On March 10, an AI actress named Tilly Norwood dropped a song called "Take the Lead." Made with Suno, timed to the Oscars, accompanied by flying flamingos and lyrics urging human actors to embrace the technology designed to replace them. Ten major outlets covered it. Deadline. Variety. TechCrunch. Gizmodo. Euronews. Forbes. ABC. NBC. Hollywood Reporter. Billboard. The reviews were uniformly brutal. The video had 4,000 views four hours after launch.
Five days later, "Golden" from KPop Demon Hunters won Best Original Song at the 98th Academy Awards. Eight weeks at number one on the Billboard Hot 100. The first K-pop song to win an Oscar. Entirely human-made.
Same week. Same cultural moment. Two completely different outcomes.
The Gap Nobody Is Talking About
Ten major outlets covered Tilly Norwood. Under 30,000 views after 24 hours. The journalists writing those reviews were not the people watching the video. The verdict was delivered before most people pressed play.
This is not a coincidence. Shank et al. (2023), Journal of Experimental Psychology: Applied, ran three experiments where listeners rated identical music. Half told it was AI-composed, half told human-composed. The AI group rated it significantly lower every time. Same audio file. Different label. Different score. The Oscars framing made the Tilly Norwood case an almost perfect controlled experiment. The cultural narrative was set before anyone heard a single note.
What "Golden" Actually Proves
"Golden" did not win because it was human-made. It won because it was exceptional. Seven writers. Eight weeks at number one. A song that had to function as a real pop hit inside a fictional K-pop film, then win at the Grammys, Golden Globes, Critics Choice, and Oscars.
That is not a verdict on AI music. That is a verdict on one exceptional song. The question nobody has a clean answer to: what would have happened if "Golden" had been AI-generated but sounded identical? Would it have been nominated? Would it have won?
We do not have that data yet. But we have 6,500+ blind ratings that suggest something worth paying attention to: when people rate music without knowing who made it or how, the scores are lower than most AI creators expect. Average 2.8 out of 5. Not because the music is terrible. Because honest judgment is harder than we think, even without the label doing the work.
Read our full breakdown of the AI composer bias research →
The Week in Full
Monday: Tilly Norwood releases "Take the Lead." Ten outlets cover it. Under 30K views.
Thursday: VoteMyAI crosses 1,000 tracks rated blind by 300 strangers. Average score: 2.8/5.
Sunday: "Golden" wins Best Original Song at the Oscars. Tilly Norwood does not attend.
The week did not prove that AI music is bad. It proved that the conversation about AI music is still almost entirely about the label, not the sound. That is the problem worth solving.
What Comes Next
The tools are getting better. Suno V5 exists. Lyria 3 from Google launched last month. The quality gap between AI and human music is narrowing faster than most people want to admit.
The bias is not narrowing at the same rate.
That gap, between what AI music sounds like and what people think it sounds like before they press play, is exactly what VoteMyAI was built to measure. If you want to hear what AI music actually sounds like when nobody knows what it is, the radio is running.
Curious about the tools being used to make the music on VoteMyAI? Check out our Creator Tools page for the platforms our community uses most.
Hear First. Judge Second.
VoteMyAI is a blind rating platform for AI-generated music. No artist names, no tool credits, no clout. Just the track — and your honest reaction.
Rate Tracks on VoteMyAI →