Spotify’s podcast and audiobook discovery will get a boost from Google Cloud’s AI

[Collection]

Illustration by Nick Barclay / The Verge

It’s a cliché at this point to say that podcasts (and, to a less publicized extent, audiobooks) have a discovery problem. To help fix that, Spotify announced it’s using Google Cloud’s AI tools to improve both content discovery and personalized recommendations for its audio offerings.

With this latest move, Spotify is using Google Cloud’s LLMs (large language models) to analyze the roughly 5 million podcasts and 350,000 audiobooks in its content library with a goal of trying to “augment” the metadata, according to a press release. (The Verge has reached out to Spotify to clarify exactly what “augment” means in this case). A podcast or audiobook’s metadata includes information like the title, the name of the host or author, show notes, and other details — all of which come up in the search results of podcast apps like Apple Podcasts or Spotify.

Google Cloud is Spotify’s sole cloud provider, and as Protocol reported back in 2021, the two companies have been working together on more services.

Identifying “harmful” content appears to be another aim of Spotify’s Google Cloud AI experiment, though the company didn’t specify much about what that will entail. The platform’s rules for podcasters and musicians ban several types of “sensitive” and “dangerous” content — and it currently uses a mix of automated technology and human reviewers to enforce these rules. The Verge has reached out to Spotify for more details and will update when we hear back.

Google Cloud’s AI tools will also be used to improve Spotify’s personalized recommendations for podcasts and audiobooks. LLMs will be used to “better understand patterns behind users’ favorite spoken content,” which will presumably result in better tailored recommendations.

Currently, Spotify will suggest new podcasts and individual episodes to listeners on their homescreen. On each podcast’s show page, there’s also a “More like this section” with similar podcasts. Both features can be hit-or-miss, as many users have noted. (For example, Spotify’s recommendations for Overheard at National Geographic include a podcast about nursing, one on bipolar disorder, and a true crime podcast.)

Recommended podcasts for listeners of “Overheard at National Geographic”.

Earlier this year, Spotify rolled out an AI tool for songs, AI DJ, which uses OpenAI’s generative AI tools to create custom playlists. So far, reviews of the feature have been mixed. Spotify’s use of AI for spoken audio isn’t a standalone feature like AI DJ, but it’s a sign that the company is serious about improving the user experience for its non-music offerings.