In the fall of 2012, Taylor Swift released Red. Her ambitions had outgrown Nashville, and so began a delicate dance: moving toward pop without rejecting the country audience that made her famous.

She went for it with her third single on that record, “I Knew You Were Trouble.” Co-written with Max Martin and Shellback, it is not a country song: Its chorus crashes into a distorted synth bass drop from Native Instruments’ “Massive” software synthesizer with programmed drums; the vocals run through distortion, compression, pitch correction, reverb and delay. In fact, the song shared a lot of DNA with Martin’s precision-engineered boy-band-pop, and fans ate it up: “I Knew” hit No. 2 on the Billboard Hot 100. 

Related

RIAA, IFPI & More Come Together to Create System for Tagging AI-Generated Songs

DJ Quik Says His Son 'Made a Mistake' Following Murder Conviction, Pleads With Fans to Stop Asking About It

Naoshi Fujikura of Universal Music Japan on Japan's Unique Superfan Culture & Global Ambitions: Billboard Global Power Players Interview 

And how did the music community react to the technology that took Swift from the Bluebird Cafe to Nissan Stadium? Nobody cared. No synthesizer backlash. No campaign to warn listeners that her voice had been processed. No demand that the song be labeled “electronically assisted music.”

Three decades earlier, the reaction was very different. In May 1982, the Central London branch of the UK Musicians’ Union passed a motion to ban synthesizers and drum machines, fearing machines would take work from working musicians (sound familiar?). American unions later fought the “virtual orchestras” threatening Broadway pit musicians.

The technology since then has changed; the anxiety hasn’t. What happens when a machine can do work that once belonged to a person?

Recently, an unusually broad coalition — the RIAA, IFPI, A2IM, WIN, IMPALA, the Recording Academy, SAG-AFTRA and the Human Artistry Campaign — proposed that streaming services label recordings “AI-generated” or “AI-assisted.” I understand why labels, artists and fans might support this. But the issue is far smaller than it seems. At best, the proposal is pointless. At worst, shortsighted. Here’s why:

1. Almost nobody is listening to this music. Streaming services now hold more than 253 million tracks, according to Luminate. Nearly half received ten streams or fewer in 2025; 88% got no more than 1,000. This is an unfathomable amount of music nobody hears: AI tracks, amateur uploads, meditation audio, karaoke, abandoned demos. Uploading a file is not building an audience. Most Gen AI tracks will land in the same graveyard as most human ones: available everywhere, heard by almost no one. 

Related

A Short History of AI-Generated Music: From ‘Fake Drake’ to Blockbuster Legal Settlements

2. “AI-assisted” is almost impossible to define. Does AI mastering, common for a decade now, count? Stem separation, pitch correction, sample discovery? One generative texture buried beneath 80 human-performed tracks? What if that AI sample is manipulated by a producer? Recordings pass among songwriters, producers, musicians, engineers and mixers, each using dozens of digital tools. Soon (if not now), nearly every professional recording may involve machine learning, and the label won’t identify anything exceptional — it will simply describe the modern recording process.

3. The detection technology will not keep up. Detection assumes generative models leave identifiable artifacts, but there is no universal fingerprint for “AI music.” Detectors must be trained on the specific model they hunt. When a model updates, its artifacts change; a new model sails through undetected. Add human engineering and fraud, and provenance blurs further. It’s an endless game of whack-a-mole, leading to false negatives and poor precision and recall. An inconsistently applied label is not meaningful transparency.

4. It creates a costly compliance system without solving the underlying problems. DSPs will spend their thin margins examining hundreds of millions of recordings nobody will hear, while pressure rolls downhill to distributors who can’t accurately certify the volume they process — an expensive charade of passing responsibility down the supply chain. After all that, an “AI-generated” label won’t improve royalties for real artists (AI-generated music is under 3% of the current pool, as found in a recent study by Deezer, and most of that is fraud, according to Deezer and other reporting); it also won’t stop streaming fraud, prevent infringement, or put a halt to voice cloning. It’s just more bureaucracy.

5. It will not meaningfully improve the fan experience. A recent Luminate survey found that 42% of respondents would become less interested in a song if they learned generative AI was involved in its creation, compared with 25% who would become more interested. I suspect that finding reflects how AI is discussed in the broader culture, more than the practical effect of AI on what fans are actually hearing.

Still, it should be taken seriously. There is a legitimate case for clearly identifying music that’s entirely generated by AI, particularly when listeners might otherwise believe that they’re hearing a human artist. But that does not justify the much broader and murkier category of “AI-assisted.” Will listeners skip a song they already like because of a small symbol indicating that AI was used somewhere in the production process? Probably not. Most do not know or care whether a song used Auto-Tune, programmed drums or AI mastering. And “AI-assisted” is so broad that it tells them almost nothing. For listeners who are especially sensitive to AI, the major DSPs may not be the best place to seek a fully curated experience. Alternatives already exist, including Bandcamp, Cantilever, NTS Radio, Discover.fm and the myriad of radio services. 

Related

How Many AI Artists Have Debuted on Billboard’s Charts?

6. It distracts from the problems that actually need solving. Artists should be protected from unauthorized voice cloning. Fraudulent uploads should be removed. Listeners shouldn’t be deceived into believing a synthetic recording was made by a real artist. Fully AI-generated music shouldn’t collect publishing royalties. And the training data legal questions remain. But these are specific problems requiring specific solutions: licensing, provenance, fraud detection, identity verification, legislation. Was the work authorized? Was anyone impersonated? Were the people who created the value compensated? A generic label answers none of those questions.

7. Technology-based labels age badly. “AI-assisted” may eventually look like labeling a recording “digital” in 1999 — technically accurate, too broad to be useful, quickly outdated. 

Taylor Swift did not betray human artistry with synthesizers and digital processing; she used available technology to execute her vision and grow her audience. Furthermore, the lesson of the synthesizer panic is not that new technology is harmless — it’s that dividing music into “human” and “machine” tends to age badly.

We should identify fraud, protect artists from unauthorized replicas, require disclosure when a synthetic voice impersonates a real person and keep fully generated music out of the royalty pool. But labeling nearly every modern creative tool “AI-assisted” will not protect human creativity — it will build an expensive bureaucracy around a distinction that’s already disappearing.

Drew Thurlow is a former senior vp of A&R at Sony Music, an advisor to the Mindset Music Tech investment fund, and a professor of Music Strategy, Innovation & AI at Berklee College of Music. His first book, Machine Music: How AI Is Transforming Music’s Next Act, was published by Routledge in spring 2026.