More than music, why Spotify AI is the secret to listener retention

Streaming music apps have ushered users into the era of artificial intelligence, with a limited track record of success. However, AI-based recommendation tools offered by Apple, Amazon, and pure streaming companies spotify Spotify’s latest approach to the future of personal music discovery is moving forward, leaning into AI prompts in multiple formats. Experts say these technology investments could be critical to Spotify’s ability to build a moat around its business as its core input, music, becomes commoditized across streaming apps.

Spotify recently announced a new ChatGPT integration that allows users to connect their accounts directly to OpenAI’s generative AI chatbot. This announcement bodes well for OpenAI in its broader efforts to turn ChatGPT into a platform for third-party apps that work within conversations. For Spotify, it’s betting that personalized music and podcast recommendations will improve through the now-familiar format of chatting with an AI and letting it know what you want.

Spotify users can request songs, artists, albums, playlists, and podcast episodes by mood, genre, or topic. The results will appear within ChatGPT and can be opened and played in the Spotify app. Users can manipulate recommendations to provide more specificity than is possible with traditional “like/dislike” feedback options.

According to a Spotify spokesperson, the prompts are “an opportunity to discover new tracks, revisit old favorites, and expand your ChatGPT conversations with a soundtrack that fits the moment.”

Spotify said the integration is opt-in and users can opt out at any time. It also said it will not share music or podcast content with OpenAI for training purposes to address industry concerns about AI and copyrighted material.

Spotify recently introduced a prompted playlist feature within its streaming music app. This is a “vibe” feature that allows users to build custom mixes using emotions and memories.

Rival streaming services connected to big tech companies are exploring similar AI capabilities.

apple has been layering AI into Apple Music in stages. The Playlist Playground beta feature is closest to what Spotify is doing, and also focuses on chat-based AI interactions that allow users to fine-tune recommendations through chat. Apple recently introduced AutoMix, which uses AI to analyze songs and automatically blend them together by matching tempos and beats, removing silence between songs, adding crossfades, and more. The company also develops machine learning tools such as lyrics translation and pronunciation functions.

Amazon Music has been offering a prompt-based playlist feature called Maestro since mid-2024. This allows listeners to generate playlists with text descriptions and emojis. It remains in beta testing rather than a full release.

Spotify executives have repeatedly described AI as central to the platform’s subscriber retention strategy. During a recent earnings call, management told investors that improving AI-driven discovery is central to keeping users engaged with the platform. “Our investments in personalization and AI are paying off,” said co-chief executive officer Alex Norstrom. “This means people are spending more days and spending more time with us during the month,” he said.

Spotify’s interactive iDJ feature, introduced in 2023, has around 90 million subscribers as of its latest earnings report, with users spending 4 billion hours on the app. Norström said Prompted Playlists “will be quickly adopted by power users.”

“If iDJ is a chat interface to Spotify and a place for casual conversation, then prompt playlists are Spotify’s deep research mode,” he said. “It allows you to write and configure the rules for your own personalized playlists. You’re literally writing your own algorithms. … There’s nothing else like it.”

Music catalog becomes commoditized, AI generates millions of songs

Analysts covering Spotify say management’s hype about AI may need to become reality for the company sooner rather than later. While there are occasional cases where musicians pull their music from a particular app as negative headline issues emerge (as happened most recently with Spotify, thanks to founder and former CEO Daniel Ek’s investments in defense technology), competitors like Apple Music, Amazon Music, and YouTube Music offer nearly overlapping catalogs and increasingly sophisticated recommendation engines.

“In the same way that Bing and Edge are almost identical to Google, Amazon, Apple and YouTube have similar catalogs to Spotify, almost identical songs,” said Michael Pachter, senior advisor for digital media, sports and entertainment at Wedbush Securities, a longtime research analyst covering the streaming industry. (Mr. Wedbush has never given a specific rating to Spotify stock.)

While Google’s search business faces AI threats, Pachter said Spotify is also the best model to look at in terms of how to maintain the user edge. “Google has managed to expand its reach by offering a number of features that make its service more reliable, such as remembering credit card and password information. I can’t imagine switching away from Google Search, and I think that’s what Spotify is trying to establish,” he said.

Switching costs may be small, but in some cases they can be significant. Users spend years building libraries, curating playlists, and training algorithms. Each additional integration with a car dashboard, voice assistant, or now an AI chatbot (which Spotify says can now connect to over 2,000 devices) further strengthens the ecosystem.

“We expect this ChatGPT integration to be widely used by Spotify users and to be a huge success,” Pachter said. “Other companies may try to do the same thing, but every time you try to build a playlist on Spotify, the switching costs increase. They expect it.”

Apple Music and other third-party apps offer tools to export playlists when subscribers want to change music services.

Others on Wall Street are less convinced than Pachter, but in recent quarters they’ve been upbeat about the Spotify AI story and less worried about the risks facing platform-disrupting AI music creation tools like theirs. Although Spotify’s stock price has fallen nearly 20% over the past year, the stock has performed extremely well since its IPO in 2018.

“Spotify has addressed this concern head-on, insisting that AI supports rather than detracts from its strategic position. Spotify appears positioned to use AI to enhance its platform by leaning toward personalization, product innovation, and economies of scale, but the pace of adoption and industry alignment will remain important variables,” Bank of America’s research team, which rates the stock as a buy, wrote in a February note after its most recent earnings.

Spotify Co-CEO Gustav Söderström said on an earnings call that by building music apps that allow users to have conversations and fully understand each listener, listening will move “from a passive experience to an interactive one.”

Mark Mulligan, managing director and analyst at MIDiA Research, a research firm that tracks the music market, said that while AI will be integral to streaming music behavior, he is less convinced that the interactive-passive distinction is a likely outcome of what Spotify will do.

“Streaming music consumption has become polarized between passive and active consumption,” Mulligan said. “But this doesn’t mean the audience is split; everyone is. Even the most active music listeners spend more than half their time passively listening.”

In fact, he says the broader trend is toward more passive consumption through curated playlists and features like artist radio stations and AI DJs. “The direction of the journey is toward more passive listening,” Mulligan said. Agent functionality could represent a compromise between “passive listening and active listening,” he said. “Instead of a large amount of ‘leaning back’ listening, users can expend a small amount of ‘leaning forward’ effort.”

While typing detailed prompts into ChatGPT may feel active, Mulligan says, “The more the algorithm learns about your listener’s behaviors and preferences, the better recommendations it will make, so you don’t have to lean in, and the needle shifts further toward passive listening.”

In this AI interface-first streaming model, the underlying content is important, but what the user ultimately finds rewarding is less important. For example, the ability to explicitly exclude artists or filter by subgenre can make AI-assisted discovery feel more customized than traditional algorithmic playlists. If a listener likes 1980s rock bands like Bon Jovi or Guns N’ Roses, but hates other bands from the same era, it’s easy to filter. Spotify can usually predict that if you like A and most people who like A also like B, you’ll probably like B too, but that doesn’t necessarily represent how your preferences are expressed. “With GPT, you could say, ‘No Def Leppard,’ and they’d take you off the list,” Pachter said.

As with any industry, predictions about how AI will impact music are just educated guesses. But it’s already clear that AI is influencing the way we think about the music catalog itself. A recent report from Rothschild & Co Redburn reported that text-to-music conversion platforms like Suno generate around 7 million songs per day, roughly equivalent to Spotify’s entire fortnightly pre-AI catalog. “This is a deluge,” said analyst Ed Vivian.

Söderström hinted that what matters most for the future are the upcoming datasets, rather than the deep tracks already in the stack. “We’re building a data set that didn’t exist before,” he said on a recent earnings call. “We had a song-to-song dataset, but no one had a language-to-song dataset. … You might think this is a canonical dataset. So there’s no factual answer to, for example, what is training music? … On average, for Americans, it’s usually hip-hop. For Europeans, it’s usually EDM. For a lot of Scandinavians, it’s heavy metal or death. It’s like metal. Again, for many Americans, at least millions of Americans, it’s also death metal.”

“You can’t just have LLM commoditize it as a fact the same way you commoditize Wikipedia,” he said. “The reality is that you need hundreds of millions of listeners in markets around the world to constantly communicate what it means to that particular person.”