An AI cover of Papaoutai crossed 80 million streams on Spotify in three months. No singer lent their voice. SACEM (the French collection society) validated the operation as a legal cover. Meanwhile, Deezer announced that 44% of new daily uploads on its platform are AI-generated. This double signal forces indie artists to rethink three things: royalties, voice ownership, and fan data capture. Here’s the breakdown.
The facts, straight
In late 2025, a track titled Papaoutai (Afro Soul), credited to Loud Urban Choir, appeared on Spotify. It’s a cover of Stromae’s 2013 hit, but the voice isn’t human. It was generated via RVC, a voice-conversion technology trained on the timbre of Arsene Mukendi, a Congolese-Russian singer. The track is distributed by a Swedish label, Unjaps AB.
By March 2026, the Spotify counter passed 80 million cumulative streams. The operation reportedly generated between $240,000 and $400,000 in gross streaming revenue according to cross-referenced press estimates.
SACEM confirmed the legality: the melody and lyrics weren’t altered, so the original rights holders, Stromae and his co-writers, collect their author royalties. The dispute isn’t about copyright. It’s about neighboring rights, the ones that pay the performer, and about capturing vocal identity.
AI volume now exceeds every industry forecast
The Papaoutai case isn’t isolated. It’s the visible peak of an already quantified wave.
« 44% of all new daily uploads on Deezer are AI-generated. »
Deezer Newsroom, April 20, 2026 (translated from French)
In volume, that’s roughly 75,000 AI tracks ingested daily on Deezer alone, over 2 million per month. The growth curve is brutal: 10% in January 2025, 18% in April, 34% in November, 44% in April 2026. The 50% threshold will likely be hit in summer 2026.
Why it isn’t (yet) draining human artists’ wallets
AI listening share remains marginal
Deezer publishes a reassuring revenue stat: AI-generated music accounts for between 1% and 3% of total streams on the platform. AI tracks are massively uploaded but barely played.
AI streams are mostly fraudulent
« Around 85% of streams detected on AI tracks are fraudulent and demonetized. »
Deezer Newsroom, April 20, 2026 (translated from French)
Translation: bots stream AI tracks to skim royalties. Deezer’s anti-fraud systems detect and cut payment. The royalty pool isn’t being diluted as severely as raw upload volume might suggest. But vigilance stays absolute, Spotify and Apple Music share far fewer numbers, and their anti-fraud policies aren’t public.
The real problem: a legal gap
The Papaoutai case isn’t an isolated incident. It’s a replicable, industrial-scale playbook. An operator armed with an open voice-conversion model can generate dozens of AI covers per day, push them through a consumer distributor, and collect author royalties without ever recording a vocal or playing a note. The legal friction sits between two rights regimes never designed for this scenario.
50 AI covers a day, zero notes played, royalties captured for life on a known artist’s name. The law hasn’t caught up.
The gap sits between copyright and neighboring rights. An AI cover that preserves melody and lyrics stays legal in SACEM’s eyes. But the synthetic voice, imitating an existing singer without consent, falls in a legal blind spot.
The proposed « presumption of exploitation » bill
Nineteen French collective management organizations, SACEM, ADAMI, SPEDIDAM, ADAGP, SCAM, SGDL and others, are backing a Senate bill on the « presumption of exploitation of cultural content by AI providers. » The text would reverse the burden of proof: the AI company would have to prove its model wasn’t trained on protected work, not the artist proving the opposite.
The timeline remains open. The bill is under Senate review in 2026.
Implications by profile
Indie artist (DIY and pro)
Priority: protect your vocal identity. Three actions to roll out now. Systematically deposit your masters and stems with the relevant collection societies (SACEM, SPEDIDAM in France, ASCAP/BMI in the US, PRS for Music in the UK). Activate Spotify for Artists and Apple Music for Artists alerts on your artist name to detect an AI clone. Monitor Shazam and TikTok for vocal sample usage.
Indie label
The focus shifts to artist contracts. New clauses to negotiate: explicit grant (or denial) of voice-cloning rights, mandatory platform disclosure on deepfake detection, revenue split if post-detection monetization occurs.
Publisher / distributor
The real work is metadata. Platforms now require an AI flag at delivery. Deezer is, for now, the only one explicitly labeling AI tracks in its public catalog. Spotify, Apple Music, and YouTube Music are working on their own signaling systems.
The band.stream angle: first-party fan data as the lifeline
In a market flooded with AI tracks and fraudulent streams, value isn’t created at the stream level anymore. It’s created in the fan relationship. An artist who knows their 500 real fans by email, territory, and listening behavior holds an asset no generative model can replicate.
Centralizing smartlinks on a proprietary URL, rather than a direct DSP link, lets you capture that data before the DSP absorbs it. Solutions like Linkfire, Feature.fm, ToneDen, or band.stream offer this layer. The differentiator plays out on three fronts: price (€5/month at band.stream vs €25-46 at legacy competitors), fan data sovereignty (CSV-exportable email base, GDPR article 20 portability right, IONOS Germany hosting, French CNIL jurisdiction), and native Meta Pixel + GA4 tracking.
Smartlinks & music marketing, no fluff.
Centralize smartlinks, ad campaigns, and analytics in a single tool. €5/month, ready in 30 seconds, data hosted in the EU.
Request access →
Key takeaways
The Papaoutai AI case isn’t isolated. It’s the symptom of a structural shift. Law is catching up. Platforms are trying to label. Artists, meanwhile, have to anticipate: vocal protection, clean metadata, first-party fan capture. The 80 million stream counter is a reminder: attention doesn’t disappear, it migrates. Who will convert it into durable revenue is still open.
