Ambiral uses AI personas to simulate Okinawa election results

Ambiral Inc. has successfully simulated the results of the September 2026 Okinawa gubernatorial election using AI personas. By utilizing only data available prior to the election date, the company’s simulation reproduced the vote share of the winner, Genta Koja, with an error margin of just 0.09 percentage points (59.51% simulated vs. 59.42% actual).

The methodology involved creating synthetic personas based on the 2020 national census, assigning them attributes such as municipality, gender, age, and occupation. These personas were also given historical voting data from the previous six elections, estimated through ecological inference. Large language models were then used to predict voting behavior based on these attributes and pre-election news reports.

While the simulation showed high accuracy for voters aged 30 to 60, it encountered a larger discrepancy among voters aged 70 and older, where the simulated vote share for Koja was 49.5% compared to the actual 35%. Ambiral aims to expand this technology through its ‘AI Resident Platform’ for marketing research, demand forecasting, and public opinion prediction.