Amid an influx of AI tracks on music streaming services, platforms are evaluating how best to manage AI-generated and assisted tracks.
The table below represents a snapshot of the current AI content policies and enforcement methods at music digital service providers (DSPs). These could change as the music industry pursues consensus and common standards on how platforms should treat AI music.
In addition to reviewing company websites and news articles, Luminate Intelligence reached out to companies to verify information relevant to each platform’s approach to moderating AI music content. YouTube, Tidal, Pandora and SoundCloud were not included because they have yet to publish detailed AI music content policies and did not respond to inquiry.
Though the industry doesn’t have a common definition for what qualifies as AI music content, DSP policies are primarily focused on moderating fully or partially AI-generated tracks. A fully AI track is 100% generated by an AI model or song generator service such as Suno or Udio from a text prompt. A partially AI track contains some AI-generated material as well as human-authored or performed elements.
Among the main concerns regarding the influx of AI-generated tracks on DSPs is its link with streaming fraud activity. Scammers can easily generate and upload large volumes of AI songs, deploying bots to artificially inflate streams, which risks diluting streaming royalties.
Though the wider music industry has recognized how AI tracks have exacerbated streaming fraud, the only DSP to quantify its impact has been Deezer, which reported in April that while AI-generated tracks detected on the platform accounted for just 1%-3% of total streams, 85% of streams on those tracks were identified as fraudulent.
While AI music has exacerbated the longstanding problem of streaming fraud, DSPs have existing systems to identify and remove fraudulent and low-quality spammy tracks, which often encompasses AI-generated tracks. This means that DSPs are likely already removing large numbers of AI tracks regardless of whether they’ve set specific AI content policies.
Current DSP approaches to AI content differ along several key lines:
- Hosting: Is AI content permitted on the platform? Most DSPs still allow AI tracks to exist on their platforms, though some are mitigating any impact on royalties paid to human artists and user trust by tagging or labeling AI content and/or excluding these tracks from recommendations and playlists.
- Transparency: Is AI content visibly tagged or labeled? Several DSPs have taken steps to disclose to users when a track is AI generated or contains AI-generated material though differ in their preferred approach to identifying and labeling. France-based Deezer and Qobuz have built and deployed proprietary detection systems to identify and tag AI music, while Spotify and Apple Music are requesting content providers disclose AI use in a track’s metadata. For detection-focused platforms, AI content labels have been directed primarily at fully AI-generated music. Metadata disclosures allow for more granular information about specific AI use, but they’re voluntary for content providers to give.
- Discovery: Is AI content recommended or playlisted? So far, only Deezer and Qobuz say they exclude AI music from algorithmic recommendations or playlists. While this helps prevent users unwittingly listening to AI music, it’s designed to minimize monetization on fully AI-generated tracks.
While DSPs bear the brunt of the AI music flood, distributors may be increasingly key enforcers of AI content policies upon the tracks they accept and deliver, as they’re upstream of streaming platforms.
Among the three biggest distributors, AI content policies also differ. UMG-owned CD Baby refuses both fully and partially AI-generated tracks. Believe’s TuneCore rejects fully or partially AI-generated tracks specifically from AI music services supported by models trained on unlicensed content. DistroKid accepts AI-generated music, though with limitations including that uploaders must own 100% of the rights to distribute the content.
Yet for DSPs and distributors to manage AI content more equally across platforms, the broader music industry — music companies across the value chain as well as artists and songwriters — needs to come together to develop common standards on three fundamental questions:
- What is AI content? Defining AI music content is complicated by the spectrum of possible AI use in music production, ranging from full AI song generation from a text prompt to partially generated tracks that contain one or multiple different AI-generated elements, such as lyrics, vocals, instrumentation or a sample.
Much of the industry expects generative AI to be used as a creative tool in music production, meaning authentic artists may also engage in some AI use, further complicating the question of what kind or degree of AI use should be considered acceptable. - How is AI content identified? The industry has yet to adopt a standard approach to detecting AI content. Between two available methods — AI detection and metadata disclosure — neither is perfect. Detection systems aren’t 100% accurate, and self-disclosure is only as reliable as those providing the information, their knowledge of how a track was produced and their willingness to share when disclosure is optional.
Both methods could be used in conjunction to increase overall effectiveness, such as by requiring disclosure from content providers and using detection systems to help verify that metadata is accurate and complete. Though its use is optional on its own platform, Spotify is leading the development of a DDEX metadata standard for AI disclosure intended for industrywide adoption toward unifying information across platforms. - Should AI content earn equal royalties? AI tracks currently receive streaming royalties equal to human-made or traditionally produced tracks on DSPs because most DSPs, including Spotify, YouTube Music and Apple Music, distribute royalties on a pro rata basis.
As AI tracks grow on DSPs, some have proposed creating separate royalty pools or tiered streaming royalties, with lower rates paid to AI track streams. Some have further advocated DSPs switching to an alternative “user-based” or “artist-centric” (rather than pro rata) royalty model that would pay based on the artists a fan streams, which it is believed would favor legitimate artists and tracks. Alternatives would require wider industry agreement before being implemented on platforms.
