YouTube Music starts labeling AI tracks — and immediately walks into the problem Deezer just hit

Updated:

YouTube Music confirmed on October 14, 2026, in a brief blog post, that it will begin labeling tracks "created primarily using generative AI," with the rollout starting that same day. Out of the gate, the platform is focusing on the easy cases: fully machine-generated audio, and outputs "using a generative model to create the composition and vocals" (Digital Music News's coverage). The label itself is an "AI" badge with an info icon below the song title on the Now Playing screen; tapping it opens a "How this was made" bottom sheet reading "Made with AI: This audio was altered or fully generated using AI," 9to5Google's Abner Li reports. YouTube says it will use its own detection tools and is "teaming up with music partners" to identify tracks, with the labels "expand[ing] to the main YouTube app" down the line — no timetable given.

The interesting part of the announcement is not the badge. It's the phrase "created primarily" — and the fact that YouTube, alone among the major DSPs so far, chose to plant its flag directly on the boundary the whole industry has been avoiding.

Two DSPs, two thresholds, one unresolved problem

Deezer rolled out its own version of this the day before — an "OnOff toggle" letting listeners filter out "fully AI-generated music," backed by an in-house study finding "nearly half of music streaming users would like to filter out fully AI-generated music" (per Digital Music News). Note the careful wording: "fully." Deezer's threshold is binary — 100% machine-made gets filtered, everything else stays.

YouTube's wording is "created primarily." That is a categorically harder standard, and YouTube Music itself acknowledges the difficulty, noting that longer term, "assuming AI adoption continues growing, determining what was 'created primarily' with generative models could prove challenging."

This is the DSP equivalent of the US Supreme Court's "I know it when I see it" problem, except applied to a spectrum rather than a category. A track where the model generated the composition and vocals is easy. A track where the model generated a demo and a human artist rewrote the bridge is not. A track where a human wrote the song and the model replaced the vocals with an AI voice — a practice already at the center of rightsholder disputes — sits somewhere in the middle and will get tagged under YouTube's rollout. A track where AI was used only for stem separation or mastering sits on the other side and won't. Every line the DSPs draw between "AI" and "human" has a plausibly legitimate work on both sides of it, and every artist whose work lands on the wrong side has a grievance.

The platforms' own detection choices underline how unsettled this is. YouTube says it will combine "our own detection tools" with "teaming up with music partners" — which is to say, a mix of automated audio fingerprinting for the fully-generated cases and, presumably, rightsholder reports for the rest. Deezer leans on its own detection tech plus distributor-side tagging. Neither approach can see inside a song to tell who wrote the bridge; both can, at best, catch the tracks that are entirely machine-made and handle disputes about the rest after the fact.

The listener-demand side

The demand driving all this is not speculative. YouTube attributed the labels directly to user feedback: "our listeners have told us they want transparency when they're hearing an AI-generated track" (the platform's blog post, per Digital Music News). On Reddit's YouTube Music community, threads asking the platform to "enable controls to block AI music" and complaining that "AI is annoying" have been running for months. Deezer's study found that nearly half of streaming users want a filter — a number big enough that the transparency question has effectively become table stakes.

A filter, though, is a different product from a label, and the gap between them is where the pressure will land next. YouTube Music has committed only to labels so far. Given that users are actively requesting blocking — and that Deezer has already shipped its toggle — logic suggests a Deezer-like filter is coming. But a filter sharpens the boundary problem to a legal edge: labeling a track costs the uploader nothing, while excluding it from a listener's feed costs reach. Misclassify a fully human track as "primarily AI" and you've silenced an artist on the platform's say-so.

The backdrop: rightsholders are already forcing the purge

What makes the labels feel less like user-service and more like triage is what else happened this week: DistroKid has been quietly taking down songs in response to UMG's lawsuit, The Verge reports — and per Digital Music News, some of the tracks being pulled belong to users who insist their work is entirely human-made, which is exactly the false-positive problem an automated "primarily AI" detector will have at scale. Tidal fully demonetized AI-generated music in June. That is the enforcement side; labels and filters are the consumer-facing side of the same cleanup.

But the DMN piece also raises the counterfactual worth taking seriously: do these moves open a market for AI-music platforms that embrace the walled garden from launch? Udio pivoted toward licensed generation; Suno's v6 rollout showed that even gen AI's own subscribers don't necessarily welcome sweeping changes; Klay Vision is building fully licensed products. Spotify, which owns listening but no generation tools yet, could plausibly house both. And ElevenLabs' UMG-partnered artist-specific offerings might find takers on the strength of the existing userbase alone. The very filtering that protects the incumbent DSPs could hand the AI-native platforms a population of listeners who actually want this music.

The uncomfortable middle

Both YouTube Music and Deezer deserve credit for moving before the question was answered. But neither has answered the question — because it may not have an answer that survives contact with real music. "Created primarily" is a spectrum being forced through a label, and the music that will cause the most trouble is precisely the music in the middle: human artists using AI tools, in ways that are already mainstream practice, whose work now has to be sorted by a detection system that cannot see intent.

The 100% machine-generated stuff — the slop, the upload-farms, the tens of millions of tracks — is easy to tag and everyone involved wants it tagged. The hard question is what happens when the badge lands on a track with a human name on it. That's when "transparency" becomes a judgment, and the judgment becomes a dispute. YouTube has just volunteered to referee it.

Sources