LANDR says Spotify-flagged fake streams fell 95% as distributors become the front line on fraud

LANDR, the Montreal distributor and mastering company, said on 23 September that artificial streams flagged by Spotify across its distributed catalogue fell 95% between Q1 2025 and Q2 2026. It is now pushing its fraud checks upstream, so suspicious releases are caught before they reach streaming services rather than after the streams have been counted. Zinstrel called distributors the new choke point for fraud enforcement. That's the right frame, and it's why this matters more than one company's metric.

What is LANDR actually doing?

The system was built with Mila, the Quebec AI Institute, during a six-month internship, and LANDR says it went live in February 2026. It runs two machine-learning models. One scores accounts on profile and behavioural data. The other looks for listening patterns that don't look like a real audience. Mila's example is a single song looping around the clock on one device.

The expansion adds risk signals at the point of upload, per LANDR:

• unusually high release frequency
• catalogues that appear overnight
• clusters of accounts sharing devices or payout details
• metadata that impersonates established artists

Audio gets checked too. ACRCloud fingerprints recordings against commercial catalogues, and IRCAM technology looks for altered copies (sped up, pitch-shifted or re-recorded) and for acoustic markers of synthetic generation. Accounts that release often face identity verification. Artists declare AI-generated content when they submit, and LANDR passes that declaration on to platforms in the metadata. LANDR says its anti-fraud team reviews flagged activity before taking any action. Confirmed fraud accounts can be closed and blocked.

How much should you read into 95%?

Less than the headline suggests. LANDR did not disclose how many streams were flagged at either end of the period, so a 95% drop could be large or small in absolute terms. It credits multiple fraud-prevention measures, not the Mila models alone, and the measurement window opens about a year before those models went live. It also counts only what Spotify flagged. No other service is named.

There's also a gap in what the number can measure. It counts fraud that got through and was caught later. It can't count legitimate releases that were wrongly held back at upload. Tighter filtering at the door moves this number in the right direction whether or not honest artists get caught too.

Why distributors, and why now?

The money explains most of it. Since 2024, Spotify has charged labels and distributors per track when it detects flagrant artificial streaming on their content. Spotify's artist support page says it sends monthly reports to distributors, who can warn, take down content or suspend accounts. Mila's write-up says Spotify and other major platforms hold distributors responsible for fraud rates, and that LANDR faced the risk of a permanent ban before its rates fell below the platforms' thresholds.

Industry policy now points the same way. On 14 September, IFPI launched the Streaming Integrity Initiative, a set of baseline practices for distributors in five areas: verify customers and rights, vet uploads, act against offenders, share fraud signals across platforms, and measure results. The 27 launch signatories include the three majors, CD Baby, Ditto Music, RouteNote, Symphonic and Too Lost. LANDR says its expansion fits these priorities, but it isn't on the launch list.

AI raises the stakes. Deezer said in April that close to 75,000 AI-generated tracks reach it every day, about 44% of daily uploads, and that up to 85% of streams on those tracks in 2025 were fraudulent.

What it means for artists

LANDR CEO Pascal Pilon's argument is the one artists should hold onto: fake streams take money from the same royalty pool that pays working musicians. Fraud isn't victimless. Every bot-farmed stream dilutes everyone else's per-stream value, so honest independent artists have the most to gain from fraud being stopped before it is paid.

The risk is in the signals themselves. Read as fraud markers, several of them also describe normal independent careers. A beatmaker who drops a track every week has high release frequency. A producer who uploads a back catalogue of 200 tracks in one go has a catalogue that appeared overnight. A small collective sharing a laptop, or a manager collecting royalties for several acts, can look like a cluster of accounts sharing devices or payout details. Zinstrel makes a similar point about legitimate artists who release AI music at volume. That's our reading of the risk, not a reported case. But it explains why the review process and the appeal path matter as much as the model.

The downstream system already shows what opaque enforcement feels like. One artist posted on X that they'd received nothing from Spotify through DistroKid despite around 70,000 streams, and that support was slow to respond. We can't verify what happened to that account, and the post doesn't say fraud was the cause. But the experience is common: the artist sees streams but no money, and gets no clear explanation. Spotify's own advice to artists who think they were wrongly flagged is to send their distributor details of how they promoted the music. In practice, that leaves artists dependent on how responsive their distributor is.

What we think

Screening before release beats clawing money back after it, and that's good for artists. A release that's stopped at upload doesn't result in withheld royalties, a fine passed on to the distributor or a wiped stream count months later. On that basis, LANDR's direction is right, and putting a human review before any action is the correct design.

It isn't enough, though, and the industry framework has a blind spot. IFPI's "Measure" pillar asks distributors to assess whether their anti-fraud work is effective. Measured that way, success means less fraud. Nobody is yet asked to publish how many legitimate artists were wrongly flagged, how long reviews take, or how many appeals succeed. A distributor that reports a fraud reduction should report its false-positive rate next to it. Artists who are flagged should be told which signal fired and get a named route to contest it.

There's also a likely shift in the market, and it's our inference rather than anything LANDR claimed. Machine-learning screening, audio fingerprinting and identity checks cost money, and Spotify's per-track charges fall on distributors whether or not they can afford the tooling. That favours larger distributors that can build or license detection, and squeezes cheap, low-friction upload models that compete on volume. Pseudonymous artists pay a quieter price, since identity checks for frequent releasers end some of the anonymity many electronic and bedroom producers rely on. For artists choosing a distributor in 2027, the fraud policy, and above all how it handles appeals, deserves as much attention as the fee and the payout split.