Vigilance towards AI voices can be nudged through a change in 'MINDSET'

AI-assisted technology is used to create synthetic voices that are highly naturalistic, making it difficult for listeners to distinguish between real and synthetic speech. While listeners often show a bias towards classifying these voices as human, this effect is even stronger when the voices use underrepresented regional or non-standard dialects—presumably because listeners are not used to such varieties being represented by speech technology. This MINDSET—Minority, Indigenous, Non-standard, and Dialect-Shaped Expectations of Technology—could leave some language communities more at risk of AI-voice based deception. To address this, the current study tested whether simple informational nudges could shift listeners’ default assumptions away from ‘Human’ and increase their vigilance towards categorizing voices as ‘AI’. Experiment 1 (N = 150) investigated whether nudges outlining AI’s ability to produce (Scottish) accents and dialects would affect human categorization responses. The results demonstrated a significant reduction in ‘Human’ responses for a nudge outlining AI’s capabilities at authentically producing these varieties. In Experiment 2 (N = 150), a vigilance-based nudge warning about the risks of AI deception was tested alone and in combination with the capability message. Only the nudge containing the capability-based information had a measurable effect, suggesting that updating expectations about what AI can convincingly reproduce is more effective than simply warning listeners to be cautious. These findings have practical applications for cybersecurity, fraud prevention, and public awareness campaigns. As AI-generated voices become increasingly used in scams, deception and misinformation, informational nudges which update expectations about AI’s linguistic capabilities may offer a low-cost, scalable way to increase vigilance—particularly in language communities historically marginalized or excluded from speech technology systems.