Skip to navigationSkip to content

Facebook ads reveal what the Democratic candidates think will sway South Carolina voters

From left, Democratic presidential candidates, former New York City Mayor Mike Bloomberg, former South Bend Mayor Pete Buttigieg, Sen. Elizabeth Warren, D-Mass., Sen. Bernie Sanders, I-Vt., former Vice President Joe Biden, Sen. Amy Klobuchar, D-Minn., and businessman Tom Steyer participate in a Democratic presidential primary debate, Tuesday, Feb. 25, 2020, in Charleston, S.C.
AP/Matt Rourke
The candidates have issues.
By Jeremy B. Merrill
AtlantaPublished Last updated This article is more than 2 years old.

The Democratic candidates have different pitches for South Carolina voters ahead of tomorrow’s primary: Buttigieg is focusing on gun control. Sanders talks about the economy and healthcare. Warren, of course, is focusing on the economy, as is Tom Steyer. Biden and Klobuchar mostly push their ability to beat Trump.

And Michael Bloomberg, who has deluged just about every medium in much of the rest of country, is sitting out the South Carolina primary but has nevertheless has spent about $76,000 on ads mostly about gun control and education.

That’s based on a Quartz review of 68,000 Facebook ads from Feb. 12 to 25. We didn’t read all those ads, but instead trained an artificial intelligence algorithm to tell us what all those ads were about based on the text within the advertisements.

Here are examples of each candidate’s ads promoting their top topic:

An ad from Pete Buttigieg on gun control and healthcare
An ad from Bernie Sanders about healthcare and the economy.
An ad from Elizabeth Warren about the economy.
An ad from Tom Steyer about the economy.
An ad from Joe Biden about beating Trump
An ad from Amy Klobuchar about beating Trump

Tom Steyer has spent so much money in South Carolina that if you examine just the estimated number of ad impressions in the state, they drown out the stats from other candidates.

Quartz is using AI to help us cover online political influence, and in this case used an AI-based classifier to “read” the text and guess which topics are discussed in each ad. Unlike algorithms based on pre-defined keywords, this one is more powerful because it can learn new keywords. For instance, even if we didn’t tell the model that the term “GI bill” relates to both education and veterans, it will detect that the phrase occurs in ads with words like “veterans” and “college tuition” and learn what “GI bill” means. That can also lead to hiccups: for some reason, our model thinks that the word “Nevada” means an ad is about veterans.

Also, the text inside Klobuchar’s and Biden’s ads don’t dive particularly deeply into individual issues—though both run ads that discuss issues within videos, which our AI can’t read.

Be the first to know.

Sign up for the Quartz investigations email and get updates 2-3 times per month.

📬 Kick off each morning with coffee and the Daily Brief (BYO coffee).

By providing your email, you agree to the Quartz Privacy Policy.