44% of New Songs on Deezer Are AI-Generated. Streaming Already Fell.

Open your streaming app, hit shuffle, and roll the dice. According to new internal numbers from French streaming service Deezer, 44% of the songs uploaded to its platform every day are generated by AI. A year ago, that figure sat in the low teens. The tipping point wasn’t supposed to arrive this fast.

Tens of Thousands of Synthetic Tracks Per Day

Deezer sees roughly 150,000 new tracks uploaded daily. Do the math: we’re talking about tens of thousands of AI-generated songs pouring in every 24 hours, most of which no human will ever deliberately queue up. The number surfaced because Deezer began deploying its own AI-detection system last year, specifically tuned to flag outputs from models like Suno and Udio.

CEO Alexis Lanternier has been consistent on where he draws the line: the problem isn’t AI music per se, it’s fraudulent streaming — tracks that nobody actually listens to siphoning royalties from the pool. When the upload pipe is flooded with machine-made filler, the economics of the whole system start to bend.

Why the Curve Went Vertical

The answer is boring but decisive: the barrier to entry collapsed. Plug a text prompt into Suno v4 or Udio, wait ten seconds, get a finished track. For roughly $10 a month, a single user can generate hundreds of songs a day. No studio, no session musicians, no mastering engineer.

Layer on the grift. Upload thousands of AI tracks, point a botnet at them, and even at $0.003 per stream the checks start to add up. Last year, U.S. prosecutors charged a man with generating hundreds of thousands of AI songs and botting his way to $10 million in fraudulent royalties from Spotify, Apple Music, and Amazon Music. He wasn’t a lone eccentric — he was a template.

The Platforms Are Responding. Sort Of.

Deezer now flags detected AI tracks, pulls them out of algorithmic recommendations, and excludes them from royalty distribution. Spotify rolled out an AI transparency policy last fall. Apple Music has started pushing labels to tag AI involvement in track metadata.

The skepticism is earned. Detection and generation are locked in a cat-and-mouse loop — every refinement in model quality, or even a light human touch-up pass, drops detection accuracy sharply. And plenty of listeners shrug: if the song slaps, why should the origin matter? That’s a harder argument to defeat than any watermarking scheme.

The Squeeze on Working Musicians

Indie artists are the ones actually bleeding. Streaming payouts were already thin. Now an effectively infinite supply of machine-generated tracks is competing for the same playlist real estate and algorithmic attention. The IFPI has projected that, on current trends, AI-generated music could erode up to 24% of professional musician income by 2028.

The picture isn’t uniformly grim. A growing share of working creators use AI tools for demos, arrangement sketches, and background beds — it’s quietly becoming standard in the production stack. The fault line isn’t “AI assistance” vs. “purity.” It’s assistance versus industrial-scale finished-product generation aimed squarely at the royalty pool.

What You’re Actually Listening To

Deezer’s 44% isn’t just a stat. It’s a signal that the economic structure underneath recorded music is cracking. Platforms will keep tightening detection. Regulators in the EU and U.S. are already debating mandatory AI-content labeling. But the more interesting question is the one every listener can ask tonight: of the last ten songs in your queue, how many were written by a person who meant them?

Would your favorite track feel different if you learned a model wrote it in six seconds? Or, in the end, is the song the only thing that matters — and the human behind it just nostalgia?