Between October 30 and November 2, 2025, our voice AI system conducted 43 semi-structured interviews with constituents located in the 5 boroughs. The AI asked about party identification, voting intentions, candidate support, policy priorities, information sources, and concerns about the city. It followed up on answers, asked for clarification, and let conversations extend when respondents wanted to elaborate. Constituents accessed the interview through a web link, spoke naturally, and ended the conversation when finished. Interview lengths varied depending on constituents’ willingness to share, but ran as long as 30+ minutes.
These are excerpts from 43 conversations between New York City voters and the Third Ear voice AI interviewer during the final days of the 2025 mayoral election…
Between October 30 and November 2, 2025, our voice AI system conducted 43 semi-structured interviews with constituents located in the 5 boroughs. The AI asked about party identification, voting intentions, candidate support, policy priorities, information sources, and concerns about the city. It followed up on answers, asked for clarification, and let conversations extend when respondents wanted to elaborate. Constituents accessed the interview through a web link, spoke naturally, and ended the conversation when finished. Interview lengths varied depending on constituents’ willingness to share, but ran as long as 30+ minutes.
These are excerpts from 43 conversations between New York City voters and the Third Ear voice AI interviewer during the final days of the 2025 mayoral election. The interviews averaged 14 minutes each and generated thousands of words of detailed constituent perspectives. We conducted this study to test whether a goal-directed, semi-structured autonomous voice AI could collect the kind of nuanced, qualitative political data that traditionally requires expensive and time-intensive human interviewers.
What Constituents Actually Said
The Housing Affordability Crisis is a Top-of-Ballot Issue
Roughly 76 percent of respondents raised housing affordability without being prompted about it. They didn’t just mention it, they explained it in vivid detail.
“I can’t afford to live in New York without roommates. I probably won’t be able to live alone ever.”
“There are people who are homeless and living in shelters who are making $50,000 a year. That’s just unacceptable.”
“Everything to me is about rent. I have to ask myself, “Do I want natural light?”
These quotes come from different interviews, but they share a common thread: housing isn’t an abstract policy issue for these voters. It’s a daily constraint that shapes where they live, how they work, and whether they stay in the city.
Safety Concerns Are Split Along Multiple Lines
Public safety appeared in 60 percent of interviews, but perspectives diverged in ways that simple polling can’t capture.
Some respondents wanted more police presence:
**“Crime has gone up ever since the COVID pandemic. I don’t feel very safe.”**
Others felt over-policed:
***“I feel safe from my neighbors. I don’t feel safe from the police.”***
Still others focused on mental health:
**“There are homeless people that are just batshit crazy. Mental health cases shouldn’t be handled with cops.”**
These aren’t contradictory views, but reveal different people experiencing different aspects of urban safety. A survey asking multiple choice questions would miss these distinctions entirely.
Socioeconomic aftershocks of the COVID-19 Pandemic Persist
Surprisingly, a recurring theme which resurfaced in several voter interviews was the 2020 lockdown.
One respondent recounted lasting psychological impacts: *** ***
“When COVID started, and for COVID, there was a curfew enacted. There was some looting going on. I felt that the city wasn’t a very safe place. And then during the lockdown, although there were no crime increases as a result of COVID, I was very on edge. Hesitant to leave my home. Which affected my personal productivity.”
Another attributes their present economic struggles directly to the pandemic:
***“We’re all struggling after the COVID virus. And I know I’m barely making it... That’s why we really would like somebody elected that would provide more opportunities.”***
Political Apathy is Increasingly Endemic
Across party lines, across candidate preferences, and across policy priorities, one theme appeared consistently: deep skepticism about political promises.
“Nine times out of ten, these guys don’t usually uphold their promises.”
“I don’t want it to be where the last mayor was corrupt.”
“Do they actually want to make change or just want the spotlight?”
“So I don’t know. When I look at political parties, I see smoke and mirrors, and I think on a deeper scale, there’s a lot more going on than what you see on the surface. Right? So for now, I guess I’m gonna (identify as) independent.
Even voters who supported establishment candidates expressed this wariness.
“I’m not sure if he (Mamdani) can really live up to all his promises, if it’s really gonna be possible. But, I think if Cuomo ends up winning, we’re not gonna get any better.”
The State of the Race
Among respondents who stated a preference, support clustered around three candidates*:
Zohran Mamdani drew ~47 percent support, largely from progressives and anti-establishment voters. Supporters cited his outsider status, his stance on Gaza, and his rent freeze proposal. But even supporters worried:
***“He just overpromises. Not gonna deliver.”***
Roughly 14 percent of constituents expressed support for Curtis Sliwa, primarily Republicans and voters prioritizing crime reduction. His Guardian Angels experience resonated:
***“He’s been working on the streets for a long time.” ***
Andrew Cuomo held ~12 percent support, with supporters valuing his gubernatorial experience:
“He was governor. He should be able to run a city.”
About 21 percent of voters surveyed remained undecided, often torn between wanting change and doubting whether outsiders could govern effectively.
*Note: Results from this study are primarily qualitative and based on a small sample of 43 voters. These figures are not sufficiently predictive of election outcomes; rather, they are intended to accompany real-life examples of voter experiences, attitudes, and positions. _________________________________________________________________________________________________________
The Era of Interviewing at Scale is Here
We now have the possibility of one-on-one interviews which are free of judgment from another person. We believe that data-gathering on sensitive topics benefits from making it possible for respondents to talk freely in such a private space. This is an exciting time for research on human data, from social science research to UX research. Third Ear interviewers are driven by a symbolic dialog planner to ensure we don’t get derailed by hallucinations in language models. This NYC study demonstrates three capabilities that are transformative for deep qualitative research:
Interview-Level Depth at Survey-Level Cost
The Third Ear voice interviewer slashes data acquisition costs by two orders of magnitude, from thousands of dollars to tens of dollars, and is driven by a symbolic dialog planner to ensure we don’t get derailed by hallucinations in language models. It can conduct virtually unlimited interviews in parallel, speeding up the process of getting from inquiry to insights from weeks to hours. This isn’t just about saving money, but about making large-scale qualitative research feasible. That changes what’s possible to learn.
Consistency Without Rigidity
Human interviewers can vary in how they follow the protocol with different respondents. They might probe housing deeply with one respondent while another rushes past another one. They might inadvertently signal approval towards one while being neutral towards another respondent.
Our AI treated every respondent consistently. Every interview used the same conversational approach, eliminating the variability that plagues human-conducted qualitative research. This consistency means patterns in the data reflect actual constituent differences rather than artifacts of who conducted which interview.
Yet, the system remained flexible to the respondents. When respondents elaborated, our AI followed up. When they finished, it moved on. The structure stayed consistent while the content varied naturally.
Breadth and Depth. Without Compromise.
Every interview is rich with explanations, stories, and arguments. Previously, research on human subjects forced a choice: larger samples with lower resolution or smaller samples with high fidelity. Our voice AI can run countless interviews simultaneously, eliminating that constraint and allowing researchers to collect hundreds of detailed responses at comparatively little cost. It took less than an hour to configure the interviewer for this specific study. The average interview lasted 14 minutes and generated over 1,000 words from respondents. Our dialog planner manages the mixed-initative interaction, giving as much flexible space for the respondent to go deeper while being guided by the goals of the researcher.
Our mission is to democratize deep research and allow researchers to peek into what lies in the hearts and minds of people. We are making the de-identified interview transcripts available for research purposes. Reach out to research@thirdear.co to request the dataset or if you would like to work with us.