In a blind test commissioned by Deezer and conducted by Ipsos, 97% of 9,000 adults failed to distinguish fully AI-generated tracks from human-made music. Each respondent heard three songs, two created by AI and one recorded by human musicians, and had to identify which was which.
The Deezer–Ipsos Blind Test
Survey parameters and key results:
- Fieldwork: October 6–10, 2025, online, through Ipsos Digital.
- Sample: 9,000 adults aged 18–65, including 6,791 music streaming users.
- Countries: United States, Canada, Brazil, the UK, France, the Netherlands, Germany, and Japan.
- Reaction to the result: 71% were surprised, and 52% felt uncomfortable about failing to tell the difference.
- Labeling: 80% want fully AI-generated music clearly labeled on streaming platforms.
- Charts: 52% say fully AI-generated songs should not appear in charts alongside human-made music.
- Listening behavior: 45% of streaming users want an option to filter AI music out, and 40% would skip it.
- Artist income: 70% believe fully AI-generated music threatens artists’ livelihoods.
AI Singers That Have Already Reached Audiences
Several AI-generated acts gained large audiences before most listeners identified them as synthetic:
- The Velvet Sundown: The project released the album Floating on Echoes on June 5, 2025, and gained more than 1 million monthly Spotify listeners within about a month. On July 5, 2025, it described itself as a “synthetic music project guided by human direction” after weeks of speculation.
- Xania Monet: Poet Telisha Nikki Jones writes the lyrics, and the vocals and music are generated with Suno. The track “How Was I Supposed to Know” reached No. 30 on Billboard’s Adult R&B Airplay chart, and “Let Go, Let God” entered the Hot Gospel Songs chart. Jones signed a multimillion-dollar record deal with Hallwood Media in September 2025.
- Breaking Rust: The AI-generated track “Walk My Walk” reached No. 1 on Billboard’s Country Digital Song Sales chart in November 2025.
Why Human Listeners Fail
The study “Echoes of Humanity: Exploring the Perceived Humanness of AI Music” by Flavio Figueiredo, Lucas N. Ferreira, and co-authors was presented in the Creative AI track at NeurIPS 2025. Its methodology and findings:
- Design: A randomized, controlled crossover blind test in which participants identified the AI-generated song in each pair.
- Material: Songs created by real users of the commercial service Suno, not tracks generated by the researchers.
- Main finding: Detection became more reliable only when the two songs in a pair were similar in style, which gave listeners a direct reference point.
- Cues used: Listener comments focused mainly on vocal characteristics and technical production artifacts.
In everyday streaming, tracks are not presented in matched pairs, which removes the reference point that improved accuracy in the study.
The Scale of AI Music on Streaming Platforms
Deezer data on fully AI-generated uploads:
- January 2025: About 10,000 tracks per day, or 10% of daily deliveries.
- November 2025: More than 50,000 tracks per day, or over 34% of daily deliveries.
- June 2026: About 90,000 tracks per day, exceeding 50% of daily deliveries on peak days.
Additional figures from Deezer’s July 2026 report:
- Fully AI-generated tracks account for 1–3% of total streams on the platform.
- Up to 85% of streams on fully AI-generated tracks in 2025 were identified as fraudulent and excluded from royalty payments.
- Deezer detected and tagged 13.4 million AI-generated tracks in 2025.
- The detection tool’s accuracy is 99.8%, with fewer than 1 in 10,000 human-made tracks falsely flagged.
Software detection therefore reaches 99.8% accuracy on the same task where 97% of human listeners fail.
Do AI Labels Change How Content Is Perceived?
Labels inform listeners, but their wording affects reactions. A study by Timo Schreiner, Yakov Bart, Koen Pauwels, Cosima Schütze, and Fabian Buder, published in the Journal of Marketing Analytics on August 27, 2026, tested explanatory AI disclosures. Details are covered in an overview of new research on whether better wording can fix the “Made with AI” problem.
Key results:
- Study 1: 339 U.S. participants rated specific explanations of AI use higher than a minimal “Created with AI” label when the statements were shown as text alone.
- Study 2A: 359 U.S. participants viewed the same disclosures inside Instagram-style ads, where the gains disappeared and some effects turned negative.
- Platform practice: Meta introduced “Made with AI” labels on Facebook, Instagram, and Threads in 2024 and later renamed them “AI info.”
Online Presence for Music Projects
Human and AI-assisted music projects publish official websites to host release pages, press kits, and statements on how their music is produced. The .co extension is the country-code top-level domain of Colombia and is used internationally as a short alternative to .com. Artists, labels, and independent producers can buy .co domain names to secure a project name across platforms.
Measures that address the detection gap:
- Platform-level AI tags, such as Deezer’s album-level labels, identify synthetic tracks without relying on listener judgment.
- Official artist websites provide a verified source of information about who created a release.
- Filtering options respond to the 45% of streaming users who want to exclude fully AI-generated music.
