The Dialogue Pulse · Episode 1
Los Angeles is heading into a mayoral runoff: Karen Bass versus Nithya Raman, November 3, 2026. We sat down with voters, one AI-moderated voice conversation at a time, and listened for what they brought up on their own.
Scroll to begin ↓
Beat One
By the time a campaign reaches a runoff, the polls have told you who is ahead. They rarely tell you why a city feels the way it feels. That was our question: not who will win, but what is actually living in voters' minds when they think about their mayor and their city.
So we ran a study. Over about five days we conducted 91 voice interviews with registered City-of-LA voters who intend to vote in the runoff. Each conversation lasted around 22 minutes and was led by an AI moderator named "George." We started every interview before naming a single issue, just to hear what people raised on their own. Only later did we test specific campaign messages.
One verdict surfaced again and again, unprompted. Voters were not arguing about left versus right. They were rendering a competence verdict: the city takes in money and resources, and it cannot turn them into anything people can see on their own street.
"The money isn't going to where it's supposed to be going to, and that could be unfortunately because of corrupt individuals."
— Participant P01
Of the 92 voters we analyzed, 77 (84%) took an anti-incumbent posture. Crucially, it held up across every age and gender group we had enough people to report on, so it is not just an artifact of a young sample. That makes it the spine of everything that follows.
Beat Two
Voice interviews are only as good as their signal. Before drawing any conclusion, every interview was scored on a credibility framework we call the Interview Quality Score (IQS), weighing the strength of what was said, how the moderator behaved, and how smoothly the technology held up.
Across 92 analyzed interviews (90 scored; one void interview removed). Higher is better.
The moderator channel is our weakest, and we say so plainly. A handful of interviews showed the AI leading or stumbling; those slips were caught and corrected before analysis. We would rather show you the soft spot than hide it.
The design matters: we captured what voters said in their own words first, then tested campaign messages, so the message-testing could not contaminate the unprompted capture.
Beat Three
When we asked people what was on their mind about their city, before naming a single topic, the same concerns surfaced over and over. Here is how often each theme came up among the 92 voters we talked to.
Share of the 92 analyzed voters who raised each theme. This is how common a theme was among the people we talked to, not its rate across Los Angeles.
Every single voter raised homelessness. But it rarely stood alone: nearly all of them tied it to the cost of simply living here. Voters disagreed about the cause, and that disagreement was real and unresolved among them.
"They keep giving permits to these big developers who then turn around and put apartments on the market for ridiculously high rents."
— Participant P78
The fires came up for 90% of voters, but almost never as a debate about forest management or building codes. The recurring image was simpler and more human: a mayor who was not there. One voter put it as starkly as anyone:
"She wasn't here when action was required of her. She was out of town on vacation."
— Participant P52
To be fair to the disconfirming voices, several voters explicitly rejected that blame. "I don't know what she has to do with it," said one (P08). The wound is about absence, and the data does not pretend the city agrees on whose fault it was.
Underneath nearly everything was a single epistemics: voters discount any claim they cannot see for themselves. One voter delivered the litany that became the spine of this whole study:
"I don't see it on the buses. I don't see it on the trains. I don't see it on my street of my apartment. I don't see it as I'm walking home from work. I don't see it."
— Participant P25
Beat Four
Halfway through each interview, after we had heard voters in their own words, we read them six real-world campaign messages, three about homelessness and three about the wildfire recovery, and asked them to react. The pattern was unmistakable. Optimism backfired. Accountability and candor landed.
Each bar below splits at the center: reject to the left, agree to the right, among the voters who actually reacted to that message.
Among those who reacted to each message, n ≈ 57–84. Bars show the share who rejected versus agreed; the remainder were mixed or neutral. Two interviews received non-standard messages and are excluded from this matrix.
"They are continuing to collect tax revenue, and it's continuing to disappear magically into the ether."
— Participant P78, on why "show me where the money went" lands
Beat Five
A study like this is descriptive and directional, not a prediction. But the shape of what voters told us points in a few practical directions.
Among these voters, "we're turning a corner" was rejected almost as fast as it was read, while "show me where the money went" and "repair the systems that failed" landed. A schedule and proof beats a promise.
The anti-incumbent feeling these voters expressed routed through a single question, can the city convert money into visible outcomes, rather than through left-versus-right. That is a different story to cover.
Voters told us things in these conversations that they said they would not tell a person. That candor, explored in the final beat, suggests a real role for this method in sensitive listening at scale.
Beat Six
All 92 analyzed voters passed a City-of-LA registration screen and intended to vote in the runoff. Every interview was completed in English. Here is the shape of the sample.
The surprising part
At the end of each interview we asked a quiet meta-question: was anything easier to say to an AI than to a person? Of the 41 voters we asked, 61% said yes it was easier, and 90% said it was easier or no harder than talking to a human. One voter, who had been homeless himself, said the thing that stayed with us:
"It was easier. I told you about homelessness and my past, that I wouldn't tell a human. It's very embarrassing, but it wasn't as embarrassing to tell you."
— Participant P77
Not everyone agreed. A few voters bristled at the machine. That dissent is exactly why we trust the finding rather than treating it as a sales pitch: people told us when it did not work, too.
When you remove the fear of being judged, some people will finally tell you the truth. That, as much as anything we heard about money or fires, is what this first episode of The Dialogue Pulse found.
The Dialogue Pulse · Episode 1 — 94 LA voters, in their own words.
Enter your email to read the full report.
We'll only email you about this report and related research — never spam, and we'll never sell or share your address.