Telltale Signs that a "Vintage" Song on YouTube is AI

Last updated September 7, 2025


The Artist and Their Videos

  1. The artist churns out new songs and albums far more frequently than humanly possible (e.g. a new song each day for months on end).
  2. The cover art looks computer-generated and/or shows a person with an iPhone face (generically attractive by modern beauty standards). Bonus points for weird or nonsensical text.
  3. The artist (or, rather, their creator) acts coy about whether or not they really exist. The Velvet Sundown is the best example of this, with their description of themselves as "sound[ing] like the memory of a time that never actually happened... but somehow they make it feel real."

Music and Song Structure

  1. The production values don't sound like those of an authentic recording from the era. Sound recordings made with analog microphones in recording booths sound different than those made with computer microphones, but AI can't tell the difference. Bonus points if the singer's voice sounds like it was processed using Autotune.
  2. The song's structure does not resemble that of most authentic songs of the era. Notably, fake jazz, imitation kayōkyoku, and 1950s-style pop songs that are generated with AI rarely follow the AABA song structure that was common in those genres in mid-20th century, nearly always following the verse/chorus/optional bridge format common in more recent pop music instead. In extreme examples (such as many of the fake 1960s and 1970s Japanese songs on the YouTube Channel utakawa), the tune consists of complete nonsense with little resemblance to human-created music.
  3. The music features flourishes from random (typically brass) instruments at random times. (These are particularly common in utakawa's songs.)

Lyrics

  1. The same words and phrases are recycled over and over from song to song, most likely the result of generating multiple songs at once using the same prompt. The Velvet Sundown's lyrics and even song titles repeat words and phrases from song to song constantly. Likewise, Frankie Flairs' songs often mention linen, ice, drinks being poured into glasses, silver, and (oddly) animate clothing doing things in the (marble or apricot) hall. Echoes and whispers are recurring themes, especially in song titles; for some reason, AI song generators seem to be fixated on both. (See the song titles on the YouTube channel utakawa, created using Udio, for some examples.)
  2. The song's lyrics are unusually vulgar for the alleged time period and/or about a subject that would have been considered taboo.
  3. The lyrics make excessive references to stereotypes associated with the period compared to authentic songs from the era. For example, I've heard many more AI imitations of 1950s and 1960s songs reference jukeboxes than do actual songs from the 1950s and 1960s. "Swish Swish" by Kitty Summers is an extreme example, mentioning jukeboxes, poodle skirts (which I have literally never heard mentioned in any real 1950s song), hairbows, greased hair, soda (whatever that is—I think it's something like pop ;)), spinning records, and neon lights, all in one two-minute song.
  4. The same word is used in two different senses within the same stanza, but not how an actual human lyricist would use wordplay. Examples that I've heard include "twirling around in a diamond ring" and "my teardrops are falling in love on my pillow." If the wordplay is used in a way that doesn't make sense the way wordplay by a human being would, it was probably written by a robot that doesn't understand what it's saying.
  5. Likewise, the lyricist doesn't seem to genuinely know what the objects mentioned in the song are and/or how they're typically used. The aforementioned "Swish Swish" provides one example in the line, "Swishing under the jukebox glow," which seems unlikely, unless the jukebox is up on a mezzanine or otherwise above the singer. Another example can be found in Frankie Flairs' "New York, That Dazzlin' Town" in the line, "I saw a kid tap-dancin' down the turnstile," which implies that the AI thinks that a turnstile is a kind of thoroughfare and that it would be possible to tap-dance on one without quickly falling on your face. In other songs, he mentions "basil on skin" and declares that "nobody tips the moon." (While doing research for this article, I asked my mom if she'd ever heard that phrase used before. She said no, and Google did not turn up any search results for "nobody tips the moon" that weren't related to Frankie Flairs, so I suspect that the AI invented it.)
  6. In the case of ballads, the story is full of inconsistencies and plot holes.
  7. Parts of the prompt are included in the lyrics. Example: "Give a trumpet solo over soft upright bass and brush snare" from "The Ice Never Melts Here" by Frankie Flairs. Bits of the prompt may also appear on the cover art, as with image-based AI slop.
  8. The lyrics fixate on material objects (stereotypical for the period or otherwise) and spend more time describing them than most human lyricists would.
  9. The lyrics fixate on details or scenarios that most human beings would find uninteresting and/or unappealing. The best example I've found for this one is Frankie Flairs' "The Quiet Rich." The first verse goes:
    The bath draws itself at a quarter to ten
    Drawn by a man I may never see again
    The silk robes warm from a waiting rack
    And the sun knows just where to land on my back
    There's citrus sliced like a crown on glass,
    A poached pear set on Burmese brass
    The eggs arrive with a silent glide
    And the toast has never once been tried on the wrong side

    Later in the song, he sings:

    The staff moves soft like a hand in a glove
    They're trained not to knock, just to understand
    I suspect that his staff all hate him. Honestly, he sounds like an insufferable boss.
  10. The lyrics are consistently vague, with the imagery changing rapidly from line to line. "Dust on the Wind" by the Velvet Sundown is a very good example of this, imitating the protest songs of the early 1970s while having considerably vaguer lyrics. While this alone isn't a good indicator that a song is AI, it is when combined with other tropes from this list.
  11. AI lyrics written in Japanese (and possibly other foreign languages) tend to be way wordier than lyrics from real vintage Japanese songs.
  12. For whatever reason, AI tends to use very simple rhymes and rhyme schemes. AI songs usually follow an AABB rhyme scheme and rarely use any internal rhymes. They also rarely use feminine rhymes (rhyming the last two syllables of words where the last is unstressed), almost exclusively using masculine ones (rhyming only the last syllable). Like vague lyrics, this by itself may be the result of either AI or a bad human songwriter, but it becomes a red flag when combined with other AI music tropes like misusing words and nonsensical song structure.


Return to Articles

Return to the Audio Room

Go home