Google Screwing Trump Through More Than Just AutoComplete

If you Googled any of the U.S. presidential candidates over the past four months, you probably noticed an information box displaying their stances on specific political issues.

It seems impartial enough: a list of topical info cards with issue-related quotes from the candidates, mostly pulled from news articles. But at the Computational Journalism Lab at the University of Maryland, we are investigating how bias in search engines can affect the political process, and we started collecting data on this new info box in April. Our analysis of data collected from the guide shows significant differences in the way candidates are portrayed. Among the issues: unequal space given to each candidate, the use of quotes that don’t really explain their stances, and overreliance on a limited set of news sources.

This matters because search engines are powerful brokers of political information that can shape an electorate’s perceptions. Political biases can emerge from the complex and opaque search algorithm, surfacing more or less supportive or critical information about candidates. With the issue boxes, Google is taking a more direct approach to curating information about the candidates. And, as we found, that comes with its own set of unique biases.

The first concern is the way the guide lists the issues. It’s not alphabetical; the first three issues for each candidate are immigration, abortion, and guns. And it doesn’t align with the public’s assessment of the importance of each issue, at least comparing it to polls conducted by Gallup and Pew. If it did, the economy would be much higher up. This matters, because items listed higher on the page will get more attention from searchers.

According to a Google representative, issues are defined and ranked using “several factors,” which include the way other websites and organizations do it (the representative specifically mentions Pew, Vote Smart, or OnTheIssues.org) and user interest. We tried, unsuccessfully, to replicate the issue ranking in Google Trends, which indicates the frequency of searches of specific terms through time. For instance, the first two items in the issues guide list are “immigration” and “abortion,” but in Google Trends the user interest is higher for “guns” and “taxes.”

The list is relatively static, suggesting that any data-driven techniques to order the issues are moderated through human editing or approval, not unlike the Facebook Trends process that has recently come to light.