newsroom describes it, this logic operates across thousands and millions of connections simultaneously, updating continuously.

Release Radar updates every Friday with recent releases from artists a listener follows or has engaged with, plus music the algorithm thinks they would like. According to Spotify for Artists, a new release is eligible to appear in Release Radar for up to 28 days after it goes live. Pitching a track through Spotify for Artists at least seven days before release makes it automatically eligible for the playlist. A track can only appear in a given listener's Release Radar once at a time.

Radio and autoplay surfaces play when a listener finishes a playlist or selects a radio session based on a song or artist. The system generates an extended queue of music it considers contextually similar, drawing on audio features, genre proximity, and the behavioral patterns of listeners who engaged with the anchor track. Spotify for Artists describes Radio and Autoplay as a context where listeners are "open to discovery," which is part of why Discovery Mode, the paid promotional tool, operates primarily in this context.

Daily Mix, Mixes, and personalized editorial playlists occupy a middle ground between purely algorithmic and editorially curated. Spotify for Artists counts these under the "Made for You" discovery pathway alongside Discover Weekly, Radio, and Autoplay. Together, these personalized sessions account for 33 percent of all new artist discoveries on Spotify, per Spotify for Artists.

The signals that feed recommendations

Spotify does not publish a weighting table for its signals. What it attributes publicly, and what its recommendation architecture makes clear, is that listener behavior after a play is the primary input.

Saves are among the strongest positive signals. When a listener saves a track to their library or adds it to a personal playlist, it tells the system that person wants to return to that music. A high save rate, saves measured against streams, signals durable appeal beyond passive listening. Spotify for Artists lists saves alongside streams per listener and playlist adds as the engagement stats visible in the Source of Streams breakdown.

Completion rate is the share of a stream in which the listener plays the track all the way through. A high completion rate signals that the track holds attention. Skips, especially in the first 30 seconds, signal the opposite: a listener who bails early is telling the system the track did not match their intent in that moment. The system reads these behaviors in context, a skip in a lean-back Radio session carries different weight than a skip in an active exploration session, but the direction is consistent: completion supports reach, early abandonment limits it.

Repeat listens indicate depth of engagement. When a listener returns to a track multiple times, especially within a short window, streams per listener rises and the system gains stronger evidence that the match between track and listener is genuine. High streams per listener is a component of listener retention, the broader signal that a listener is becoming a real fan rather than a one-time passer-by.

Source of streams matters alongside volume. Spotify for Artists includes a Source of Streams breakdown in the Engagement tab, separating streams from active listener-driven sessions from streams from programmed or algorithmic sources. Music that is sought out directly, through an artist profile, a listener's own queue, or a fan-made playlist, generates strong active signals that reinforce algorithmic confidence.

How new releases enter the system

When a track goes live, the recommendation system has no historical data for it. It draws on what it knows about the artist's existing audience, the metadata submitted at pitching, and early listener behavior to begin forming a picture of where the track fits.

If early listeners engage strongly, high saves, high completion, return listens, the system treats that as evidence the track can be recommended reliably to more listeners with similar profiles. If early engagement is weak, confidence to expand reach does not build.

As Spotify for Artists notes: there is nothing specific an artist can do to guarantee placement on an algotorial playlist. The system makes continuous assessments about fit, not awards for hitting a benchmark.

Why artificial streaming actively backfires

Because the system reads engagement quality rather than raw volume, manufactured streams provide no benefit and carry significant penalties.

Spotify for Artists states this directly: when Spotify detects artificial streams, those streams do not earn royalties, do not count toward public stream numbers or charts, and do not positively influence recommendation algorithms. Spotify defines artificial streaming as streams that do not reflect genuine user listening intent, including any attempt to manipulate streaming counts through automated processes such as bots or scripts.

The practical consequence is that buying streams does not move an artist up in algorithmic surfaces because the system is looking for behavioral evidence that real listeners want the music. Manufactured plays produce no such evidence. They dilute the royalty pool, harm other artists, and as of April 2024 expose distributors and labels to direct per-track charges from Spotify when flagrant artificial streaming is detected, according to Spotify for Artists.

What artists can actually feed the system

The operator conclusion is simple: you cannot game the system, but you can give it honest data to work with.

Pitching through Spotify for Artists before release ensures editorial consideration and automatic Release Radar eligibility. Growing a genuine follower base means new releases reach listeners who already have evidence of interest, generating cleaner early engagement data.

Quality of early streams matters more than raw volume. Music that real fans encounter during a release window, through social posts, pre-save campaigns, or direct artist communication, generates authentic saves and completions the algorithm reads as reliable. Broad, untargeted promotion may produce streams without producing the engagement signals that expand reach.

Strong listener retention and streams per listener across a catalog signal that an artist's audience is genuinely engaged. That signal compounds as the catalog grows: a tight, consistent catalog is easier for the system to recommend confidently than a scattered one. The algorithm rewards the work of building a real audience, not the appearance of one.

FTSMusic analysis is based on anonymized aggregate artist data, internal campaign observations, and publicly available industry documentation. Individual outcomes vary by catalog, genre, audience quality, and release strategy.