Election surveys based on limited samples frequently diverge from final outcomes. In the current age of AI, enormous volumes of online activity data from likes to buying habits can forecast political leanings with notable precision.

Voters casting ballots in Israel's 25th Knesset election

(photo credit: REUVEN CASTRO)

ByDR. SHAY HOROVITZ

AUGUST 5, 2026 09:55

Before each voting cycle, numerous surveys appear that draw from only a modest number of participants. Survey organizations claim their selections mirror the broader electorate by factoring in elements like age, gender, location, schooling, and religious observance. Still, these efforts often fall short when compared to actual election results.

With advances in artificial intelligence and data analysis, approaches to gauging public sentiment may shift. Rather than querying a small group about voting intentions, algorithms can examine digital traces from vast numbers of individuals to detect trends that would be hard for people to spot manually.

Insights from online traces

Each online move creates a record. Likes, viewed clips, queries, transactions, and followed pages seem minor alone. Yet when billions of such records accumulate across extended periods, they paint a detailed portrait of individuals, including their preferences, routines, and even fundamental beliefs and political orientations.

This idea has moved beyond theory. Recent research indicates that traces left on social platforms can reveal political positions. A notable 2013 Cambridge University project found that Facebook users' political leanings could be predicted at roughly 85 percent accuracy using only like patterns.

The system does not target likes for a particular party. Instead, mixes of music tastes, sports affiliations, shows, products, and varied interests reveal clusters typical of those sharing similar outlooks. Follow-up work confirmed that pursuits, followed pages, and even non-political group activity offer strong hints about political leanings.

This leads to a key inquiry: Might such insights improve the accuracy of public opinion tracking?

Might AI supplant conventional surveys?

Standard election surveys rest on the notion that querying a representative group allows estimation of nationwide voting. This technique carries known drawbacks. Surveys capture one moment and depend on self-reported answers. Some decline involvement, others mask their positions, and still others give responses they view as acceptable. Question phrasing, sequence, and even interviewer dynamics can sway results.

Data-driven techniques avoid reliance on momentary replies. They review long-term behavior patterns from far larger groups, emphasizing actual actions over stated views at one instant. While not free of potential skew or mistakes, this method enables opinion measurement at unprecedented scope.

Artificial intelligence already alters work, education, information retrieval, and choices. Whole sectors adapt, so it follows that election surveying would advance similarly.

The issue now centers not on whether technology can discern political preferences, but on whether opinion tracking will stay limited to brief phone replies or embrace methods that learn from millions of digital behavior signals. Care must also be taken to prevent these tools from being applied to influence rather than merely assess public views.

The author leads the Data Science Program in the Faculty of Computer Science at the College of Management Academic Studies.