Election Season Is Coming: Is Your Polling Place Finder Ready?
The most common question voters ask shouldn’t require your staff to answer it
Every election cycle, the same pattern repeats. Voter registration deadlines approach, early voting opens, and then Election Day arrives—each milestone triggering a wave of calls to county election offices. The question is almost always the same: “Where do I vote?”
It’s a reasonable question. Polling locations change. Precincts get redrawn. People move. And for all the investment in voter information systems over the years, most jurisdictions still rely on web interfaces that require voters to find the right page, enter their address correctly, and interpret the results. Many voters give up and just call.
This creates a familiar problem for election administrators: the weeks when public interest in elections is highest are exactly when staff capacity is most strained. Temporary workers handle overflow. Hold times lengthen. Mistakes happen. And every minute spent telling someone their polling location is a minute not spent on the hundred other tasks that make elections run.
The math doesn’t work
Consider the arithmetic. A mid-sized county might have 500,000 registered voters. If just 2% of them call with polling place questions during the three weeks before an election, that’s 10,000 calls. At five minutes per call—looking up an address, confirming the location, spelling out directions—that’s over 800 hours of staff time. For a question that has exactly one correct answer sitting in a database.
The traditional responses to this problem have been partial solutions. Better websites help, but many voters don’t find them or don’t use them correctly. Automated phone systems help, but voice recognition struggles with addresses. Social media posts help, but reach only those already following. Each approach captures some fraction of the demand, leaving the rest to pour into phone lines.
What changes the calculation is a system that can handle natural language questions, at scale, around the clock, drawing on the same authoritative data that staff would consult. The technology to do this now exists, and it’s simpler to deploy than most election officials might assume.
Your data already has the answers
Here’s what makes this tractable: if your jurisdiction uses ArcGIS for elections data—and most do—you already have everything necessary. Polling place locations, precinct boundaries, address-to-precinct mappings. The data is there. It’s maintained. It’s authoritative. The question is just how to put it in front of voters in a form they can actually use.
An AI agent connected to your ArcGIS services can accept questions in plain language—“Where do I vote if I live at 742 Evergreen Terrace?”—and return the same answer your staff would give after looking it up manually. The difference is that it can do this for thousands of voters simultaneously, at 2 AM on the Sunday before Election Day, in whatever language the voter prefers to use.
That last point is worth pausing on. A voter who speaks primarily Spanish, or Vietnamese, or Tagalog, can ask their question in their language and receive an answer in their language. The underlying polling place data is the same; the AI handles the translation as a natural byproduct of how language models work. No separate translation layer. No additional configuration. A voter types their question, and the system responds.
What this looks like in practice
Tarrant County, Texas—population over two million—has pioneered this approach. Their elections AI agent connects directly to county GIS services, providing real-time answers about polling locations and voting precincts. A resident can ask a simple question and get back not just a polling place name and address, but driving directions, hours of operation, and early voting options.
The implementation runs on the county’s existing data infrastructure. There’s no migration, no data transformation, no parallel systems to maintain. The ArcGIS services that power the county’s internal GIS applications are the same services that power the voter-facing AI. When staff update a polling location in the source system, the change is immediately reflected in voter responses.
Delaware is now following a similar path, adapting the same architecture to state-level elections data. The pattern is transferable because it’s built on standards—ArcGIS REST services that most jurisdictions already expose, wrapped in an AI integration that translates voter questions into spatial queries.
The questions administrators are asking
When election officials first see this, the concerns tend to be consistent. What about accuracy? What about security? What about cost?
On accuracy: the AI draws exclusively from your authoritative GIS data. It’s not making guesses. If a voter asks about a location, the system runs the same spatial intersection query that your GIS analyst would run—checking which precinct contains that address, and which polling place serves that precinct. The AI adds a conversational interface; it doesn’t add a new source of truth.
On security: the system is read-only. It queries your GIS services but cannot modify them. Voter interactions are transient—there’s no database of who asked what. The architecture is simpler, and therefore smaller in attack surface, than many existing voter information systems.
On cost: this is where the comparison becomes stark. A part-time seasonal worker answering phone calls costs money. Software that answers the same questions at 3 AM costs considerably less per interaction. The economics favour automation most strongly during exactly the high-volume periods when elections offices are most stretched.
Beyond the obvious question
Polling place lookup is the obvious starting point, but it’s not the only application. The same infrastructure can handle adjacent questions that voters ask: “Am I registered to vote?” leads to a lookup of voter registration data. “What’s on my ballot?” leads to candidate and measure information. “Can I vote early?” leads to early voting locations and hours.
Each of these questions follows the same pattern: public data that exists in authoritative form, currently trapped behind interfaces that voters struggle to navigate. An AI layer doesn’t replace the underlying data systems; it makes them usable.
And because the technology is built on open standards—MCP for AI integration, ArcGIS REST for data access—jurisdictions aren’t locked into proprietary solutions. The same codebase that serves one county can be adapted for another. The cost of the second implementation is a fraction of the first.
The window is open
For the 2026 election cycles, the window is open in a way it wasn’t before. The AI technology has matured. The integration patterns are proven. Several jurisdictions have already deployed production systems. The path from pilot to production is measured in weeks, not years.
What makes this moment unusual is the combination of mature technology and acute need. Election offices are being asked to do more with less, facing a staffing environment that has only gotten harder since 2020. Tools that multiply staff capacity aren’t luxuries; they’re necessities.
The question of “where do I vote?” has always had an answer. The challenge has been getting that answer to voters who need it, when they need it, in a form they can use. For the first time, that’s becoming straightforward.
The elections AI agent is available now for jurisdictions using ArcGIS. Implementation typically requires no data migration—only configuration of the connection to existing GIS services.