Can Google predict the future?

It’s hard to imagine going a day without Google. Directions to a restaurant, a recipe for dinner or a search for a news story you’re curious about – Google’s resourcefulness has become ubiquitous with modern living. There’s a reason ‘Google’ is a verb recognized by the Oxford English Dictionary. While innocuous online interactions might seem forgettable to the individual user, collectively, they can provide deep insight, signal larger trends and even predict offline behaviour.

The first iteration of Google Trends was launched in 2006, and has been continually upgraded since. Its current version allows users to track news stories and the words and phrases used in searches, breaking them down by time and location. Google Trends also provides data on the type of search (web, image, news, Google Shopping or YouTube) and related top and rising topics and queries.

This resource has inspired researchers of all kinds to apply its open data to their fields of study. Sociologists have used to it examine parents’ curiosity surrounding their child’s weight or intelligence, while investors have tried to find connections between behavioural changes with searches and stock prices. News editors and producers can use the tool to map out how stories and pop culture trends are resonating. After Donald Trump’s success on Super Tuesday, The Globe and Mail, as part of its analysis, reported that Americans searching ‘moving to Canada’ spiked.

In all cases, this data enables researchers or curious members of the general public with the ability to monitor and understand regional breakdowns and related searches. The ムmoving to Canada’ sentiment was most popular in Norfolk, Virginia, and related searches mentioned Donald Trump or Raven Symone, a celebrity whose musing on the subject grabbed headlines. Further, the ムmoving to Canada’ search has skyrocketed since the beginning of February and the volume is still strong moving into March.

Beyond exploring trends, can Google actually forecast the future? Could it, for example, predict the outcome of an election?

Recent updates to Google Trends now provide enough granular, real-time data to allow users to track behaviour in key regions and correlate it with expected electoral outcomes. The New York Times recently examined Google searches in states prior to their caucuses and primaries, finding that electoral results in New Hampshire, South Carolina and Nevada closely aligned with their respective shares of search volume.

Obvious doubts about this predictive model persist: just because someone Googles Donald Trump, doesn’t mean they’ll vote for him. However, with recent data showing the power earned media coverage can have on popular appeal, it’s not hard to see why someone’s interest in a candidate on Google can correspond with real-life support at the ballot box. Here in Canada, Google Trends accurately forecasted Justin Trudeau’s win in October. While Stephen Harper began the campaign as the most searched leader, Trudeau ended the campaign far ahead of his Conservative and NDP counterparts. Notably, despite leading public opinion polling in August, Thomas Mulcair remained a distant third in Google Trends for the duration of the campaign, forecasting his party’s relegation to third party status.

While promising and certainly interesting, Google Trends creates what some experts call ‘Big Data Hubris’, the false assumption that new tools can replace, rather than complement, traditional research methods. One notable example of this is Google Flu Trends. First launched in 2009, it tracked regional searches of flu-related terms and was able to predict outbreaks two weeks before the United States Center of Disease Control was able to do so. However, shortly after, this predictive model was found to be deeply flawed, as it was unable to predict the swine flu epidemic and missed the peak of the 2013 flu season by 140 per cent.

But why was Google so wrong? Simply put, because humans are the ones doing the searches. In understanding how Google Trends failed, researchers concluded that people weren’t as sick as they thought. Indeed, only 8.8 per cent of people who exhibit flu-like symptoms actually have the virus – explaining why doctor’s offices are unnecessarily clogged during flu season and a user who Googles flu remedies isn’t necessarily inflicted.

Until the crystal ball is perfected, amateur and professional researchers alike will experiment with and refine methodologies. Sourced data will still require human analysis, extraction and perspective for proper qualification and application. At Navigator, we use a library of digital tools to augment our traditional research. While social media listening, Google Trends and other online analytics do not replace our quantitative and qualitative research methods, they do provide insight into how our client’s issues are being consumed and discussed in the age of Twitter and the 24/7 news cycle — insight that many can ill afford to live without.