A small city doesn’t necessarily need a standalone AI department to start using the technology. But local officials do need to know which tools are safe, where human oversight matters, and how to protect residents’ information.
Georgia’s new Local Government AI Playbook gives cities and counties a place to start. The 56-page guide offers practical advice on setting AI policies, reviewing potential projects, training employees, and putting safeguards in place before using new tools.
It also gives local governments a framework for deciding whether AI is the right solution in the first place, how to assess potential risks and when projects need additional review and human oversight.
A LOCAL REQUEST BEGAN IT
The idea took shape during a Google-hosted roundtable in Atlanta earlier in 2026, when state Rep. Dar’Shun Kendrick, a member of the Georgia Technology Authority’s (GTA) Oversight Committee, heard local officials asking for help navigating AI. She asked GTA to turn its state agency guidance into a downloadable resource for cities and counties.
The goal was to give local governments practical advice, even those with few technology employees or no dedicated AI staff. Smaller communities shouldn’t feel they have to become AI experts or use the technology for everything, Kendrick said: “Take what you can and start.”
The playbook’s practical guidance comes with a reminder that AI adoption must keep people at the center. In it, state CIO Shawnzia Thomas wrote that “AI should support public employees, not replace the judgment, accountability, and community knowledge that effective government requires.” For Chief Digital and AI Officer Nikhil Deshpande, who helped develop the guide, the effort was personal. Having joined state AI leadership while its program was still taking shape, he wanted to create the kind of starting point he wished he’d had.
START WITH THE PROBLEM, NOT THE PRODUCT
One of the playbook’s central recommendations is that local governments should resist buying AI simply because a vendor is selling it or another community is using it.
Instead, officials should first identify the problem they are trying to solve. Is a department spending too much time answering routine questions? Are employees struggling to find information? Are residents waiting too long for a service? And is AI actually the right tool, or could the problem be addressed by improving a form, updating a website or changing an existing process?
The playbook lays out an eight-step project process that begins with defining the problem and considering alternatives. Only after doing so, it says, should a government assess the risks, select or build a solution, test it, and decide whether to expand.
FIRST, KNOW WHAT EMPLOYEES ARE USING
Before a city or county starts approving new AI projects, the playbook recommends finding out what is already happening inside the organization.
That means creating an inventory of AI tools, including products that may not be marketed as AI. The list could include chatbots, meeting transcription software, cybersecurity products, document management systems, permitting applications and AI features added to existing software. It should also account for tools employees may have started using on their own.
The playbook recommends that local governments adopt similar basic controls: approved accounts, clear rules about what information can be entered, employee training and a way to report problems. It also calls for reviewing vendors, including where their tools are hosted, what models they use and how data moves through their systems.
Policies alone are not enough, Deshpande said. Organizations need visibility into actual use, not just an assumption that employees are following the rules.
The guide proposes a phased 90-day starting plan. During the first phase, governments should name a leader responsible for AI, appoint a coordinator, form a review committee, issue interim employee guidance and begin the inventory. The next phase focuses on creating repeatable review processes, setting risk levels, establishing vendor requirements and providing basic training. The final phase calls for launching a small number of pilots, publishing an initial AI webpage and setting up incident response and reporting.
Source: Georgia Local Government AI Playbook
Deshpande stressed that the 30-, 60- and 90-day marks are not strict deadlines. They are meant to describe three phases that communities can move through at a pace that fits their resources.
NOT EVERY AI PROJECT CARRIES THE SAME RISK
The playbook also makes clear that a tool used to draft an internal email should not face the same review process as a system that could affect someone’s benefits, employment or access to public services.
It recommends four risk categories: low, moderate, high and prohibited. Low-risk uses, such as drafting or summarizing approved information with employee review, can follow simpler controls. Moderate-risk projects should receive review from a cross-functional AI committee. High-risk projects should get a more extensive assessment and executive approval. Prohibited uses should not move forward unless the governing body formally changes the policy after legal and public review.
Source: Georgia Local Government AI Playbook
The guide identifies situations that should automatically receive high-risk review, including systems involving law enforcement, surveillance, public safety, sensitive information at scale or decisions affecting people’s rights and access to services. It also calls for extra scrutiny when an AI system can take action, such as modifying records, sending official messages or spending money.
Human oversight is a recurring requirement, it says. AI-generated information should be checked by a qualified employee and AI should not be the sole decision-maker on matters involving benefits, employment, enforcement, safety or access to government services.
Kendrick said residents should know when they are interacting with AI and have a way to reach a person. “There should always be a human that is making part of the AI decision,” she said.
MAKING HELP AVAILABLE BEYOND STATE AGENCIES
The playbook is not intended to merely sit on a shelf. GTA hopes, Deshpande said, to offer office hours where local officials can ask questions and get help adapting its recommendations.
The guide includes templates and references that communities can adapt, including an AI use case intake form, vendor review questions, policy examples and training recommendations. It also encourages local governments to create a public register of significant AI uses so residents can understand where the technology is involved in government services.
Kendrick sees the playbook as a resource that can evolve, not a mandate for local governments to adopt AI. She doesn’t want communities to feel pressured to move faster than they’re ready. Instead, she wants local officials to weigh the risks and benefits and decide what works for their residents. Deshpande measures success by whether technology makes workflows easier, speeds up services and prepares employees for what’s ahead. More than simply making existing processes faster, he hopes AI will help governments rethink how they deliver services and reduce friction for residents and employees.
