More music is being created than at any other time in human history. More than 100,000 songs a day are submitted to streaming services.
And we are listening to less and less of it.
The numbers
Luminate, the entertainment data company that has tracked music streaming since 2008, found that in the first half of 2026, almost two thirds — about 60 per cent — of music played in both the UK and the US was released five or more years ago. The industry calls this “deep catalogue”.
In the UK, 18 of the year’s top 20 bestselling songs were released twelve months ago or earlier. The year’s eighth biggest hit is Zara Larsson’s Lush Life, which came out in June 2015. Michael Jackson’s Billie Jean (1983) sits at 22. Fleetwood Mac’s Dreams (1977) is at 30.
In the US the pattern holds in the two most-played genres: about 60 per cent of R&B and hip-hop streams are catalogue.
It is not an AI flood
The obvious explanation is wrong, and it is worth saying so clearly. Spotify says it has removed 75 million “spammy tracks” from its catalogue, and Deezer reports that AI-generated music accounts for just 0.5 per cent of streams on its platform. Real music made by people is still almost all of what gets played.
The problem for new artists is not that machines are crowding them out. It is that listeners are choosing the familiar.
Winners take even more
When a new song does break through, it stays. Sam Fender and Olivia Dean’s duet Rein Me In has spent a record 23 non-consecutive weeks at UK number one. In the US, Ella Langley’s Choosin’ Texas has become the longest-running US number one in chart history.
Reaching the top now requires millions of streams a day. The UK’s Competition and Markets Authority found that nearly two thirds of streams go to music recorded by only the top 0.4 per cent of artists.
Going viral is not the same as building a career. “It’s easier [today] for folks to be passive fans,” said J Erving, founder of the distribution company Human Re Sources, in a recent interview with Billboard. “For you to consider yourself broken [as a mainstream artist], people need to care about you beyond the song. Where is the connectivity?” A platform like TikTok is built to create a moment around a single track, not to carry an artist.
The playlist problem
Streaming services now let subscribers generate playlists from a text prompt. Musicologist Anna-Kaisa Kaila thinks “there absolutely is potential for bias” in that. “With AI systems, the bias can either come from the data or from the system itself. If there’s not enough training data in a particular style that is perhaps not so commercially dominant, it will be that much harder for the system to pull up tracks in that style.”
Which is the mechanism, stated plainly: under-represented music gets recommended less, so it accumulates less data, so it gets recommended less.
“They’re based on everyone’s listening behaviour and data, right?” said London DJ and curator Tim Garcia. “So, you’re probably going to get given a very similar recommendation to an artist you’re already listening to.” Dan Mackta, managing director of the human-curated platform Qobuz, put the same point in one line: the technology “is never going to want to throw a curve ball at you”.
What the industry is trying
Oliver Schusser, head of Apple Music, says the service is working to champion “tomorrow’s artists” in its curation. “You know, we do the Super Bowl halftime show, we need to make sure that we still have performers in 10, 20, 30 years,” he told the BBC. “Because if you look back at the performers we had in the last four years, they all started somewhere.”
Sofar Sounds, which puts on small pop-up gigs worldwide, is launching a Human Discovery Movement campaign that sends voice-note recommendations from human curators to anyone who texts a number. East London rapper J Casa, who played its Shoreditch fringe festival this month, argues a live room builds “a lasting connection” that a feed cannot.
Why this is familiar here
Anyone releasing Bangla music on Spotify or YouTube Music is operating inside the same machinery, with one difference that cuts deeper: Bangla is exactly the kind of catalogue Kaila describes as thin on training data. The pull towards the proven hit is not a national problem and not a taste problem. It is what a recommendation system does when it is rewarded for being right rather than for being surprising.
Related: the clippers who turn other people’s work into reach.
