Streaming platforms have spent two years arguing about AI music policy while the volume of machine-generated tracks on their platforms kept climbing. Spotify has publicly refused to add a filter. Labels have scrambled to write licensing frameworks. Meanwhile, listeners have been largely unable to tell the difference, which is no longer just a subjective impression but a documented fact.
Deezer commissioned a study with Ipsos and found that 97% of listeners could not identify AI-generated music when it was played alongside human recordings. That number, published in April 2026, gave the Paris-based streamer both a problem statement and a commercial opening. The company had already built internal detection infrastructure. Now it is selling access to it.
What the Tool Does
Deezer's AI music detection tool analyzes audio at the signal level to identify statistical and spectral markers associated with generative music models. It does not rely on metadata, upload flags, or artist declarations, which are all trivially spoofed. The tool ingests audio files directly and returns a confidence score indicating the likelihood of machine generation. Deezer has commercialized it as a B2B product, licensing the detection layer to other platforms, distributors, and rights organizations that want to audit their catalogs or gate uploads.
How It Works in Practice
A distributor using the API submits a track at upload. The system runs analysis in near real time and returns a result before the track enters a release queue. The score can trigger a manual review, a block, or a label applied to the file depending on how the licensee configures the pipeline. Deezer has not published the full technical architecture, but the Ipsos study tested the detection model against a controlled set of AI and human recordings, where the tool outperformed human listeners by a wide margin across all tested genres.
Where It Fits for Agencies and Rights Teams
Music supervisors at creative agencies clear dozens of tracks per project. AI detection at the clearance stage catches catalog entries that distributors passed through without flagging, reducing legal exposure before a spot goes to air. For publishers and PROs managing royalty eligibility, the tool provides an audit layer that can be run against existing catalogs, not just new submissions. Streaming platforms building their own content policies can license the detection model rather than developing proprietary infrastructure from scratch, which Deezer's executive team has described as a multi-year engineering investment. Brand safety teams at media agencies that produce branded audio content or license music for social campaigns have a direct use case in verifying that third-party tracks meet client standards before placement.
Limitations and Trade-offs
Detection tools and generation tools exist in an adversarial loop. As generative models improve their spectral and temporal coherence, detection accuracy will degrade unless the detection model is retrained continuously. Deezer has not disclosed its update cadence or how it handles hybrid recordings, which mix AI-generated stems with human performance, a format that is increasingly common in commercial music production. The tool also generates false positives on heavily processed or electronically produced human music, which means any automated blocking pipeline needs a human review stage to avoid wrongful removal. There is also a jurisdictional dimension: detection scores carry no legal weight in most markets, so the tool informs policy decisions but does not resolve ownership disputes on its own.
Verdict
Deezer has done something the larger platforms have avoided: it turned an internal engineering capability into a product and put a price on the problem. The 97% figure from the Ipsos study is the most concrete data point the industry has produced on listener detection failure, and it makes the commercial case for third-party tooling clearer than any policy statement. As the volume of AI-generated content on streaming platforms continues to grow, detection infrastructure is becoming a standard component of any distribution or rights workflow, not an optional add-on. Deezer is positioning itself as the vendor of record for that layer, and the timing is precise.