AI is a moving target, and the tricks that help us identify it can quickly become outdated. When Suno launched their music generating platform late in 2023, one obvious tell was what you could call the Forgetful Bassist, who seemed to never remember what had been played before, and chose a different groove every time the bass returned. Now, that tic has largely disappeared. But while the artifacts of AI have become more subtle, they aren’t too hard to find (obligatory: yet).
These days one of the best places to start is guitar, while drums help corroborate the evidence.
AI Tells in Guitars and Drums
A guitar string might squeak as the plastic pick strikes the steel alloy, and all of those noises arrive at slightly different times than the sustained note held by the vibrating string. A snare hit comes with a cascade of different frequencies that tell our ears more than just how the drum heads are tuned, like how big the recording studio might have been, and how hard the snare was struck. AI can struggle with all of this extra texture; including it when the instrument stands out, forgetting it in a crowd. The generators also tend to focus on the “main” range of frequencies, to just “play the note.”
You can hear drum and guitar tells in songs by The Velvet Sundown, which in 2025 became one of the first generative AI acts to go viral — until a spokesperson/creator/conman(?) admitted that the project ran on Suno. In “Drift Beyond the Flame,” the drums are suggestive of AI, with the main snare — the sound that the whole song is built around — coming out quiet, dull, and flat. Medieval armies marched to snare drums, and even your average bedroom producer would tend to record something with a bit more oomph, if only by accident; it’s a naturally powerful instrument. Still, while this meek snare could be a production choice, that’s not true of the morphing guitar.
Related Video
The texture of the guitar evolves throughout the track, and especially as the music swells — as if the extra detail is the first thing to go when the generator has more to keep track of. During verses you can hear strums, but at the end of the chorus and parts of the bridge, the instrument sounds more like keyboards, and then shifts with the smoothness of a programmed synthesizer. Skip between about 35 seconds and 1:30 and hear a notably different tone. Amid the festival of other noises it forgets to hold the shape of a guitar; the audio equivalent of swirling faces and six-fingered hands at the edge of an AI video crowd.
The AI music detectors also seem to have an easier time with some instruments. A Barcelona team recently tested how well a neural network could spot machine-made stems hiding inside an otherwise totally human mix. The authors wrote that “drums and guitar carry strong codec-artifact signatures, while vocals and bass are less detectable.”
Which instruments give it away depends on where their energy sits in the spectrum: the bright, fast, high-frequency stuff gives the machine away, and the low and midrange mostly doesn’t. That high end happens to be the same end of the spectrum where a human ear does its best work, which is why drums and guitar are good entry points for all the various intelligences trying to crack this problem.
Optimized Vocal Loops
When it comes to vocals, however, humans distinguish ourselves over the machines (obligatory: for the moment). One of the stronger AI tells comes in long, looping melodies — the kind that don’t offer a chance to take a breath, and generally would not occur to songwriters with a handicap like lungs.
A viral example was the 2025 country oddity “Walk My Walk,” credited to Breaking Rust. The song has a couple of AI hallmarks — the generators like a bit of stomp with their clap, just like an IPA-addled Millennial — but the most obvious is the breath. Partway through the chorus, on what appears to be a single inhale, the singer plows through all of these phrases with the intensity of mid-vocal-run Chris Stapleton:
Ain’t changing my tone, ain’t changing my song
I was born this way, been loud too long
You can hate my style, you can roll your eyes
But I ain’t slowing down, I was born to rise
So kick them rocks if you don’t like how I talk
I’m gon’ keep on talking and walk my walk
And yes, awful lyrics are another tell. But there are plenty of guides to AI writing, so let’s focus on the breathing.
It’s not literally impossible to sing this way, it’s just inhuman — in the way that chess engines running the h-pawn up the board or AlphaGo’s move 37 felt inhuman. The AI looks to optimize the melody; the squishy human knows it can drop a word to sneak a breath.
These looping melodies are one of the few tells that let us know AI was used as a songwriting aid before a human recording session. If Suno helps come up with the melody, and the melody is recorded with professional musicians, the detectors wouldn’t note the same odd decoder fingerprints, and you wouldn’t be able to catch the guitar morphing.
According to Pat McAfee’s producer ERNEST, McAfee’s 2026 album The Diary of a Polarizing Figure started as “Suno demos,” with ERNEST offering to “cut this stuff for real.” We don’t know if the producer or friends generated new toplines (McAfee’s reps did not respond to a request for comment) but the looping melodies give it a surprisingly familiar feel.
McAfee’s “Blue Collar Symphony” has a chorus that echoes Breaking Rust’s, even so far as to contain similar loops that rise to emphasize a word and then step back down the scale. He needs autotune to get the breathless effect, and while it’s not exactly the same melody, it comes from the same inhuman impulse to optimize the loop.
The tools built for this aren’t quite built for this, if you know what I mean. Deezer flags generated audio on its platform, Billboard consults that flag when it decides whether a charting act is a person, and the whole apparatus sorts music into made by a machine or not. Breaking Rust trips it. McAfee’s record doesn’t, because there is nothing in the audio to trip it with — real musicians played every note, and the only thing Suno left in the room might be the shape of a melody.
So listen to the guitar, and listen to the drums, and listen for the singer who never breathes. Those tells will help (obligatory: for now).
