You Can Now Use AI to Create Playlists on Spotify

There's a new AI Playlist feature in Spotify. Credit: Spotify

You may have noticed generative artificial intelligence making its way into one or two apps and gadgets in recent times, and you can now use the tech to create new playlists in Spotify: The AI Playlist feature is now live in the U.S. on Android and iOS, after testing in the UK and Australia, though you do need to be a Premium subscriber to get access to it.

This works in the same way as AI text and image chatbots, in that you come up with your own text prompt that forms the basis of a new playlist—it might be "songs for a wistful mood" or "songs for a birthday party with a bunch of 5-year-olds." Here's how to get started with the feature, plus some old-school music discovery ideas that might work better.

Creating AI playlists in Spotify

This new Spotify feature is another reminder that the term "AI" is a broad one. The algorithms that already power your recommended playlists in Spotify also rely on artificial intelligence of a different type—but this new AI Playlist tool takes the idea up a notch. Instead of just seeing songs linked to the music you currently listen to and like, you can be much more specific about what you want.

"The more specific you can be, the better," Spotify says. "The most successful playlists are generated with prompts that contain a mix of genres, moods, artists, or decades, so get creative! You can also reference places, animals, activities, movie characters, colors, and even emojis." That gives you a lot of scope to work with.

However, there are some guardrails in place: You're not able to create prompts using references to brand names or current events. There are also safeguards in place to cut out offensive prompts, so it's probably not a great idea to try and test the boundaries of decency with this new tool.

Assuming you're a Spotify Premium user, open up the app on Android or iOS, then go to the Your Library tab. Tap the + (plus) button in the top right corner, and pick AI Playlist from the menu that pops up. Now you have to get creative: Describe what you want from your new playlist with as much detail as you can. You can base it around moods, events, genres, your favorite artists, a time of day, an activity, or whatever you like.

You'll also see some suggestions above the text entry box, which you can use if you're not getting any inspiration for your own prompts. Swipe left to see more recommendations—you'll see you're able to limit your playlists to a particular duration or to a particular decade of music, for example, which then might give you more ideas about how to generate your own custom playlists.

Submit your prompt, and a playlist appears. You can refine it further using the prompt box at the bottom of the screen, tap on any song to hear a preview clip of it, or swipe left on a song to remove it from the list. When you're ready to accept the list of songs you've been given, tap Create to save it.

AI playlists vs. traditional music discovery

Based on my limited time testing the AI Playlist feature, the underlying algorithms don't seem to be too different to those deployed across the rest of Spotify (when you ask it to automatically fill in a playlist, for example). Mention "indie rock," and you'll get a lot of the same songs as you would with a general search. However, the text prompt interface gives you more flexibility when building a playlist.

This is AI, so of course there are misfires (Spotify has stuck a "beta" label on it too). When asked for "songs like R.E.M. that aren't R.E.M.," the AI returned—you guessed it—a playlist that included some R.E.M. tracks. Some refining is often required, as is usually the case when producing text or images with generative AI, and the ability to add to prompts can be useful.

It seems that AI Playlist does take your previous listening history into account when making playlists, and it will mix in songs you've heard before with tracks that are popular elsewhere. The prompt "upbeat songs for a long winter drive" returned tunes from artists including Future Islands, Taylor Swift, Coldplay, and the 1975—so a mix of stuff I've already heard and stuff I haven't.

This is certainly an interesting idea, and even if the playlists aren't massively different from what you can already find by digging through the recommendations on Spotify, it's fun to play around with certain ideas: Prompts like "songs to help me forget the past" or "the perfect playlist for gazing out across a rainy cityscape," for example (the latter returned Bon Iver, Tracy Chapman, and Radiohead, by the way).

As with much of AI, how this actually works isn't clear, but I can't help feeling a little turned off by algorithm-based, AI recommendations in general. They're based on a lot of data points, sure, but they can often lack the serendipitous or left-field discoveries that used to come from picking up a music magazine, flicking through music video channels, or getting mixtapes made by friends.

Personally, I still prefer the old ways of discovering music, including reading album reviews and band interviews, and tapping friends for recommendations (the Blend feature in Spotify is actually really good for this). Or, to give another recent example, checking out bands that are supporting bands I already like—a choice curated by human beings and their tastes, which often leads to some new favorites.