New data from music marketing platform intellijend reveals that while 106,000 tracks are uploaded to streaming services daily, only a fraction gain traction, and AI-generated content now accounts for 44% of uploads but just 1% to 3% of streams. The platform, which works with over 6,000 artists including Milky Chance, Ben Böhmer and BLOND:ISH, processes more than $1 million in monthly marketing spend and recently launched rosterOS for labels, managers and agencies.
Streaming’s Supply Glut
Approximately 106,000 new tracks are added to streaming platforms each day, pushing the total catalog to roughly 253 million songs by the end of 2025. Last year, 88% of those tracks were streamed fewer than 1,000 times, and 120 million tracks, nearly half, received 10 plays or fewer. At the top end, exactly 29 tracks surpassed a billion streams in 2025, two fewer than the previous year.
AI’s Growing Footprint
On Deezer, around 75,000 daily uploads, or 44%, are now fully generated by AI, amounting to more than two million machine-made songs per month. In April 2026, an AI-generated track reached number one on iTunes in five countries simultaneously. When Deezer tested listeners’ ability to distinguish AI from human-made music, 97% could not tell the difference. Despite this upload volume, AI-generated music represents only 1% to 3% of actual streams, and 85% of those streams are flagged as fraudulent, according to intellijend.
The Spotify Popularity Score and Velocity
Spotify assigns every track and artist a hidden popularity score from 0 to 100, accessible via its developer API (application programming interface) but not visible in Spotify for Artists. Intellijend analyzed a year of data from over a thousand artists and found that velocity, the speed at which streams accumulate, is the primary driver of the score, not total stream counts.
A concentrated burst of genuine streams in a short window can spike the score, and crossing thresholds in increments of 10, such as moving from 29 to 30, significantly expands a track’s placement in algorithmic playlists like Discover Weekly and Release Radar. Each track-level spike also lifts the artist score, which compounds over time.
Meta Ads: Creative and Testing
Generating velocity requires real listeners quickly. The platform identifies Meta Ads as the most reliable method to attract targeted audiences on demand, with campaigns aimed at specific genres, countries, and age groups. Users do not open Instagram expecting advertisements, so overproduced, glossy creative is often skipped. Shaky, native phone footage that feels human and blends into the feed is more effective. The visual should match the audio: techno music belongs in a rave setting, not a beach. Leading with the strongest part of the song is critical.
Effective ad copy formats include:
- RIYL (Recommended If You Like) references
- “For fans of [a comparable major artist]”
- “A song for overthinkers”
- A friend’s genuine reaction
- B-roll footage matching the mood with lyrics overlaid
Testing is essential. Intellijend recommends running multiple creatives, such as ten ad variations against four audience segments, totaling 40 combinations. Underperforming ads should be cut quickly, and winning ads duplicated rather than having their budget sharply increased, as Facebook’s algorithm can treat a sudden budget hike as a new ad and destabilize performance.
Cost per play benchmarks: under approximately $0.30 is considered good, under $0.20 is “evergreen” and can be sustained for months. Each campaign yields audience data, making subsequent efforts more efficient and less expensive.
Budget Benchmarks and Long-Term Value
A release budget of roughly $300 is a minimum floor, $600 is a solid target, and $1,000 or more is excellent. The platform emphasizes that this spend is a marketing investment, not merely a purchase of stream counts, and that the audience insights gained reduce costs over time.
