Podcast advertisers target what you actually played, not what you remember
A British data firm combines AI reading of show descriptions with tracking inside everyday phone apps to group listeners by what they actually played.
PodBiz | Where's The Money In Podcasting? · How Privacy-First Data Is Changing Podcast Ads with Emma Raz of NumberEightForget asking listeners what they heard last week. The more reliable record is what their phones already saw them play.
That is the approach described by Emma Raz of NumberEight, a British company that sells podcast audience data to advertisers. Raz says the company was born after GDPR, Europe's strict General Data Protection Regulation for technology privacy, and had to build its system under those limits from the start.
The first step is reading, not listening. Every podcast has an RSS feed, essentially a long text file with the show's titles, descriptions, and keywords. NumberEight feeds all of that text into so-called embedding models, a type of artificial intelligence that reads language for meaning, and sorts each show into contextual categories that advertisers can buy.
The second step is watching what people actually do. NumberEight puts an SDK, a small bit of software, inside ordinary publisher apps where people catch up on news, stream songs, or play podcasts. That software can draw on hardware sensors in the phone, such as the gyroscope and the accelerometer that sense motion and rotation, along with location when it is on. Raz says the apps are not only podcast apps, and can include something as everyday as a news reader, with permission covered in the app's terms and conditions and with users logged in.
From those apps, the company gets three things. It infers behavior, such as whether someone is a regular shopper, an international traveler, or a desk worker. It receives basic demographics, such as age and gender, that the user gave the publisher. Most important, it records real content consumption, every podcast, stream, or article that was actually opened.
Put together across millions of users, those observed plays form what Raz calls podcast neighborhoods, groups defined by what people truly consumed. They are much larger than a survey sample, and more trustworthy, because seeing beats remembering.
Questions people ask
Is this kind of in-app tracking legal if I never heard of the data company?
Yes, when it is disclosed and the user agrees. In both Europe and the United States, apps are generally required to explain in their terms and privacy policies what data they and their partners collect, and users must accept those terms and can often opt out of tracking in phone settings.
Do advertisers get to see exactly what I listened to?
No. Advertisers buy grouped audiences, like regular shoppers who listen to business shows, not a named person's playlist. The system is designed to share patterns across millions of users for ad targeting, not to hand a brand your individual listening history.
Why is watching behavior better than just surveying listeners?
Memory is unreliable and samples are small. People forget titles, confuse shows, or say what sounds impressive, and a few thousand survey answers cannot represent millions of listeners. Observed plays are automatic, detailed, and continuous, so they capture habits people would never report accurately.