As happens with social media data, reporters often use Google Trends to imply more than should be inferred from this tool, while omitting important caveats. Over the past several years, I have worked to better understand what Google Trends can really tell us about political events and distinguish between the questions it can answer and those for which it is of little or no help. Here's what I've learned.
What is Google Trends data?
Google Trends uses a scale from zero to 100 to illustrate how much people use Google to search for information about a chosen topic over a given period. Journalists often use Google Trends to compare different topics with one another in the same chart. Thus, this tool doesn’t just reveal how much one topic was searched over time; it enables users to find out whether a specific topic — for example, a presidential campaign — was searched more often than another one was.
Google Trends scores are based on search volume data — but they give no indication of the absolute volume. In other words, from Google Trends data alone there is no way of knowing whether a surge in the number of searches for a given topic represents dozens, thousands or millions of new searches.
What can Google Trends tell us about elections?
Despite these limitations, Google Trends can offer very valuable information on searches associated with key political events, particularly TV debates. First, it indicates when Google users are particularly likely to search for information about a certain candidate. This makes specific search peaks more interesting and potentially more useful than a candidate’s average Google Trends score during an entire debate. Campaign managers may find these search peaks offer particularly useful insights, such as whether to focus more closely on the topics that generated more searches in the days following the debates.
According to my research, unusual, dramatic or particularly entertaining episodes during debates tend to generate outsize Google Trends peaks. Due to the specific nature of these search triggers, Google Trends peaks do not equate directly to an increase in interest for a candidate. Rather, they are better interpreted as a sign of “curiosity” that may or may not generate ongoing interest.
Consider an example from the second night of the first 2020 Democratic debate on June 27. That night, Sen. Kamala D. Harris (D-Calif.) and author Marianne Williamson generated some of the biggest Google Trends peaks. Harris was searched during a heated exchange with former vice president Joe Biden; Williamson was searched during the five total minutes of airtime she received. After the debate, Harris rose noticeably in the polls. Williamson did not — showing that searches for her reflected mere curiosity for a different type of candidate.
Similarly, research I carried out about the 2012 election debates showed that Republican candidate Mitt Romney’s “binders full of women” gaffe gave him a clear edge in Google searches. But that curiosity did little for him; he went on to lose the election.
What are the limits of Google Trends?
Google Trends data cannot be used to replace opinion polls or, worse, predict election results. To do this, we would need to assume that all Americans use Google for political searches in exactly the same way, which is obviously unrealistic. In addition, Google Trends data offers no real information about exactly how much support exists for a certain candidate. For instance, a great deal of media coverage noted that, during the first 2020 Democratic debate, Google searches for Julián Castro increased by 2,400 percent. Impressive as this may seem, it actually means very little. That’s because we don’t know how many searches Castro started from — and therefore how many he was getting when the night ended.
Google Trends also offers no information about search results and, therefore, tells us nothing about the sites that users are most likely to click on. While some of the reasons people are searching can be inferred from search timing, where they go after those searches and what they do with that information remains unclear. Google Trends’ list of “related queries” can suggest some of the next steps people take. For example, my work on the 2013 Italian elections showed that people who searched for the populist Five Star Movement were especially likely to also search for its website and video-streaming channel instead of, say, its Wikipedia page — thus suggesting that users were genuinely interested in its politics. However, we rarely get such specific and informative related-search information. Typically, we know little about exactly what information users take away after completing a Google search.
How does Google itself use Trends?
Here's a final important point: Google itself tries to use Trends data to present a certain narrative from key election events such as TV debates. Since 2015, Google has used the @GoogleTrends Twitter handle to publicize important search trends, posting frequently during and after each debate. In addition, Google has also set up a Trends blog about the 2020 election, which it uses to report on the debates in a “horse race” fashion, using dynamic charts and highlighting key search trends in early primary states including Iowa, New Hampshire, South Carolina and Nevada.
This information constitutes not only an influential resource to be seized upon by reporters, but also a data showcase that’s probably intended to interest campaigns in collaborating more closely with Google’s election team.
As the 2020 campaign heats up, it is important that audiences — both lay and professional — fully understand the meaning and limitations of Google Trends data, which shapes so much relevant coverage.
Filippo Trevisan (@filippotrevisan) is assistant professor in the school of communication at American University and author of “Disability Rights Advocacy Online: Voice, Empowerment and Global Connectivity” (Routledge, 2017).