Deezer to Let Listeners Switch Off AI-Generated Music
Deezer will let listeners block AI-generated music completely with an on/off toggle in its app by the end of the year.
The control is set by each user and stays on until changed. When enabled, it keeps fully AI-generated tracks, albums and artists out of recommendations, search results and playback, according to the company's October 2026 newsroom announcement titled 'AI music toggle lets users filter content'. The toggle can be switched off to restore the full catalog. Engadget
On scope, this is not a label or a lower ranking. It is a hard filter in the presentation layer, the software stage that decides what appears on screen. Deezer said users will be able to exclude fully AI-generated artists and albums, and block AI-generated music from search results. Music Business Worldwide
The upload flood behind the filter
Deezer now reports that more than 50 percent of tracks uploaded daily are AI-generated, up from around 18 percent last year. In absolute terms, that is 90,000 fully AI-generated tracks per day, totaling more than 42 million tracks in the catalog pipeline.
The rise has been fast. In April 2026, the company put the AI share at 44 percent of new uploads. Passing 50 percent six months later points to continued acceleration in generative output, not a plateau. For catalog operations, the intake and checking of new music, that shifts the work from handling exceptions to sorting most incoming tracks.
Deezer CEO Alexis Lanternier said there is clear user demand for the ability to filter out AI-generated music. A company study found nearly half of streaming users want AI-generated music completely gone from their feeds. A separate study by Deezer and Ipsos found that AI fools 97 percent of listeners, which helps explain why manual avoidance does not work. Discovery feeds and autoplay decide what plays in most sessions.
How Deezer detects AI audio
Deezer deployed its AI detection tool for music streaming in January 2025. The system is built to spot artificially created music from generative models such as Suno and Udio. The company described the risk of a false positive, where a human-made track is wrongly flagged as AI, as very low. That low error rate is needed for any filter that removes content from search and recommendations.
The company launched one of the industry's first AI music detection tools and has offered to loan it to rival streaming platforms.
The wider point here is detection works at ecosystem scale only if distributors, rights holders and services share signals and tooling.
Deezer has already put detection to use outside its own app. In June 2026 it launched a free online tool to detect AI-generated tracks in playlists, available to users of all major streaming platforms. Reuters Its consumer-facing AI labelling notifies users instantly when a song was generated by AI. In March 2026, Deezer announced a partnership with EJI to filter out recordings created by AI, extending the same classification task into rights enforcement.
The broader context here is catalog integrity, not taste. Streaming catalogs were built for human-scale ingestion, with metadata review, duplicate detection and fraud controls sized for thousands of releases per week. Generative systems produce at machine scale. Without automated classification at ingest, every downstream system degrades. Recommendations get noisier. Search gets polluted. Royalty pools get diluted by tracks with negligible marginal production cost.
Looking at what this means for platform design, Deezer is treating AI provenance, information about origin, as a user preference, like explicit-content filtering, rather than as a single rule for all. That preserves choice. Listeners who want synthetic soundtracks for focus or workouts keep access. Listeners who want only human-made recordings can opt out entirely. The classifier does the hard work once. The presentation layer respects the toggle.
In my view, this is the pragmatic path, and other services will have to answer it. Centralized removal would trigger endless boundary disputes about assisted production, mastering tools and hybrid workflows. Deezer is sidestepping that by limiting the toggle to fully AI-generated content and leaning on its low false-positive claim. The edge cases remain difficult. Co-writes, AI vocals over human instrumentals, and heavily processed samples will test any binary classifier. Worth flagging for engineers, provenance metadata and detector scores need to travel with the track object through ingestion, search indexes and recommendation features, or the filter will leak.
The long arc here favors better tooling. I covered the spam wars in email, the content-farm era in search, and the bot waves in social. Each time, the open system looked unmanageable until detection, reputation and user controls caught up. Music is now going through the same cycle. A reliable detector plus a simple user control does not solve attribution or compensation. It does restore a basic condition for trust. Listeners know what they are hearing, and artists compete in a catalog where origin is at least visible to the system, if not always to the ear.
