For the past ninety years, election forecasters have had one tool in their toolbox: surveys. After the third consecutive US presidential election, in which this methodology underestimated support for Donald Trump, there are reasons to doubt whether this is still useful. It seems anachronistic to call up a small group of people and ask them what they're going to do in a world where tech giants collect billions of online data points to predict consumer actions – giants that often know consumers better than they do.
Consider the problems with polls. First, people don't answer them. The response rate to the survey has fallen to just 2%.
Second, people tamper with it. Younger people especially enjoy giving wrong answers. One academic study found a link between adoption and several problematic behaviors in survey data; When it turned out that 19% of those who said they were adopted were just joking, the survey was withdrawn.
Third, people lie to pollsters to protect their self-image, something called social desirability bias. The level of deception can be shocking. Research shows that roughly four in ten people who did not vote in an election will report in surveys that they did so. People are also known to dramatically overreport how much sex they have, their propensity to give to charity, and their academic performance.
These non-respondents, cheats and liars likely all contributed to Trump's better-than-expected performance. There are indications that his supporters are less likely to answer surveys, more likely to fiddle with them and less likely to admit they support him.
Does this make surveys useless for predicting what will happen in elections? Not quite. As noisy and flawed as they can be, surveys do contain useful information, especially in helping us understand where support for candidates might be changing. I compared Trump's actual performance to that predicted by FiveThirtyEight, a poll aggregator. About half of the state-level change in support for Trump between 2020 and 2024 could be predicted by surveys. Surveys rightly showed that his vote share would skyrocket in Kentucky, New York and Massachusetts. And there are trends that are simply overlooked. Some seemingly surprising election patterns, such as Trump's strong performance in heavily Hispanic districts, would have been less surprising to those who had paid close attention to the surveys that largely predicted this shift.
That said, polls clearly have difficulty fully predicting what will happen in elections. And in an age when Internet-crawling humanity produces more than 400 terabytes of data every day, it becomes increasingly strange to pin hope on the responses of a few thousand people who happen to answer the phone, and who may or may not be honest with pollsters, or against himself.
For the past 15 years I have been studying Google searches. I and others have found that search data is often much more predictive than surveys. Google searches for “votes” and “votes” can predict who will actually vote, not just those who say they will – just as searches related to suicide can predict suicide rates better than survey reports of suicidal ideation. Google searches revealed where racism was highest in America and predicted Mr Trump's early rise. And in April 2020, I used them to discover a new symptom of Covid-19: eye pain, a finding that was confirmed by health researchers months later.
There is already evidence that search data can provide rich predictive power around elections, and not necessarily by asking simple questions. For example, Stuart Gabriel of UCLA and I found that the order in which candidates are searched for on Google, over many presidential cycles, is itself an indicator. People who search for “Trump Harris debate” are more likely to support Mr. Trump than people who search for “Harris Trump debate.” The most fascinating thing about this indicator is that again the data can reveal something that the searcher may not even realize. Seemingly undecided voters can express their support based on which candidate they list first in their search.
In Mississippi, a Trump stronghold, more than 65% of searches with both candidates' names included “Trump” first — the highest of any state. In Vermont, a Harris stronghold, 58% of searches with both candidates' names first included “Harris” — also the highest of any state. In total, 24 of the 26 states most likely to list “Trump” first in their Google searches went with two names to Trump. Nineteen of the 25 states most likely to list “Harris” first in their searches went for Harris. And we've seen four elections in a row for which adding this indicator would have improved state-level forecasts compared to poll averaging alone.
We are only at the beginning of research into how online data can help understand and predict human behavior. But it is abundantly clear that the 2024 elections will be among the last in which polls alone are used to predict outcomes.
Seth Stephens-Davidowitz is a data scientist and author who previously worked at Google.
© 2025, The Economist Newspaper Limited. All rights reserved. From The Economist, published under license. Original content can be found at www.economist.com