Start-Track’s Guide on How to Spot AI Music

AI music is fast food: Quick and cheap, and looks like real food, but at the end of the day is nothing more than slop meant for mass production and consumption...

Imagine: Your friend sends you a new song from an up-and-coming artist. Upon the first listen, you love it: In fact, you think it’s one of the best songs you’ve ever heard. And from such new talent! Then you take a second listen, and something seems off, but you can’t tell what. Upon further research, you find out the song, the artist, and even the album art were all generated using AI tools. From that point, the song is no longer in your library, and that person is probably no longer your friend.

AI music has become more advanced due to improvements in platforms like Suno and Udio for full-song generation, InstaComposer for MIDI generation, and the usual suspects like ChatGPT and Claude for lyric writing. This has made AI-generated music harder to distinguish from 100 percent human-created art, prompting platforms to take action against AI-generated music. For example, the French streaming service Deezer now tags songs it detects as AI-generated. And most notably, Bandcamp has adopted a new policy banning any music “generated wholly or in substantial part by AI.”

But artificial intelligence still manages to slip through the cracks and land in the laps of music lovers, especially with constant improvements in AI tools. As more critics pick at the faults of AI music, it will sound more like the real thing. Therefore, the tips in any article you read about how to tell AI from human-made music - including this one - may become obsolete within a year, if not sooner. But you may choose to follow them nonetheless.

The artist’s info seems sketchy. Before worrying about the music, it helps to look up info about any musician or band. Where do they come from? How did they get their start? What do they look like? Most human acts will have at least this much information about them on a press release, website, or blog post. AI acts may have this info as well, but a keen eye will help you separate them from humans. If their bio seems vague and their profile picture looks too polished (or doesn’t exist), you might have a bot on your hands.

The artist has released a lot of music in a short amount of time. While this doesn’t always mark an artist as AI, people who create AI music tend to drop dozens of singles or even full albums at an impossible rate. Look up the discography of the artist and see if they’ve released a lot of new music lately. If so, that means someone’s burning through their Suno credits.

If your artist has an iron-clad backstory and a reasonable amount of releases under their name, it’s time to take a closer listen to the music. Turn up your headphones and follow these tips:

The music contains strange artifacts. Often, AI music contains microscopic artifacts - glitches or hiccups - in the vocals or instruments. You may hear a breath or a word where it doesn’t belong, or an instrument cut off in a strange place. One of the biggest telltale signs in AI-generated music centers on vocal mixing: every song has a hiss layered over its singer’s voice, which comes out stronger in softer songs. While these artifacts may sound like editing errors, it wouldn’t make sense for a keen-eared producer to make the song sound glossy yet forget to remove those mistakes. Platforms like Suno have refined their models to prevent artifacts from appearing in the music and have created tools like Suno Studio to let artists work out these kinks themselves, as they would in a DAW. However, some artifacts still slip through the cracks.

Everything sounds too glossy. Despite the occasional blips, AI music does its best to sound too perfect. This tip comes from Dylan Paul Ware, who creates music under the name The Infinity Chamber. “AI songs tend to have a certain sound: Too slick, too manicured, too clear in a way they often tend to share.” Vocal transitions can sound too polished, as if the auto-tune got turned up to eleven, or abrupt, as if the singer tried to cram too many words into too little time. And while the backing band might sound professional, the mixing makes each instrument sound too hollow or compressed in a way that squeezes the life out of the music. It may take a real audiophile to hear what this sounds like; however, any ear will pick up on the artifacts and glitches more easily because the rest of the music just sounds too good.

Certain parts of the music sound too familiar. Because AI models scrape from existing materials - basically the entire Spotify catalog - they tend to rely on popular sounds in each genre to generate results. Therefore, it’s no surprise that the up-and-coming (AI) artist sounds like Adele, or why the guitarist is laying down solos like Slash. But this isn’t a telltale sign of AI; human artists borrow from each other. Of course, humans can work out an arrangement if one artist copies another. When AI copies someone else’s work, the law becomes foggy.

The music sounds too generic. At the end of the day, because AI takes from music already out there, it ends up sounding like the average of everything in its database. Therefore, the soul of the music doesn’t exist. Rock music has no edge, even when everything sounds edgy. Pop music glistens, but has no shine. Hip-hop doesn’t come from the heart, and the rhymes don’t land as hard. Even electronic music - seemingly a slam dunk for AI - ends up hollow. And the more AI music you hear, the more you will hear similar musical phrases, vocal riffs, and styles copied across each track.

AI music is fast food: Quick and cheap, and looks like real food, but at the end of the day, it is nothing more than slop meant for mass production and consumption to make a buck for someone higher up the ladder. But unlike fast food, which you can tell is lower quality, AI music is becoming better at hiding its flaws, and the prompt engineers are getting more creative with their artists’ identities, online presences, and storylines. If you want to watch out for AI music, you should keep all of the tips above in mind and stay vigilant, especially in an age when truth and fiction have become harder to separate.

Once again, Start-Track stands against AI and supports art created by humans for humans. This article originally included examples to show what to notice in AI-generated music, but they were removed to avoid providing any support whatsoever to artists who use artificial intelligence.

---

Written by Will Sisskind

P.S.: Please consider donating to our blog.