Last night was the first Democratic presidential debate and almost immediately afterward the “winners and losers” articles started appearing. But how do we know who’s really winning?

Right now, polls indicate that several of the Democratic candidates would beat Trump if the election were held today. But polling famously got it wrong in 2016. And as pundits, reporters and political consultants seek to assess the state of the race they are increasingly likely to supplement polling data with other types of information, such as search, social engagement and, especially, location data.

Search data useful but ambiguous. Search volume and query data can indicate a number of things about popular interest in candidates. While useful, there are also some problems with relying on search data alone, such as ambiguity inherent in the data and the often uncertain relationship of queries to subsequent behavior.

Search volumes may also not be entirely representative of the voting public. This is similar to the way that online social media activity around brand sentiment doesn’t reliably predict offline buying behavior.