Can AI Predict How We Vote?

Every election season brings dozens of opinion polls, typically based on surveys of just a few hundred people. But what if, instead of asking hundreds of people how they intend to vote, algorithms could analyze the digital behavior of millions and identify patterns that are almost impossible for humans to detect?

Every digital action we take leaves behind a data footprint. The likes we give, the videos we watch, the searches we conduct, the purchases we make and the accounts we follow may seem insignificant individually. But when billions of these digital traces are combined over months and years, they create a surprisingly detailed picture of who we are: our interests, habits, values and even our political attitudes.

This is no longer a theoretical possibility. Over the past decade, research has demonstrated that information collected through social media can reveal significant insights into people's political preferences.

One of the best-known studies, conducted at the University of Cambridge in 2013, demonstrated that Facebook users' political orientation could be predicted with approximately 85% accuracy based solely on their likes. The algorithm does not necessarily look for likes on political parties or politicians. Instead, it identifies patterns across seemingly unrelated interests, including music, sports teams, television programs, brands and other activities.

Subsequent research has further demonstrated that hobbies, followed accounts and participation in non-political online communities can provide meaningful clues about political preferences.

But this raises a more interesting question: Can this information be used to measure public opinion more accurately?

Election polling is built on a relatively simple assumption: if we question a representative sample of the population, we can estimate how the entire country will vote. In practice, however, this approach has well-known limitations.

Traditional polls capture public opinion at a single point in time and depend on answers provided by respondents. Some people refuse to participate, others conceal their actual preferences, and some may provide answers they consider socially acceptable rather than expressing their genuine views.

The wording and sequence of questions, and sometimes even the interaction with the interviewer, can influence the results.

Data science offers a fundamentally different approach. Rather than relying exclusively on answers provided during a brief telephone conversation, algorithms can analyze behavioral patterns accumulated over extended periods and across much larger populations.

In other words, they can examine what people actually do, rather than relying solely on what they say they intend to do.

This approach also has limitations. Digital behavior is not necessarily representative of the entire population. Algorithms can reproduce existing biases, misinterpret online activity and draw incorrect conclusions. Not everyone uses social media in the same way, and political preferences cannot always be inferred reliably from seemingly unrelated activities.

Nevertheless, the combination of large-scale data analysis and artificial intelligence creates new opportunities for understanding public opinion on a scale that was previously impossible.

Artificial intelligence is already transforming the way we work, study, search for information and make decisions. Entire industries are undergoing technological transformation. It is only natural that election polling will evolve as well.

The real question is no longer whether technology can learn something about our political preferences. It is whether we will continue measuring public opinion primarily through a few hundred survey responses, or embrace tools capable of learning from millions of digital behavioral patterns.

At the same time, society must confront an equally important challenge: ensuring that these technologies are used not only to measure public opinion, but also not to manipulate or manufacture it.

The future of election polling may depend on understanding not just what voters say, but what their digital behavior reveals — while protecting their privacy and democratic freedom.