“We constantly hear from podcast listeners that they have one or two podcasts that they like but can’t find any other ones like them—this is the problem that we’re solving,” Peter Birsinger, Podible’s chief technology officer, wrote in an email.
Founded in 2017, New York-based Podible has been working to crack the recommendations game from its earliest days. To improve its selections, Podible picks new podcasts for listeners based on an algorithm that culls data from their listening history, including played episodes, saved episodes, and followed shows. The company’s recommendation engine starts by transcribing the podcasts that users listen to, because when it converts audio to text, its AI can mine the data to make better, more personalized recommendations. It then extracts topics from each podcast episode based on the transcripts and adds the data to a “Podcast Genome,” which is a massive database that shows how podcasts intersect.
The Podcast Genome drives the recommendations, allowing the company to make recommendations of either a single episode or an entire show, based on past listening experience. “The more a listener uses Podible, the more data we have and the better the recommendations we can provide,” explained Birsinger.