The Certified Human: Who Actually Labels AI Music

Overview

Streaming services are being buried. Deezer reported in April 2026 that it receives roughly 75,000 fully AI-generated tracks every day — about 44% of everything uploaded to the platform, and more than two million new tracks a month.

Almost nobody listens to them. AI-generated music accounts for between 1% and 3% of total streams, and Deezer detects 85% of those streams as fraudulent, generated to farm royalties rather than by anyone choosing to play a song.

The uncomfortable finding sits alongside it: research indicates 97% of listeners cannot distinguish fully AI-generated music from human-made music. The flood is not being rejected on quality. It is simply not being found.

No streaming service has introduced a 'certified human' badge. The industry moved in the opposite direction — labelling AI involvement through a standardised credits system rather than certifying its absence, which is a meaningfully different approach with different consequences.

How 10,000 Tracks a Day Became 75,000

Deezer built an AI-music detection tool and began publishing what it found, which makes it the only reliable public window into the scale of this. The trajectory it recorded is close to vertical.

In January 2025, when detection launched, roughly 10,000 AI-generated tracks arrived per day. By September that had reached 30,000, by November 50,000, by January 2026 around 60,000, and by April 2026 approximately 75,000 — a sevenfold increase in fifteen months.

The proportion tells the same story from the other direction. AI went from a noticeable minority of uploads to 44% of everything new arriving on the platform. On current trends, the majority of music uploaded to streaming services will shortly be machine-generated.

Why 44% of Uploads Is Only 3% of Listening

The gap between those two numbers is the most important fact in this entire subject, and it is routinely omitted from coverage announcing that AI has taken over music.

Forty-four percent of supply converts to somewhere between 1% and 3% of demand. Uploading is essentially free and infinitely scalable; attention is neither. Generating a track costs a fraction of a cent and takes seconds, so the marginal cost of flooding a catalogue approaches zero, while the number of hours humans spend listening has not changed at all.

What the upload statistic measures is the collapsed cost of production, not a shift in taste. Music catalogues are being filled with tracks that were never intended to be heard by a person, and treating that as evidence of AI's popularity confuses the size of a haystack with demand for hay.

The 85% Nobody Talks About

Of the small share of streams AI tracks do receive, Deezer detects 85% as fraudulent and demonetises them accordingly.

This reframes the phenomenon entirely. The dominant use of generative music on streaming platforms is not artistic and not even really commercial in a conventional sense — it is royalty fraud. Upload enormous volumes of plausible-sounding audio, drive artificial streams through bot networks, and collect fractional payouts at scale.

That is why upload volume rises so much faster than listening. The business model does not require anyone to enjoy the music, only for the tracks to exist and appear to be played. It also explains the platform response: Spotify deleted more than 75 million tracks it classed as spam, an action aimed at fraud rather than at artistic competition.

97% of Listeners Can't Tell the Difference

A widely reported study found that 97% of listeners could not distinguish fully AI-generated music from human-made music in blind comparison.

This demolishes the comfortable assumption underlying most commentary on this subject — that audiences possess some reliable instinct for authenticity, that synthetic music carries an audible absence of feeling. Under blind conditions, essentially nobody demonstrates that ability.

It also explains why provenance became a labelling problem rather than a quality problem. If listeners could hear the difference, disclosure would be unnecessary and the market would sort itself. Because they cannot, the only way anyone knows what they are hearing is if someone tells them — which is precisely why the industry's response has centred on metadata rather than on detection at playback.

There Is No Certified Human Badge

The intuitive solution — certify human-made work and let it command a premium — is not what the industry built. Spotify instead adopted DDEX, an existing industry metadata standard, extended to carry structured AI disclosures within track credits.

The distinction matters. Certifying humanity requires proving a negative across an entire creative process and would place the burden on every artist. Disclosing AI places the burden on whoever used it, and integrates into systems labels and distributors already operate.

When an AI Act Topped a Billboard Chart

The symbolic threshold was crossed in late 2025, and by more than one act. These were real chart placements, and their scale is frequently overstated.

Why Chart Placement Was Easier Than It Sounds

A number one is a number one, but the specific chart matters enormously and the significance of Breaking Rust's placement was questioned immediately by people who understand how the rankings work.

Country Digital Song Sales measures paid downloads within a single genre — a narrow, low-volume metric in an era when almost nobody buys individual tracks. The number of purchases required to top it is a small fraction of what a placement on the Hot 100 would demand, which makes it unusually accessible to a coordinated push.

The Velvet Sundown is the more genuinely interesting case, because 1.4 million monthly listeners represents real passive listening rather than purchases. Those listeners largely did not know what they were hearing, which is a demand-side result rather than a chart-mechanics artefact — and it is the closest thing to evidence that synthetic music can hold an audience on its own.

Who Settled, and Who Got Nothing

The legal reckoning over the training data behind these tools resolved along predictable lines. Universal Music Group settled with Udio in October 2025. Warner Music Group settled with Udio and with Suno in November 2025. As of mid-2026, Sony Music had settled with neither.

The settlements converted infringement claims into licensing arrangements, which means the major labels are now commercial partners of the platforms they were suing. It is a rational outcome for both sides and a striking one: build a valuation on unlicensed catalogue, then buy retrospective permission from the rights holders large enough to force the issue.

Independent artists received nothing. They lacked the standing and resources to bring comparable actions, their work was used in the same training corpora, and no settlement reached them. The music that taught these systems came from everyone; the compensation went to three companies.

What Provenance Can and Can't Fix

Labelling is a genuine improvement over the previous situation, where a listener had no way of knowing and no one was obliged to say. Granular credits let anyone who cares make an informed choice, and prohibiting unauthorised voice clones addresses the most direct harm to living artists.

The limits are structural. Self-reported disclosure depends on the honesty of uploaders who are, in 85% of streaming cases, already committing fraud. Nobody running a bot-driven royalty operation is going to accurately tag their catalogue, and a labelling regime that works only for compliant participants does little about the actual problem.

The more durable defence is the one already visible in the numbers. Forty-four percent of uploads capturing 3% of listening suggests discovery, curation and human attention remain scarce in a way that generation no longer is. Provenance data helps people who want to choose. What protects human musicians is that being heard was always the hard part.

Conclusion

The story is not that AI music has conquered streaming. It is that generating music became free while listening to it stayed finite, and the resulting flood — 75,000 tracks a day, 44% of all uploads — is mostly aimed at royalty systems rather than at audiences.

The genuinely unsettling finding is not the volume but the indistinguishability. With 97% of listeners unable to tell the difference in blind comparison, the comforting idea that human-made music carries some audible signature of its origin does not survive testing. Provenance became a metadata problem precisely because it stopped being an audible one.

The industry's answer is to label AI rather than certify humanity, which is the more practical design and depends entirely on self-reporting by the least trustworthy participants. Meanwhile the labels settled, the platforms kept their valuations, and the independent artists whose work trained these systems were left out of every arrangement made on their behalf.

Data Source and Attribution

Deezer NewsroomMusic Business WorldwideBillboard

Upload volumes, AI share of streams and fraud detection rates come from Deezer's published newsroom data and its AI-music detection reporting. Platform policy details, the DDEX disclosure standard and track removal figures come from Spotify's announced AI policies and subsequent reporting by TechCrunch and Music Business Worldwide. Chart placements are as reported by Billboard, and litigation status reflects publicly reported settlements as of mid-2026.

FactsFigs reviews, cleans, and cross-checks every source dataset before shaping it into a data story. Each visualization is created and designed in FactsFigs Design Studio — an internal tool developed and owned by FactsFigs — and is the original work of a FactsFigs author, not an AI-generated copy of any existing graphic. Individual assets within a visual may or may not be produced with AI tools, but the design of the visual itself is solely FactsFigs' own.

Figures are estimates at the time of publication, provided for information only — nothing here is legal or financial advice.

2026-07-20