In the pre-internet days, even a mega-sized record store stocked perhaps 100,000 titles of all eras and all genres. That is a quaint concept now, given that streaming music platforms now draw on a library of about 250 million songs, a number that used to increase by about 100,000 tracks a day as musicians — pro and amateur, good and bad — around the world sought to distribute their music.

Overwhelming, yes, but nothing compared to what’s happening today, thanks to AI.

Using tools such as Suno, Google Magenta, Loudly, Mubert, and perhaps a dozen other online sites, creating a new song is as easy as entering a few text prompts. Results are delivered within seconds. And an astounding number of these tracks are being uploaded to Spotify, Apple Music, YouTube Music and every other platform.

Deezer, the Paris-based streamer, appears to be watching the situation more closely than most. Back in January 2025, the company, which employs in-house AI-detection software, reported that 10,000 AI-generated tracks were being uploaded daily. By April, that number was 15,000. In September 2025, the company found that the number had increased to 30,000, which climbed to 50,000 by November. This year started with 60,000 AI uploads in January. The most recent figure is 75,000, which has surely already been eclipsed.

Again, that’s the number of AI-created songs per day. To put it another way, these tracks account for 44 per cent of all downloads to Deezer. Meanwhile, the number of human-created songs sits at around 95,500 per day, a figure that has barely budged over the last 15 months. The robots are gaining. Fast.

And if Deezer is experiencing this problem, you can bet that all the other streamers are, too. Deezer has offered to make its AI-detection software available to all streamers, which, by the way, has sniffed out 13.4 million AI tracks since the beginning of 2025.

You may ask why this is a big deal. First, let’s consider the energy costs. Each of these fake takes electricity, anywhere from 0.001 to 0.01 watt-hours, which may not seem like much, but if you do the math, the overall amount of daily energy consumption creating junk that almost no one will hear is significant. Add in the costs to run servers for storage and transmission and the problems multiply.

Then there are the bots deployed to “listen” to these tracks, which pump up streaming numbers and fraudulently sucks up royalties. They suck up a lot of energy, too, but that’s not the biggest issue. This is where we encounter streaming fraud.

Get breaking National news

Get breaking Canada news delivered to your inbox as it happens so you won’t miss a trending story.

Last month, a 54-year-old North Carolina man pleaded guilty to masterminding a multi-year scheme that sucked away millions in streaming royalties. Over seven years, Michael Smith used AI to create hundreds of thousands of junk songs, which he then posted to all the streaming platforms. He then used several battalions of bots to “listen” to the fake songs (that is, to stream them automatically). This “listening” was distributed over thousands of different automated accounts, which made the fraud very difficult to detect.

When things were moving at maximum speed, Smith was deriving revenue from over 660,000 streams a day, which worked out to about $1.2 million a year. All he had to do was let the software run and watch the money gush into his accounts. Who got hurt? This money came from the pool of royalties from which all legitimate musicians are paid. This meant that payments to humans were lower than they should have been, per the rules of payouts.

He was finally caught in September 2024. Part of his sentence was the forfeiture of more than US$8 million in royalties. We’ll see if he gets any jail time when he again appears before a judge on July 29. The maximum penalty is five years.

Meanwhile, human performers are being victimized. Paula Toledo is a Vancouver-based singer whose music was co-opted and re-released by bots using a fraudster’s account. She had all kinds of trouble proving to Spotify, Apple, etc., that she was the rightful owner of that music.

Grace Mitchell, an American-Australian singer-songwriter, was shocked to discover that her music had also been stolen and re-released by persons unknown who then profited from royalties that should have gone to her.

A while back, I ran across an “artist” on Spotify that sounded suspiciously like Van Morrison. It was, but someone had simply uploaded a legitimate Van Morrison album under a different name. I have a singer-songwriter friend who recently discovered that her music had been hijacked, too.

Finally, there are the fake singers who do not exist IRL. Xania Money, SiennaRose and Eddie Dalton are all cyber creations. IngaRose is an AI-generated thing that hit the top of the iTunes sales chart in Canada, the U.S., the U.K., France and New Zealand earlier this year with the song Celebrate Me, which was also used in nearly 300,000 TikTok videos. Breaking Rust, an AI-generated act, hit the top of the Billboard Country Digital Song Sales chart back in November with a song called Walk My Walk.

Last year, people were faked out by a generic-sounding California folk rock band called The Velvet Sundown. And just last week, I ran into Promptgenix, a mysterious group that released three albums in two months. No bio, no info, nothing. It turned out that they were created by an AI engine called Promptgenie by someone who complained that he was doing this because he couldn’t find any song by a human that he liked.

The situation is only going to get weirder. A survey from last November — a million years ago in the AI music evolution —said that 97 per cent of listeners were fooled by AI music. And while many people say they disapprove of “art” being made by software, possibly up to 50 per cent of U.S. listeners aren’t bothered by it, according to a 2025 report by Luminate.

Meanwhile, Warner Music has settled its lawsuits with Suno and Udio (another AI music company) and has entered into some kind of partnership regarding AI music. No one knows what that means yet, but it can’t be good for human musicians. Universal, meanwhile, also has a deal with the Udio platform after initially trying to sue it out of business.

Years ago, when we were all still making mixtapes on cassette, Memorex boasted that their tapes were so good you couldn’t tell if it was a live performance or a cassette. The campaign was called, “Is it live or is it Memorex?”

We’ve now entered the age where the question is: Is it living or is it software?