Commentary: The real political risk of AI isn’t deepfakes

What we measured was affective polarisation – how warmly or coldly voters felt toward coalitions, parties and leaders after being exposed one of these experimental conditions.

The results were striking. First, exposure to deepfakes alone did not increase polarisation. Respondents who read about the opposing party’s use of deepfakes did not become significantly more hostile toward rivals than those in the control group. This finding challenges the common assumption that deepfakes automatically inflame partisan divides.

Second, voters’ approval of AI use mattered. Those who expressed higher approval of candidates using AI in campaigning showed lower levels of polarisation across coalitions, parties and leaders. In other words, far from intensifying divides, being more receptive to AI use correlated with a depolarising effect.

Third, demographic differences surfaced. Women consistently reported lower levels of polarisation than men, while older voters showed slightly reduced hostility toward rivals compared to younger ones. These results suggest that gender and age are key, and currently overlooked, variables in understanding how AI reshapes political attitudes.