What 43 Voters Told an AI About the Future of New York (www.thirdear.co)

Between Oct. 30 and Nov. 2, 2025, Third Ear deployed an autonomous voice AI to run 43 semi-structured interviews with New York City voters across all five boroughs, averaging 14 minutes and generating ~1,000+ words per interview. The system asked about party ID, voting intentions, policy priorities and concerns, followed up naturally, and let respondents speak freely. The qualitative results highlighted housing affordability as the dominant, visceral issue (76% raised it unprompted), a fractured picture on public safety (60% mentioned safety but with divergent views), lingering COVID-era socioeconomic effects, and widespread political skepticism. Among this small sample, support clustered around Zohran Mamdani (~47%), Curtis Sliwa (~14%), Andrew Cuomo (~12%), with ~21% undecided — framed explicitly as illustrative rather than predictive. For the AI/ML community this study demonstrates scalable, low-cost qualitative data collection: the voice interviewer was configured in under an hour, runs interviews in parallel, and reportedly reduces data-collection costs by two orders of magnitude. Key technical features include a goal-directed, symbolic dialog planner to keep conversations on track and mitigate LLM hallucinations, and a mixed-initiative design that balances consistent protocol adherence with flexible probing when respondents elaborate. Implications: enables high-resolution, large-N qualitative research and reduces interviewer variability, but raises methodological and ethical caveats (small non-representative sample, privacy/consent considerations). De-identified transcripts are available on request for further research.