When Michael Smith‘s lawyers argued last month that his AI-generated streaming scheme caused “tiny, immeasurable losses” spread across millions of songwriters, they were, without meaning to, describing the streaming fraud data correctly. Every stream on a subscription platform is paid out of a shared, pro-rata royalty pool, not a fixed per-play rate, and the actual streaming fraud data and research now on the record, from the platforms’ own disclosed numbers to the analysts who audit them for a living, shows that the harm is real, it is measurable in aggregate, and it always lands on somebody else’s stream. Here is what that data says, how the mechanism actually works, and why it bites hardest in the markets where the royalty pool is thinnest to start with.
The mechanism a defense argument accidentally explained#
Most subscription platforms pay out on a model researchers call pro-rata: each rightsholder’s share of a market’s royalty pool for a given period equals their share of that market’s total streams. A rightsholder’s payment is, in its simplest form, their own stream count divided by every stream counted in the pool, multiplied by the revenue available to distribute. Nobody is paid a fixed price per play. Spotify has told reporters no fixed per-stream rate exists at all, for exactly this reason.
That structure is what a fraudulent stream actually attacks. It does not steal a specific artist’s specific royalty. It adds a stream to the denominator of every legitimate artist’s fraction, without adding a real listener’s share of subscription revenue to the numerator on the other side. A 2025 paper accepted at the International Conference on Machine Learning, “Fraud-Proof Revenue Division on Subscription Platforms”, by researchers Abheek Ghosh, Tzeh Yuan Neoh, Nicholas Teh and Giannis Tyrovolas, opens with the Smith case itself as its motivating example: a musician who built “hundreds of thousands of songs created using AI” and “a complicated network of over a thousand bot accounts” to inflate streams across Amazon Music, Apple Music, Spotify and YouTube Music. The paper’s finding is not that pro-rata is merely exploitable. It is that the standard rule, which it labels GlobalProp, “not only fails to prevent fraud, but also makes detecting manipulation computationally intractable,” and it cites industry estimates that stream manipulation costs the business $300 million a year.
What the streaming fraud data and the analysts who study it actually show#
No two companies measure fraud the same way, and the disclosed figures do not agree with each other, which is itself a finding: the industry has no shared definition of what counts as a fraudulent stream.
- Deezer says fraud accounted for 8% of streams across its entire catalogue in 2025, and up to 85% of streams on fully AI-generated tracks specifically, depending on the month, up from about 70% a year earlier. Deezer chief executive Alexis Lanternier has said “every fraudulent stream that we detect is demonetized so that the royalties of human artists, songwriters and other rights owners are not affected,” and the company has since begun licensing its detection tool to other platforms.
- Apple Music vice president Oliver Schusser said in February that the service demonetized as many as 2 billion fraudulent streams in 2025, at a rate he put at “well below 1%” of total streams, around 0.3%. Apple doubled its penalty scale for distributors caught in manipulation, from a 5 to 25% fine to a 10 to 50% fine. “When we find fraud we remove the stream counts, we remove from the charts, and we take the money and put it back into the pool so that it goes to honorable artists,” Schusser said, calling streaming fraud “a zero-sum game.”
- France’s Centre national de la musique, a public body, ran an 18-month study with Deezer, Spotify, Qobuz and a panel of major labels and distributors and found between 1% and 3% of streams in France in 2021, as many as 3 billion streams, showed signs of manipulation. It stated plainly that this is a floor, not a ceiling: the study measured only fraud its participating platforms actually detected. Amazon Music, YouTube and Apple Music declined to share their data for the study.
- Beatdapp, an independent fraud-detection firm whose clients include Universal Music Group, the Mechanical Licensing Collective and Beatport, puts the number highest of all. Co-chief executive Morgan Hayduk has estimated that streaming fraud is now “probably more 5-10%, getting closer to 10” of global activity, that some mid-sized platforms run fraud rates as high as 25 to 30% in a given month, and that a handful of “obscure distributors” have posted fraud rates “in the 99% range.” Beatdapp’s own estimate is that the practice diverts roughly $2 billion a year in royalties that would otherwise reach real artists.
The spread between “well below 1%” and “closer to 10%” is not noise. It reflects different definitions of fraud, different measurement points (whole catalogue versus published charts), and, bluntly, different incentives: a platform disclosing its own number has reason to report a low one, while a vendor selling fraud detection has reason to report a large addressable problem. Both can be true at once, and the honest reading of the streaming fraud data available right now is that nobody outside the platforms and their vendors can independently verify either end of that range.
Why the same fraudulent stream cuts deeper in a smaller pool#
Every one of the figures above comes from the United States, Western Europe or a global blend. None of Deezer, Apple Music, Spotify or the CNM has published a fraud rate specific to Nigeria, South Africa, Kenya or any other African market, and this desk found none in the course of this reporting. That gap is itself worth stating plainly, because the mechanism above does not require a published African number to matter to African artists.
A pro-rata pool divides a fixed amount of revenue by however many streams are counted against it. Where that pool is large, one bad actor’s fraudulent volume is a rounding error. Where it is small, the same absolute volume of fraud is a much larger share of the denominator, and depresses everyone else’s payout by proportionally more. This desk’s own reporting has priced exactly how thin some of these pools are in practice: at Duetti’s published global blended rate, Nigeria’s entire Spotify home chart paid African-origin artists an estimated $36,395.16 combined on 22 September, and South Africa’s paid $2,201.57. A volume of fraudulent streams that would be invisible against a market the size of the United States is not invisible against a pool that size.
The clearest evidence that this risk is taken seriously inside the industry is Audiomack‘s own response. Audiomack, which reports more than 40 million monthly users and carries some of the largest Afrobeats and amapiano catalogues on any platform, brought in Beatdapp in June 2025 specifically to analyze its streaming activity, strip fraudulent plays from royalty reporting before payout, and inform which tracks make its internal and external charts. “We take our role in providing accurate royalty reporting very seriously for the benefit of every creator who earns income from our platform,” Audiomack chief executive Dave Macli said of the partnership. Hayduk framed the stakes in general terms that apply directly to a platform built on African audiences: “No one notices that a few pennies are going to this song and a few pennies are going to that song but, in aggregate, they can steal billions of dollars.” It is exactly the argument Smith’s own lawyers tried to invert in his sentencing filing, that no single loss is perceptible, made instead by the company building the detection.
What this means for artists#
An artist or label reading the streaming fraud data correctly should take three things from it. First, a stream is not a fixed unit of income; it is a claim on a shared pool, and anything that inflates the pool’s total stream count without adding real listening dilutes every legitimate claim on it, including your own. Second, buying manipulated streams, from a booster tool, a click farm or a “guaranteed plays” service, is not a victimless shortcut. It draws down the same pool every honest artist on the same platform is paid from, which is why platforms increasingly treat it as an offense against the whole roster, not a private matter between a buyer and a seller. Third, check whether your distributor or DSP discloses how it handles detected fraud, whether flagged streams are excluded before your statement is calculated, and what recourse exists if a bad actor’s activity depresses a chart or a payout you were counting on. None of the four organizations cited here publish a market-by-market fraud rate, which means an African artist currently has no way to verify how much of this problem is landing in their own pool specifically. That absence of local disclosure is, on the evidence gathered for this piece, the single clearest gap in the public record.
Michael Smith’s sentencing, delayed since July with no new date on the public docket, will be the first US test of how a court treats AI-assisted streaming fraud. The data above suggests the question his lawyers posed, whether anyone was actually harmed, has already been answered by the platforms policing the same pool he drew from. The harder question, which platform will be the first to publish a number for an African market, remains open.
