Pandora is starting a new advertising service that would allow political organizations and candidates to target the majority of the Internet radio service’s listeners based on their likely voting preferences. Pandora is doing this by comparing election results with subscribers’ ZIP Codes and musical preferences. Then Pandora is labeling subscribers’ political preferences based on their musical taste, and if an artist is more popular in a largely Republican or Democratic area.

Pandora is in fierce competition with Spotify and other Internet radio services, and believes political advertising is a unique way to increase revenue.

“Targeting users is basically the currency in data right now,” explains Jack Krawczyk, director of product management for Pandora.

“He says companies like Pandora and Facebook, which know users’ names, and can track their media consumption or stated preferences across computers, tablets, and phones, have an advantage over companies relying on Web browsing cookies,” reports The Wall Street Journal.

Pandora already began allowing advertisers to target certain demographics in November, such as subscribers who live in ZIP Codes with a large Hispanic population and listen to salsa, and mothers based on subscribers who are women in their thirties who listen to children’s music.

Pandora’s premium subscribers who pay $3.99 per month are able to opt out of politically targeted ads, while users of the free service cannot. All users are able to opt out of cookie-based ads.

“Do political leanings correlate with music preferences? Mr. Krawczyk said he believes Pandora’s predictions are between 75 percent and 80 percent accurate, but the ‘true test’ will be how the ads perform,” notes WSJ. “Pandora uses county election results in assembling its profiles because it has found they are better predictors than more-local results, such as from a voting precinct.”

So far Pandora has teamed up with two political advertising firms, Bully Pulpit Interactive and Precision Network, to test out the new advertising algorithm.