Skip to Content

Here’s How Well Search Engines Predicted Super Tuesday

Republican Candidates Take Part In Debates At Reagan Library In Simi ValleyRepublican Candidates Take Part In Debates At Reagan Library In Simi Valley
Republican presidential candidate Donald Trump. Photograph by Justin Sullivan—Getty Images

Increasingly, we are what we search.

As primary and caucus results trickled in around the country on Tuesday night, Google and Microsoft Bing had already picked their own winners based on how people search the Internet. While the search engines got the general gist of Super Tuesday voting right, they weren’t quite perfect when it came to knowing exactly how things shake out behind the voting curtain.

Microsoft’s (MSFT) Bing search engine used a mix of search data, polling, and predictions to pick out big winners, while Google Trends (GOOGL) tracked search interest in candidates in real time.

But both Bing and Google over-estimated the appeal of frontrunners Donald Trump and Hillary Clinton. Bing’s tool missed that Republican Ted Cruz would win in Oklahoma and Alaska and didn’t predict Marco Rubio would get Minnesota.

Similarly, the tool called Clinton’s sweep of the seven southern states that were voting, but didn’t guess Bernie Sanders would win Colorado, Minnesota, or Oklahoma.

MORE: More election coverage from Fortune

That’s a little worse than the search engine crystal ball has fared so far this election cycle: Before Tuesday, Bing had correctly predicted seven of the eight primary and caucus voting states, as CNET reports.

But researchers like psychologist Robert Epstein say that search engine results can have a lot of sway when it comes to certain kinds of voter opinions and can change the way a searcher sees a candidate for the better. Epstein found more positive search results can shift undecided voter opinions by 20% or more.

Get Data Sheet, Fortune’s technology newsletter.

Still, based on Super Tuesday returns, it seems search engines don’t know all our secrets. Not yet, anyway.