How Government Agencies Can Build Public Trust With AI Before Misinformation Fills the Gap

Author: Mary Frances Coryell, CGCIO

Mary Frances Coryell stands at the forefront of government technology as Chief Revenue Officer at Citibot, an AI-powered chat solutions leader serving government agencies nationwide. She brings nearly a decade of experience driving municipal digital transformation and fostering trust between governments and residents. Coryell's work has facilitated data-driven improvements across California and beyond, earning recognition for advancing innovation in resident engagement, accessibility, and operational efficiency.

Key Summary

AI has arrived at the government's digital front door. Residents are asking questions through websites, chat, text, voice, and search, often at the exact moment they need an answer about a service, deadline, fee, emergency, or public resource. Yet many government agencies are attempting to deploy AI on top of sprawling digital estates shaped by years of decentralized publishing, aging PDFs, broken links, unclear ownership, inconsistent language, and inaccessible content.

This white paper argues that public trust in AI is built before a resident ever opens a chat window. It is built in the accuracy of a fee page, the accessibility of a form, the clarity of a service instruction, and the discipline with which an agency maintains its official information. Citibot Refresh provides a practical model for that discipline. It continuously scans government websites and PDFs, flags outdated, inaccessible, inaccurate, or inconsistent content, and recommends specific fixes so teams can improve their digital front door without months of manual review.

For City Managers, Chief Information Officers, City Administrators, County Managers, and Communications Directors, the opportunity is larger than automation. It is to create a dependable information system that makes public service easier to reach, easier to understand, and harder to distrust.

Key Conclusions

Government agencies should treat their websites as operational infrastructure, not as a communications archive. A broken link, conflicting eligibility rule, inaccessible PDF, or stale service instruction is not a small editorial defect when a resident needs to pay a bill, report a concern, apply for a permit, or respond to an emergency.

AI strategy must begin with information governance. A government AI assistant can only be as trustworthy as the content, data sources, and service rules that ground its responses. Citibot's walled-garden approach is designed to keep resident interactions rooted in verified government information rather than unsupported guesses or invented details.

Accessibility is now a leadership imperative, not a cleanup project. Under the Department of Justice's ADA Title II rule, state and local government digital content must meet WCAG 2.1 Level AA requirements on a schedule tied to population thresholds. Government agencies serving populations above 50,000 faced a compliance deadline of April 24, 2026, while many smaller agencies face an April 26, 2027 deadline.

The strongest use of AI is not to imitate an open-ended consumer chatbot. It is to shorten the path from a resident's question to an accurate, useful next action. That requires accountable content owners, continuously monitored source material, visible knowledge gaps, and a shared standard for accuracy, currency, persona, and accessibility.

Key Quote:

“For too long, government agencies have made content and service decisions based on anecdote. Today, we can replace assumptions with evidence, using real resident questions to improve the information and services they depend on.” - Jonathan Wiersma, Chief Product and Marketing Officer, Citibot

The Trust Problem is not AI

The public debate about AI tends to begin at the moment a system answers a question. Did it sound human? Was it fast? Did it save time? Did it hallucinate?

For government agencies, that is too late in the story.

Trust begins earlier, in the quiet infrastructure of public information. It begins when a resident searches for a trash schedule and finds the right day. It begins when a parent tries to register for a recreation program and discovers that the information is current. It begins when a person using assistive technology can navigate a form, understand a deadline, and complete a task without asking someone else for help. It begins when an agency's website speaks in one clear voice instead of the overlapping dialects of departments that have each built their own corner of the digital city.

A government website is no longer simply a place where information is posted. It is a service counter, a records room, an alert system, a compliance surface, and often the first encounter a resident has with public institutions. Residents do not experience it as a collection of departmental responsibilities. They experience it as a government.

That is why the trust question has become so urgent. In an ICMA survey of 635 government practitioners, 55 percent identified resident engagement as the area with the greatest AI potential, including streamlined service interfaces and AI chatbots. At the same time, 56 percent said public perception and trust were concerns associated with AI use, while 77 percent cited limited awareness and understanding of AI as a barrier to adoption.

The contradiction is familiar to every public-sector leader. The demand for better digital service is clear. The risks of getting it wrong are equally clear. An agency may want to offer a modern chat experience, but it cannot afford an assistant that distributes incorrect permit information, misstates a payment deadline, gives an outdated emergency instruction, or sends a resident toward an inaccessible document.

The problem, then, is not that government agencies lack AI. The deeper problem is that they often lack continuous visibility into the information that AI is expected to represent.

The Website is a Living System

A government website can look complete while being quietly unreliable.

The homepage may be polished. The navigation may be modern. A resident may even find the right department. Yet beneath the visible surface are pages that have not been reviewed since a policy change, event notices that never expired, PDF forms that cannot be read by assistive technology, conflicting fee amounts, old departmental language, and links that lead nowhere. These are not unusual failures. They are the natural result of a website that grows faster than the organization's ability to govern it.

Citibot describes this condition as website drift. It is what happens when operational change outpaces content review. New programs appear, policies evolve, staff roles shift, facilities close, services move, ordinances are updated, and information fragments across teams. A periodic redesign can improve appearance and navigation, but it cannot solve a problem that renews itself every day.

The scale of the accessibility challenge illustrates the point. Research cited in Citibot's public materials found that 94.8 percent of homepages had detectable WCAG failures and that the average homepage contained 51 accessibility errors. Those figures are not limited to government, but they reflect a broad digital reality. Accessibility problems rarely exist in isolation. They signal gaps in templates, document practices, image tagging, content workflows, and accountability.

This is why the old model of website maintenance is no longer enough. Annual audits, inbox-based requests, informal departmental reminders, and one-time remediation projects produce temporary gains. They do not create the operational habits needed to keep information healthy after the project ends.

How a Walled Garden Builds Confidence

There is a temptation to treat government AI as a smaller version of the consumer internet. The logic goes something like this: residents use conversational AI elsewhere, so public agencies should provide a similarly broad, open-ended experience.

But government is not a search engine. It is accountable for the information it provides.

Jonathan Wiersma has described Citibot's approach as a walled garden. The phrase may sound restrictive in an era that prizes unlimited answers, but for government agencies it names the central safeguard. A resident should not need an AI system that can speculate elegantly about a zoning question or improvise an answer about a tax exemption. The resident needs the quickest path to the most accurate answer available from official government sources.

That distinction changes what success looks like. The goal is not a longer conversation. It is a resolved need. The best resident experience may be a brief exchange that identifies the right service, retrieves a verified answer, provides a location-specific detail, and directs the resident to the next official action.

Citibot's agentic V3 experience is designed around that practical standard. Through integrations with systems such as geographic information systems, waste-management platforms, and parks and recreation catalogs, Citibot can provide location-specific and live information rather than merely pointing residents to a generic page. A resident asking about collection service can receive guidance tied to an address. Someone seeking a recreation class can be directed toward current availability and registration. The AI becomes more useful because it is connected to the systems that hold official answers, not because it is encouraged to invent them.

Jordan Schinstock, Citibot's Director of Product Management, described the shift plainly: "Not only are we able to get those regular Q&A things handled with the bot, but now we're able to go that level deeper."

For public leaders, that is the appropriate promise of AI. Not novelty for its own sake. Not a substitute for public judgment. A more dependable route from need to service.

Four Signals of a Trustworthy Digital Front Door

Citibot Refresh approaches website governance through four interdependent signals: accuracy, currency, persona, and accessibility. Together, they give government agencies a clearer picture of whether their information is actually prepared to support resident service and AI-enabled engagement.

Accuracy is the first signal because official information must be correct before it can be useful. It includes inconsistent service descriptions, duplicate or conflicting content, broken links, and pages that no longer reflect current policies, programs, schedules, or fees. When accuracy fails, every downstream channel inherits the failure. A chat assistant can repeat the wrong information more quickly. A search engine can surface it more prominently. A resident can share it more widely.

Currency asks a different question: not only whether a page is correct, but whether anyone knows how recently it was reviewed. A page about emergency preparedness, utility billing, permitting, public safety, seasonal recreation, or elections has a different risk profile than an archived news release. Citibot Refresh helps agencies see aging content and prioritize work based on the pages and documents most likely to affect service delivery.

Persona recognizes that residents should not have to decode government's internal structure. A website can be technically complete and still be difficult to use if each department writes in a different voice, uses unexplained acronyms, or describes services through the lens of internal process rather than resident action. Citibot Refresh helps surface inconsistent messaging so agencies can move toward one plainspoken, unified public voice.

Accessibility is both a legal obligation and a statement about who public service is for. The Department of Justice rule requires state and local government entities to make web content and mobile apps accessible under WCAG 2.1 Level AA. Yet the deeper principle is straightforward. A service that cannot be independently accessed is not fully available. Government agencies need a process that can identify missing alt text, heading and structural problems, inaccessible forms, and difficult PDFs before residents are forced to bear the cost of those failures.

The power of the four-pillar model is that it replaces a vague sense of digital disorder with a common operating language. City management can talk about risk.

Communications can talk about clarity and consistency. IT can talk about accessibility and technical integrity. Departments can see the pages and documents for which they are accountable. The resident experiences the outcome as a site that simply works.

From Anecdotes to Evidence

Government agencies have always had informal ways of learning what residents need. Staff hear questions at the counter. Elected officials receive calls. Public meetings reveal frustration. Communications teams monitor social media. Each source matters. None, on its own, provides a complete view.

AI-supported analytics can add a more systematic signal. Citibot's knowledge-gap and departmental analytics capabilities are intended to show what residents actually search for, where official information is insufficient, and which department owns the underlying content. The result is a shift from intuition to evidence.

Schinstock calls this the "ear to the ground" function of the platform. The phrase is useful because it captures a quality that government agencies often struggle to scale. A resident question that cannot be answered is not merely a failed chat interaction. It is a signal that a page may be missing, a policy may be unclear, an integration may be needed, or an existing service description may be written in language residents do not recognize.

This is where Citibot Refresh and Citibot's resident-facing AI tools reinforce one another. Refresh helps improve the quality of official source material. Analytics show where residents are still getting stuck. Departmental filters help agencies turn those signals into accountable action. The system becomes a feedback loop rather than a publishing workflow.

Prince William County's experience points to the operational potential. In Citibot's customer discussion, the county was cited for centralizing 311 service and processing more than 90,000 interactions in its first year with the bot. The larger lesson is not simply volume. It is that a centralized, measurable digital service model can reveal demand patterns and reduce the burden on residents and staff alike when the information behind it is governed well.

How Leaders can Make Trust Operational

The most effective government agencies will not treat AI as a standalone initiative owned by one office. They will make it part of a shared operating model for public information.

The first step is to establish a baseline. Agencies should identify the web pages and PDFs that matter most to residents and to organizational risk. High-traffic service pages, payment and permitting information, emergency and public-safety content, public meeting materials, and frequently downloaded documents should receive attention first. Citibot's 90-day guidance recommends beginning with the top 100 pages by traffic, then using that view to focus limited capacity where it can improve resident experience most quickly.

The second step is to give every important content area an accountable owner. This is not about making departments defensive. It is about making maintenance possible. When an outdated page can be clearly associated with a department, a category of issue, and a recommended correction, communications and IT teams can shift from chasing updates to enabling better decisions. The governance conversation becomes concrete: here is the page, here is the risk, here is the action, and here is the resident consequence.

The third step is to treat PDFs as an active inventory, not as a forgotten archive. Government agencies often hold documents that are legally important, historically valuable, operationally necessary, obsolete, or all four at once. Citibot Refresh supports a more disciplined workflow by helping teams identify, track, archive, convert, or remediate documents rather than assuming every legacy file deserves the same manual effort.

The fourth step is to make optimization continuous. Citibot's customer team emphasizes that an agentic AI experience is not a set-it-and-forget-it deployment. An effective implementation requires review of resident questions, knowledge gaps, feedback signals, source material, and integrations. "These aren't just a one-size-fits-all," Schinstock noted. "That's why it's really important that we have that optimization call and that you are in touch with your client success manager."

The fifth step is to define success in resident terms. Page counts and chatbot sessions can be useful measures, but they are incomplete. A more meaningful scorecard asks whether residents can find answers faster, whether they complete digital tasks with less assistance, whether recurring questions decline after content improvements, whether departments resolve known issues, and whether accessibility barriers are removed before they become complaints.

The Next Standard of Public Service

The next standard for government websites will not be whether they were redesigned recently. It will be whether they can be trusted continuously.

That standard will require leaders to see content not as a collection of pages but as the information layer beneath every public interaction. Websites, chat, voice, text, search, forms, alerts, service portals, and future AI experiences all depend on the same foundation. If that foundation is inaccurate, inaccessible, stale, or inconsistent, adding new channels only expands the reach of the problem.

Citibot Refresh offers government agencies a different path. It combines continuous AI-powered scanning with a practical governance framework built around accuracy, currency, persona, and accessibility. It helps identify the specific information problems that weaken resident confidence, then directs teams toward concrete remediation rather than another static report.citibot+1

The larger idea is simple. Trust is not produced by an AI interface. It is earned through the thousands of small moments in which government gives a resident information that is clear, correct, accessible, and ready when it matters.

Citibot's mission is to build trust between governments and residents by providing accessible, AI-powered chat solutions that improve efficiency, enhance customer service, and facilitate real-time communication. Citibot Refresh extends that mission upstream, to the information that makes every resident interaction possible.

In the years ahead, government agencies that lead with governance will be better equipped to use AI with confidence. They will not ask residents to trust a technology because it is new. They will show residents, one accurate answer at a time, that their government can be trusted to serve them.