The editorial argues that streaming royalties are a fixed pie, and AI-generated tracks that accumulate streams directly siphon money from human artists. With near-zero marginal production costs and tools like Suno enabling hundreds of tracks per day, the economic dilution is not theoretical but mathematical certainty.
Folgueira has been one of the most vocal industry leaders calling for AI-generated content to be clearly labeled. He has proposed that such content be potentially excluded from the standard royalty pool, separating it from human-created music to protect artist compensation.
The editorial notes that Deezer's detection relies on audio fingerprinting and ML classifiers, but generative music tools improve faster than classifiers can keep up. The actual percentage of AI-generated content is almost certainly higher than 44%, as tools like Suno and Udio crossed the 'good enough to fool casual listeners' threshold in late 2025.
The editorial highlights that a single person with a $20/month AI subscription can generate hundreds of tracks daily and distribute them for pennies through services like DistroKid or TuneCore. The streaming model was designed around human production constraints that no longer exist, enabling playlist stuffing and volume-based revenue capture at unprecedented scale.
Deezer, the French music streaming platform with roughly 10 million paying subscribers, disclosed that 44% of songs uploaded to its platform daily are now AI-generated. The company has been tracking the trend for months using a combination of audio fingerprinting, metadata analysis, and machine learning classifiers trained to detect synthetic audio signatures.
The number is staggering but, if anything, conservative. Deezer's detection catches what it can identify — the actual percentage of AI-generated content making it onto the platform is almost certainly higher. Detection is an arms race, and generative music tools improve faster than classifiers can keep up. Suno, Udio, and dozens of smaller tools can now produce radio-quality tracks in seconds, and the output quality crossed the "good enough to fool casual listeners" threshold sometime in late 2025.
Deezer CEO Jeronimo Folgueira has been one of the more vocal industry leaders on this issue, previously calling for transparency requirements and proposing that AI-generated content be labeled and potentially excluded from the standard royalty pool.
The streaming royalty model is a fixed pie. Every platform — Spotify, Apple Music, Deezer, Tidal — uses some variant of pro-rata or user-centric payment, where total subscription revenue is divided among tracks based on their share of total streams. When AI-generated tracks absorb streams, they directly dilute payments to human artists — it's not a theoretical concern, it's arithmetic.
Consider the economics. A single person with a $20/month Suno subscription can generate hundreds of tracks per day and distribute them through services like DistroKid or TuneCore for pennies per track. If even a fraction of those tracks accumulate streams — through playlist stuffing, algorithmic recommendation, or sheer volume — they siphon money from the shared pool. The marginal cost of production has collapsed to near zero, but the royalty pool hasn't grown to compensate.
This mirrors what happened to web content after GPT-3.5 went mainstream. Google saw an explosion of AI-generated SEO content that degraded search quality. Stack Overflow saw AI-generated answers that were plausible but wrong. Reddit saw AI-generated posts farming karma. The pattern is consistent: any platform with a shared resource (search ranking, reputation, royalty pool) and low barriers to contribution will be flooded once generation costs hit zero.
The music industry's response has been predictably fragmented. Major labels want outright bans on AI training with their catalogs. Independent artists are split — some see AI tools as democratizing, others as existential threats. Platforms are caught between wanting to grow their catalogs (more content = more engagement) and protecting the ecosystem that makes their product valuable.
If you're building any UGC platform — not just music, but video, text, images, code — Deezer's 44% number is your canary in the coal mine. The questions you need to answer now:
Content provenance at ingest. The C2PA standard (Coalition for Content Provenance and Authenticity) provides cryptographic content credentials that can travel with files. It's not bulletproof — metadata can be stripped — but it raises the bar. If your upload pipeline doesn't include provenance verification today, you're accepting that your platform's quality will degrade at the rate generative AI improves.
Economic incentive design. Pro-rata payment models are particularly vulnerable because they create a tragedy of the commons. If your platform has any shared resource allocation (ad revenue splits, visibility algorithms, ranking systems), model what happens when 44% of inputs are zero-cost generated content. Then model 70%, because that's where this is heading.
Detection as ongoing ops, not a one-time feature. Audio classifiers, text detectors, image forensics — these aren't ship-once features. They're adversarial systems that need continuous retraining. Budget for this like you budget for security: it's an ongoing operational cost, not a project with a finish line.
Deezer's approach of separating AI-generated content into a distinct tier — still available but not competing in the same royalty pool — is one architectural pattern worth studying. It doesn't solve detection perfectly, but it changes the incentive: if AI tracks earn $0.00001 per stream instead of $0.004, the economic motivation for flooding evaporates.
The 44% figure will look quaint within a year. Generation tools are getting better, cheaper, and more accessible every quarter. The platforms that survive this transition will be the ones that solve two problems simultaneously: detection (knowing what's AI-generated) and economics (ensuring the answer to "what's AI-generated" actually changes how value flows). Deezer is at least naming the problem publicly. Most platforms are still pretending it isn't happening to their content pools. It is.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.