Fact-Checking Trump’s Claim That Google ‘Manipulated’ Millions of Votes for Clinton

What President Trump Said

Mr. Trump lost the popular vote by almost 2.9 million in the 2016 election and has long attributed this loss to baseless claims of voter fraud. Now Mr. Trump suggests that he would have lost the popular vote by a much smaller margin, or even won it, were it not for Google’s machinations.

He appears to be referring to a disputed estimate given by Robert Epstein, a psychologist and former editor in chief at Psychology Today who says he supported and voted for Hillary Clinton.

In a 2017 white paper, Mr. Epstein examined how Google Search might have influenced undecided voters in the 2016 election by providing more positive results for one candidate than the other. He estimated that the search results may have swayed “at least 2.6 million votes to Clinton.” When he testified to Congress this summer, Mr. Epstein added that the estimate was a “rock bottom minimum” and that up to 10.4 million votes could have been shifted to Mrs. Clinton.

In an interview, Mr. Epstein took issue with Mr. Trump’s characterization of his work.

“I’ve never said Google manipulated the 2016 elections,” he said. “The range of numbers he listed in the tweet is also incorrect.”

Panagiotis Metaxas, a computer science professor at Wellesley College, emphasized that the white paper showed a possibility — “what such an influence could have been if Google was manipulating its electoral search results” — not a conclusive fact.

“I and other researchers who have been auditing search results for years know that this did not happen,” Mr. Metaxas said. “I think that, in his congressional hearing, Dr. Epstein is misrepresenting the situation.”

He noted that Google does “sanitize” its search results, prioritizing more trusted sources while devaluing low-quality information sources. (Here’s a more detailed explanation of how Google Search works.)

The white paper also came with huge caveats. First, it was not peer-reviewed or rigorously evaluated by other researchers.

It was based on the daily online searches of just 95 participants, 21 of whom were self-described undecided voters — a small sample size to extrapolate to millions of voters, experts said. (Mr. Epstein says that the statistical significance of his findings was high.)

Their election-related search results were then given to another group of people who evaluated whether the results were biased toward Mr. Trump or Mrs. Clinton, and concluded that most favored Mrs. Clinton. Searches conducted on other websites like Yahoo and Bing did not display the same bias, according to the paper.

Then, Mr. Epstein applied his previous research demonstrating that biased search results could shift voting preferences by 20 percent or more to reach his baseline estimate of 2.6 million people affected.

But it takes “a leap of faith” to connect Mr. Epstein’s experimental results to actual election outcomes, said Nicholas Diakopoulos, an assistant professor in communication studies at Northwestern University. (Mr. Diakopoulos’s own research does show that Google Search results favor Democrats.)

“There’s a substantial shift in context, and a difference in asking someone in an experiment about likely voting behavior and how they might actually act during an election,” Mr. Diakopoulos said.

A more meticulous scientific analysis measuring the impact of Google Search on election outcomes could, for example, take into account voter history or other sources of election information.

Given the difficulties in disaggregating the impact of Google and the paper’s lack of methodological detail, Mr. Diakopoulos said, “I am skeptical of the validity of the estimates.”

Curious about the accuracy of a claim? Email factcheck@nytimes.com.