Google With a Grain of Salt

“Google” might be synonymous with “online search”—but don’t blindly trust its results this presidential election season, say UMD journalism researchers.

“People question stories they hear about in the media, they question television advertisements and things the candidates themselves say,” says lecturer Sean Mussenden M.Jour. ’00. “But there’s a perception that Google is likely to deliver less biased information when that’s not the case.”

He’s working with Daniel Trielli M.Jour. ’16 on a team led by Assistant Professor Nicholas Diakopoulos to analyze Google’s search results (favored by more than two-thirds of Americans, according to comScore) for the candidates running in the 2016 primary races.

Their initial study, which surveyed the first page of search results for 16 candidates on several days in December, showed an apparent bias favoring Democratic candidates. More positive results appeared on the first page for U.S. Sen. Bernie Sanders, for example, while U.S. Sen. Ted Cruz received more negative results.

While Google is deliberately opaque with its algorithms, these results suggest that it favors official and social media pages—and candidates who aren’t prioritizing search engine optimization might be losing out. Sanders’ positive results on the first search page included multiple links to his official site, Facebook profile, Twitter feed and YouTube account. Cruz, on the other hand, had just a Facebook profile and two official website links. The rest of 
the results were news stories, which skewed negative.

Now, “the question is what kind of news does Google rank?” says Diakopoulos. “What is the distribution of support or opposition across all news articles on a candidate? If Google only knows about The New York Times, Politico and maybe 50 other outlets, that’s the news world according to Google, but obviously we know that’s not the full reflection of reality.”

Though their initial methodology required crowdsourcing the analysis of hundreds of pages and tagging them as positive or negative to establish a baseline, their goal now is to design a program to scan more pages automatically, perhaps daily or hourly. They hope to have a prototype working by the two nominating conventions in July.

“We don’t know how those positivity ratings change over time and how that relates to other measurements of public perception of the candidates,” Trielli says. “If we investigate further and see these connections, maybe Google favorability is indicative of a candidate rising in the polls in a couple weeks. If we have a tool that can map that, we can make journalists better prepared for those shifts.”