A Memorial Day for Music

Last week I noticed something off on my Discover Weekly. A song came on that wasn’t bad. It was, somehow, a little too good. Clean, hooky, perfectly mixed, and completely soulless. Like the title track from “That Thing You Do!”, the 1996 Tom Hanks movie about a manufactured one-hit-wonder band. A song engineered to sound like a hit, not to be one.

I opened the artist page. No interviews, no live videos, no photos that weren’t suspiciously clean studio shots. Around 800,000 monthly listeners and not a single concert listed anywhere on the web. I’d been served an AI artist, and the algorithm liked that I hadn’t skipped.

That’s the problem in one paragraph. Spotify won’t tell you when a track is generated, and once you’ve listened to one without flinching, the algorithm assumes you want more.

There’s no label, by design

Spotify rolled out an AI disclosure tag in April 2026. The catch is that it’s voluntary, mobile-only, and buried inside the Song Credits view. Spotify openly states that the absence of a tag does not mean a track is AI-free. Artists who don’t want to disclose simply don’t.

Apple Music, YouTube Music, and Pandora aren’t meaningfully better. Apple rolled out mandatory disclosure tags in March 2026, but enforcement runs through distributors and there’s no user-facing filter. YouTube Music requires disclosure too, and its users are loudly complaining that the algorithm keeps pushing AI tracks at them no matter how often they thumbs-down. Pandora hasn’t published a public AI policy at all; meanwhile its users are openly asking the company for a way to filter AI artists out. The big four have all decided that disclosure is the artist’s job and filtering isn’t your right.

Two smaller services drew a different line. Deezer automatically detects AI tracks, labels them, and excludes them from algorithmic recommendations by default. Qobuz published an “AI Charter” committing to 100% human-curated recommendations, with industrially generated AI content kept out of playlists and featured sections. Both treat AI music as something a listener should be able to opt out of. The majors treat it as something an artist might choose to disclose.

The result is a watermarking gap. There’s no consistent, audible or visible, platform-wide signal on the services most people actually use that says “a human made this.” Until that exists, you’re guessing.

One play poisons the well

Spotify’s recommendation system weights early skips heavily. Skips before the 30-second mark are one of the strongest negative signals it tracks, and the 30-second mark also happens to be the threshold for a play to count as a stream at all. Listen to a song for 31 seconds and you’ve cast a vote for more like it.

What this means in practice: if you let an AI track play past the half-minute mark, the algorithm marks it as a positive signal. It doesn’t know the track was generated. It just knows you didn’t bail.

Now multiply that across a Discover Weekly with one or two AI artists slipped in. You don’t skip. The next week’s recommendations lean a little further in that direction. Within a month, a meaningful slice of what’s being surfaced to you is synthetic, and Spotify can honestly say they’re just giving you what you listen to.

Why the toggle doesn’t exist

This is where I get cynical. AI tracks are cheap to acquire, pay the same per-stream royalty as human-made music, and pad the catalog at near-zero cost. A toggle that filters them out is technically possible. Deezer built one. Spotify has not, and has not announced plans to.

The official framing is that defining “AI music” is hard, that disclosure should come from artists, and that an industry metadata standard called DDEX will sort out transparency over time. The unofficial reading is that the platform with the largest catalog and the thinnest margins has no incentive to help you remove a growing chunk of that catalog from your ears.

The EU AI Act forces labeling of AI-generated content starting August 2, 2026. Spotify’s compliance plan is, as of this writing, unclear. I’d bet on the bare minimum: a disclosure tag for EU listeners, no toggle, no recommendation filter.

What you can do on Spotify right now

If you’re sticking with Spotify, three things move the needle:

1. Skip suspicious tracks before the 30-second mark. Early skips are one of the strongest negative signals in the system. Skip late and you’ve trained the algorithm in the wrong direction.
2. Hide songs on algorithmic playlists. On Discover Weekly, Daily Mix, or Release Radar, tap the ... menu and choose “Hide song.” That’s a stronger explicit signal than a skip.
3. Don’t play this artist. On an artist page, the same ... menu has “Don’t play this artist.” Use it on accounts that look generated.

Look for the tells: stock or AI-generated profile images, no interviews or live performances anywhere on the web, suspiciously high monthly listeners with thin social presence, and a genre that’s almost but not quite a specific human subgenre.

Next time you get an unknown artist on Discover Weekly with a polished sound and no backstory, open their page. If you can’t find a single interview, music video, or live performance, that’s your answer. Skip fast, hit “Don’t play this artist,” and ask whether the platform you’re paying for is on your side.

Want to test your own ear? Sightengine built a quick game that plays you snippets and asks you to call them: human or AI. It’s harder than you’d think. That difficulty is the whole problem with leaving disclosure up to the artist.

Happy Memorial Day. Spend a little of it listening to humans.