Governments Can Use AI to Repair Public Trust. If They Listen to People First

On any given Tuesday night in America, a local government is hosting a meeting in a high school gymnasium to engage the public in something – from a new development, to traffic calming measures, to a master plan. There is lively discussion, notes scribbled, and dots placed on maps. There are online surveys and interactive tools.

Local governments reliably solicit people’s voices and collect data. For nearly two decades, I have worked with them to design creative tactics to bring people to the table. I have developed games, art, even theater to get people to show up and speak their minds. I am proud of this work. The problem is, in my experience, once governments assemble people and collect their data, they are not inclined to do anything with it. There is political power in holding a meeting, but very little in analyzing the data they collect. This is simply not sustainable. Governments are sitting on mountains of digital data. But because of a growing anti-AI sentiment, and the fear that listening to constituents will lead to unsustainable accountability, institutions are not eager to change their tactics.

This is exacerbating one of the most significant problems in democratic governance: people don’t trust government information or its messengers. A recent poll from the Partnership for Public Service showed that only 27% of Americans trust the federal government; that’s down from 33% in 2025. The newest logic is that an AI chatbot could change that by creating a better user experience and providing better public information. But according to Pew Research 63% of Americans believe AI is advancing too quickly, and 71% believe their personal information is less secure.

Consider Trump’s new America.gov website, then, with this context. The federal chatbot states that “despite 250 years of American progress, the citizen experience remains one of our nation’s greatest unrealized opportunities.” That’s true. The relationship between citizens and their public institutions is in crisis. Polarization and partisanship aside, public institutions need to take the production of trustworthy and reliable information seriously. But automation alone isn’t the answer. In fact, deploying more tools for one-way communication (chatbots included) without meaningful engagement in the limitations on use of data and opportunity for values alignment, more automation could be part of the problem.

It’s instructive to look at the local level. The most notorious case is the MyCity chatbot in New York City that was launched in 2023. It was designed to answer residents’ questions about starting a business, housing, and workers’ rights. But it gave answers to users that were wrong and in some cases illegal. It reportedly told landlords they could refuse tenants with housing vouchers, said employers could take a cut of workers' tips, and said businesses could refuse cash. All of these actions are against NYC or state law. As of last month, the chatbot has been taken down.

Well beyond New York City, city chatbots are used in hundreds of local agencies. For instance, the company Citibot promises “faster, more accessible, and trusted service” to governments and their citizens through a chatbot interface. Governments like tools such as these because they provide quick solutions to real administrative problems. But do they repair trust? People access information through technical proxies, but access alone does not lead to trust. Trust in public information requires that people understand the intentions of the institution with which they’re interacting and that they believe they share some common values.

Before governments can be trusted to deliver information, they need to be trusted to collect, interpret, and act on it. This is why, in Evanston, Illinois, when the city was given access to disaggregated public health data in 2022 that revealed that the majority Black and historically redlined Fifth Ward neighborhood had a life expectancy 16 years below the city’s average, they partnered with the community instead of simply releasing the data. The City organized data walks, where people in the neighborhood had the opportunity to scrutinize the data, add local nuance and discuss repercussions, prior to its public release. The result was a general willingness of residents to partner with the City to address the problem, as opposed to a suspicion of the data’s accuracy, and the intentions of the government.

In the Dorchester neighborhood of Boston, Massachusetts, because of a long history of neglect, trust in local government is low. Representations from mainstream news outlets tend to be negative, and the community’s access to government data is minimal and perceived by many locals to be unreliable. Instead of relying on mainstream chatbots to bypass these sources for local information, several community-based organizations have started building a trusted, local information source of their own. They curated trusted data sources from community organizations, open public data, and neighborhood leaders, and will soon be offering them up via a neighborhood chatbot interface called “On the Porch”. The community knows all too well that there is no shortcut to trusted information; it has to be cultivated.

At this moment of intersecting trust crises, in government and in AI, governments need to redirect their efforts from simply disseminating data to acting with accountability on the mountains of input, feedback and data they already have. People don't trust information; they trust messengers. And messengers become trustworthy when they authentically listen before they speak.

Government agencies, local and federal alike, should keep investing in better data quality, access, and legibility. Chatbots can help with that work. But they are not a replacement for listening to people. In my book, How Institutions Listen, I offer a framework for listening with accountability in ways that don’t further extractive surveillance, but demonstrate listening as a reciprocal process, beyond the intake of information. It is a relationship in which the speaker trusts the listener's intention to receive, comprehend, and remember what is being said. That relationship doesn't have to be small or intimate. It can happen at scale, through the kind of social scaffolding we saw in Evanston and Boston.

To confront our very real trust deficit, it would be misguided for governments not to deploy new technologies. But to deploy those technologies without prioritizing the human relationships that lead to institutional trust would be even worse.