Spotify’s AI-powered DJ now takes song requests

Since it launched two years ago, Spotify’s AI DJ has been a one-way experience. It curates old favorites and helps listeners discover new tracks based on past listening experience and what similar users like. But now it’s getting interactive. 

Spotify unveiled the ability to request songs from the DJ based on mood, genre, and vibe. The feature, which launched across 60 markets, is exclusive to Spotify Premium users, who can access the DJ by searching for the tool in the app. 

It’s the latest AI feature to come from Spotify, which introduced an AI-generated playlist builder for Premium users in the United States last fall. But Molly Holder, Spotify’s senior director of product for personalization, says that tool was designed for people who want to take an active role in their listening experience. The new DJ request feature, by contrast, is designed to give users more input into an essentially “lean back” experience.

“We know that even in a ‘lean back’ moment, users still want some semblance of control,” she says.

How does the DJ request feature work?

After a user searches for and calls up the DJ in the app, they can make a request by holding down the DJ icon and speaking a prompt. (The feature requires microphone access.) After receiving the prompt, such as “upbeat songs for running” or “ambient for a rainy day,” the DJ will “think” for a bit before launching into a tailored playlist. 

Focusing on mood or moment along with a genre tends to be the best approach for using the feature. Wanting to give it a slam dunk, I asked for “early 2000s patriotic country” and immediately got Toby Keith’s “Courtesy of the Red, White and Blue.” Not too hard. But when I asked for “undiscovered African gems for a summer barbecue,” expecting Afrobeats and Amapiano, the DJ instead served songs from the African diaspora. (The first was by Timbuktu, who Spotify describes as “one of Sweden’s most well-known hip-hop artists.”) Once I specified the genre, the results were better, but I’d forgotten to make clear I wanted undiscovered artists, so I got a lot of Tyla.

With several of the requests, the DJ seemed somewhat buggy, abruptly stopping a song and launching into its more default mode. When I asked for “DIY indie rock from the mid-2010s,” it played a few bars from an early Mitski song, then reset itself and introduced my top songs from 2023.

Pairing insight with a human touch

Spotify’s Holder says that while the DJ’s content-surfacing abilities are powered by AI, many of the comments and insights that it delivers in between songs—the bits that make the tool feel like a personalized and expert experience—are powered by humans. 

“We have music experts and editors who have built what’s called the ‘writer’s room’ for the DJ product,” she says. “These folks come together with generative AI to build the commentary that DJ offers throughout the experience.”

This commentary can increase listener engagement—which can be beneficial for artists who are trying to break through as well. “In that writer’s room, when we build that commentary, it helps to bring artists’ stories to life to bring a little bit more context to the content we’re recommending,” Holder says. “One thing we see is that users who hear commentary about a track in the DJ experience are more likely to listen to a track they might have otherwise skipped.” 

Holder also positions the new feature in Spotify’s long history of using machine learning to build personalization into the user experience, from Discovery Weekly to Daylists and now the DJ requests. “We’re really applying AI strategically across our product portfolio in ways that enhance the value proposition of our products.”

View the full article