Deezer and Ipsos found most listeners cannot spot AI-made tracks

The audio task

A platform weighing whether to label AI-generated tracks, and a rights holder gauging whether audiences even notice the difference, both need an answer to a specific question: when people actually listen, can they tell?

What the documents show

A Deezer newsroom post published 12 November 2025 reports a survey Deezer commissioned from Ipsos Digital, fielded 6 to 10 October 2025 among roughly 9,000 adults across eight markets: Brazil, the United States, Canada, the United Kingdom, France, the Netherlands, Germany, and Japan. In the survey's listening test, respondents heard three tracks and were asked to identify each as AI-generated or human-made; Deezer reports 97% failed to correctly distinguish all three, with 71% saying they were surprised by their own inability and 52% saying it made them uncomfortable. On disclosure, Deezer reports 80% of respondents want AI-made music clearly identified, 73% want to know if a streaming service is recommending AI content, and a majority said AI tracks should not appear in the same charts as human-made music, though only 11% actively supported equal chart treatment. The post also reports majorities calling it unethical for AI companies to train on copyrighted material without artist approval, and expressing concern about AI's effect on artist compensation.

Rights status

The post states listener opinion, not a legal position: it reports what surveyed people said about disclosure and fairness, not any determination that training on copyrighted music without consent is unlawful, or that any specific licensing arrangement is required. Deezer's own framing ties the results to its argument for transparency, which is the company's editorial position, not a court or regulator finding.

What to check before you use it

This is an editorial checklist. The survey was commissioned by Deezer, a company that separately sells its own AI-detection product, so treat the figures as company-reported market research rather than independently peer-reviewed research, and check Deezer's newsroom for the exact question wording before quoting a percentage out of context. Confirm which of the eight markets a figure applies to, since national results such as Brazil's reported curiosity about AI differ from the eight-market average. The listening test used three tracks per respondent, a small per-person sample that supports an aggregate population estimate but says nothing about any individual's ability to learn to detect AI music with practice.

• Is a percentage being quoted for the eight-market average or for a single country's result?
• Does citing this survey acknowledge Deezer's own commercial stake in the transparency argument it supports?
• Is “can't tell the difference” being confirmed against the actual three-track listening test, not assumed from a headline figure?

Whatever a reader concludes about disclosure policy, the survey's own numbers separate two distinct findings worth keeping apart: an inability to identify AI tracks by ear, and a stated desire for labelling regardless of that inability.