Local governments expand AI use while addressing transparency concerns
City halls are adopting AI to sort service requests, manage records, translate communications, and answer routine questions. The goal is simple: faster responses, clearer information, and better allocation of staff time.
Early momentum has been strong. "There was general enthusiasm at first, particularly from elected officials, to try to integrate these tools quickly, to get more organizational efficiency out of them, and to try and scale," said Chris Jordan, program manager for AI and innovation at the National League of Cities.
What's working right now
- Dearborn, Michigan: A translation bot helps staff serve residents who don't speak English.
- Washington: An AI assistant makes open-data portals easier to search.
- Tucson Water, Arizona: AI flags pipes most likely to fail so crews can act before outages.
These are practical, narrow tasks with measurable outcomes. They free up people to handle exceptions and higher-value work.
Public trust is earned with clear standards
Jordan's team has seen AI perform best when cities set baseline rules: AI-specific privacy protections, consistent labeling in public-facing content, and plain-language explanations of how tools work. "Public listening sessions or task forces can also be appropriate for cities to use if there's a general sentiment of anxiety or distrust about certain technologies," he said.
Some cities are publishing their playbooks. Lebanon, New Hampshire posts a list of every AI tool in use. San Jose runs an annual review to gauge how algorithms affect residents. These are small moves that signal accountability.
The trust gap: accuracy and accountability
Mistakes are the biggest risk-both the perception and the reality. "Workforces do not trust generative AI outputs in many contexts and sensitive use cases," said Joe Scheidler, who is building Helios, an AI tool for policy work. His team focuses on reducing hallucinations and drift, while building verification, traceability, and provenance into the product experience.
For government teams, that means guardrails first, features second. If staff can't audit a result, they won't use it-nor should they.
Policy moves cities can implement now
- Adopt an AI risk framework and tier uses by impact (e.g., informational, operational, rights-affecting).
- Require plain-language notices and visible labels wherever AI touches the public.
- Stand up a cross-functional AI review group (IT, legal, DEI, records, comms, operations).
- Run pilot programs with clear exit criteria, success metrics, and a human-in-the-loop.
- Log every AI tool: purpose, data sources, retention, vendor, model version, contact owner.
- Mandate procurement checklists: bias testing, security attestations, audit access, uptime, and support.
- Set records retention rules for prompts, outputs, and training data derived from city content.
- Publish an annual AI accountability report and a public tool inventory.
Metrics that matter
- Service performance: time-to-first-response, resolution time, backlog reduction.
- Quality: accuracy rates, escalation rates, human edits per output.
- Equity: language coverage, accessibility compliance, disparate impact checks.
- Trust: resident satisfaction, complaint volume, clarity of AI disclosures (surveyed).
- Cost and capacity: staff hours saved, avoided overtime, vendor spend vs. outcomes.
Resident engagement that builds confidence
- Host listening sessions before launches, not after.
- Explain what the tool does, what it doesn't, and how a person can step in.
- Offer simple contact paths for corrections or appeals.
- Publish FAQs with examples of acceptable and prohibited uses.
Procurement and deployment checklist
- Run a privacy impact assessment and security review.
- Test for bias using representative datasets and edge cases.
- Provide staff training and a quick-reference guide for safe use.
- Enable audit logs, versioning, and reproducibility.
- Set rollback plans for outages or unexpected behavior.
Why now
According to the National League of Cities, interest is high-96% of mayors are exploring generative AI. The lesson from early adopters: start with narrow use cases, publish the rules, keep a human in the loop, and measure everything.
"For the most part, constituents want faster and more effective city services that make them feel good about how their tax dollars are being spent," Scheidler said. Deliver that, and trust follows.
Helpful resources
- NIST AI Risk Management Framework for risk tiers, controls, and governance structure.
- National League of Cities for municipal policy guidance and peer examples.
Upskilling your team
If your department is formalizing AI roles or building internal expertise, curated learning by job function can speed things up. See AI courses by job for structured options your staff can adopt without extra overhead.
Your membership also unlocks:
