Busting the one-voice-fits-all myth: Effects of similarity and customization of voice-assistant personality

Despite the increasing sophistication of voice assistant (VA) technology, most major VAs subscribe to a one-voice-fits-all model of interaction. This study examines if offering users a VA similar to them, or letting users customize the VA's voice personality, would affect their perceptions and experience. We test this in a unique scenario where a VA delivers misinformation about COVID-19. Data from a pre-registered experiment (N = 401) suggest that extroverted users appreciate being matched with an extroverted VA, whereas introverted users do not have a specific preference. In addition, perceived homophily in voice increases user attraction toward the VA, and enhances credibility perceptions for those who customize their VA. Those not given the option to customize prefer VAs with an extroverted voice. Data also suggest that automated similarity matching of VA personality can evoke user resistance toward the persuasive information provided—in our case, changing as many as 38% of unvaccinated individuals' mind toward vaccination after exposure to misinformation.