Apple Music Playlist Playground: An AI Playlist Generator That Doesn't Listen to the Prompt

Apple Music has rolled out Playlist Playground, an AI-based playlist generator meant to pick music from text prompts. The feature is in beta, and early tests show it performs noticeably worse than its competitors.

An editor at The Verge asked the service to build a playlist of "atmospheric instrumental black metal for work." The result was telling: three metal tracks with vocals, one field-recording sound experiment, ambient electronica and a slice of doom jazz. Not a single track matched the request exactly. The prompt said nothing about vocals, yet Apple Music added them. "Instrumental" means no words, but the algorithm ignored that condition entirely.

For comparison, the same prompt was given to YouTube Music. Its AI stumbled only on the fifth track, with one vocal track, while the first four fit the genre strictly. It's not perfect, but it is substantially better than what Apple offered. With YouTube Music, the error was the exception to the rule; with Apple Music, it was the rule itself.

Playlist Playground arrived with the iOS 26.4 update and is available to Apple Music subscribers. Users type a text prompt (a mood, genre, activity or image) and get an automatically generated playlist. The idea isn't new: similar features already run on Spotify, YouTube Music and Amazon Music. Apple entered this niche later than its rivals and, judging by first impressions, its algorithm is not yet where it needs to be.

The problem goes beyond this one black metal test case. Users online share similar stories: ask for instrumental and get vocals, ask for quiet background music and get loud rock. The algorithm seems to rely on broad genre tags and ignore the nuances of the request. "Instrumental," "atmospheric" and "for work" are concrete semantic constraints that an AI system is obliged to respect. For now, Playlist Playground doesn't.

For the past two years Apple has actively positioned itself as a company betting on personalized AI, and Apple Intelligence is one of the central themes in its marketing. But a gap still separates the promises from the practical results. Music recommendation is one of the hardest cases for AI: it takes more than genre classifiers, requiring an understanding of the prompt's semantics, the listening context and fine subcategories within a genre.

There is also the question of what counts as success in such systems. If the generator's goal is to offer something "close enough," the algorithm may treat black metal with vocals as an acceptable answer to a request for black metal in general. But a user who explicitly asks for instrumental puts that very word first. This is where AI recommendations still lose to a human curator: they hear the words "black metal" but not the words "no vocals."

The feature is still in beta, and Apple could in theory refine it before a wide release. Even so, the start was awkward: a direct competitor handled an identical prompt noticeably better, and users noticed. For a company that builds its image on precision and quality of user experience, this is a signal worth heeding.