Not long ago, the phrase “artificial intelligence” in a city hall conversation conjured images of a distant, expensive future. Today it sits on the desktop of nearly every employee who has opened a web browser. 

Surveys suggest that a majority of government workers are already using generative AI tools in some form during a typical week. The question facing managers is no longer whether our organizations will use these tools, but whether we will use them well.

In my years managing cities of very different sizes, I learned that the most powerful technologies are the ones that amplify good judgment rather than replace it. Generative AI is exactly that kind of tool—but only if we ask it the right way. 

The most common reason a manager comes away unimpressed is not that the technology failed; it’s that the request was vague. A better question produces a better answer, and asking clear questions is a skill local government managers already possess.

It Starts with a Better Question

When I coach colleagues who are new to these tools, I offer a simple structure that turns a shrug-worthy response into a useful draft. Think of it as four parts: role, task, context, and format.

Give the tool a role (“Act as a municipal budget analyst.”) State the task plainly (“summarize,” “compare,” “draft,” “critique.”) Provide context—the specifics only you know (“for a council work session in a city of 40,000 with a structural deficit.”) And specify the format you want back (“one page, in plain language, with a short list of risks.”)

Put together, a weak prompt like “help me with my budget message” becomes: “Act as a city budget director. Draft a 300-word budget message introduction for a city of 40,000 facing a modest structural deficit. Tone: candid but reassuring. End with three priorities for the coming year.” The difference in output is dramatic—and it took only 15 seconds more to ask.

Where AI Earns Its Keep

The uses that hold up best are the ones that take routine drafting and synthesis off a manager’s plate, freeing time for the work only a manager can do.

Council and governance. Feed a long staff report to the tool and ask for a one-page summary of key points, risks, and fiscal implications for councilmembers. Ask it to play a skeptical councilmember and pose the five toughest questions you should be ready to answer before a work session. Turn rough notes into clean draft minutes.

Budget and finance. Ask the tool to translate three budget scenarios into plain language a neighbor could follow, or to generate the probing questions you should put to a department head during a budget review.

Public communications. Draft an email to neighbors explaining a new program, a social media post inviting people to a hearing, or a rewrite of a dense notice at an eighth-grade reading level. These are the tasks that pile up and delay everything else.

Resident inquiries. Draft a courteous, empathetic reply to a heated complaint that de-escalates while staying factual, or summarize the themes across dozens of public comments and group them by issue.

Policy analysis. Ask for the potential unintended consequences of a proposed ordinance and who might be affected, or a side-by-side comparison of two policy options.

In every one of these cases, the AI produces a first draft. You supply the judgment, the local knowledge, and the final word. That division of labor is the whole point.

The Guardrails that Matter

Enthusiasm without discipline is where managers get into trouble. A few rules have served me well.

Treat every output as a first draft, never the final word. 

These tools can be confidently wrong—the industry calls it “hallucination.” Verify facts, figures, names, and especially anything with legal weight before it leaves your building.

Keep sensitive information out. 

Personnel records, neighbor personal data, security details, attorney-client material—none of it belongs in a consumer AI tool. Assume that anything typed into a free service is not private.

Remember public records law. 

In many states, the prompts you enter and the outputs you receive may themselves be public records subject to retention and disclosure. Check your state and local rules before weaving AI into routine workflows, and bring your clerk and attorney into the conversation early.

Be transparent. 

When AI meaningfully shapes a public-facing decision or communication, your neighbors deserve to know—and your credibility depends on it.

Start Small and Scale to Your Community

None of this requires a large budget or a data science team. Many capable tools are free or low-cost, and the paid enterprise versions—which offer stronger data protections and are worth the investment for government work—remain modest compared with most technology line items. A community of 5,000 and a city of 500,000 can begin the same way: pick a handful of low-risk, high-volume tasks, such as drafting routine correspondence, and let staff practice there first.

As comfort grows, so can ambition. Designate someone to stay current on the tools and your policy. Write a short, plain-language acceptable-use policy before wide adoption, not after. And treat this as an ongoing conversation with your team rather than a one-time rollout. The organizations that succeed with AI will not be the ones with the fanciest technology. They will be the ones whose people were prepared to use it thoughtfully.

The Judgment Stays Yours

For all the understandable anxiety about artificial intelligence, I find it clarifying to remember what our job actually is. Managers are paid for judgment—for weighing competing interests, reading a community, and standing behind a recommendation. No tool does that for us. 

What a well-prompted AI can do is hand us a faster first draft, a sharper set of questions, and a little more time for the human work of governing. Asked well, it is one of the most useful assistants a manager has ever had. Asked poorly, it’s a novelty. The difference, as it so often is in our profession, comes down to how we choose to communicate.

Resources

1. International City/County Management Association, Local Government AI Strategy Workbook (2024).

2. Government Technology, “50 ChatGPT Prompts for State and Local Government.”

3. University of Michigan Ford School of Public Policy and the Michigan Municipal League, Artificial Intelligence Handbook for Local Government (2024).

4. National League of Cities, AI in Action: Empowering Local Governments.

LEE R. FELDMAN, ICMA-CM, is senior advisor for local government strategy at Zencity and a past president of ICMA. He previously served as city manager of Gainesville, Fort Lauderdale, Palm Bay, and North Miami, Florida.

Readers who would like the author’s two-page prompt cheat sheet, adapted from this article, may request a copy by emailing him at leefeldman@zencity.io.