Spatial intelligence has always belonged to specialists. That may finally be changing, but there are some hard problems.

Last month, I spoke with a public agency in a frontier county in the Southwest that serves a large, geographically dispersed rural population. Its team carries enormous responsibility for the health of the people who live there, and they knew geography was central to their work. They wanted GIS and had wanted it for years. What they lacked was the budget to hire GIS specialists and the capacity to implement, learn, and maintain a complex platform. This was not an unusual or isolated case.

According to NACCHO’s 2024 Public Health Informatics Profile, only 38% of local health departments reported that staff within their own department use GIS, while 28% said GIS is not performed at the department at all. With more than 3,300 agencies meeting the definition of a local health department, that points to a large and mostly invisible gap between the need for geographic reasoning and the ability to act on it. And it is not only public health.

I see the same gap across nearly every healthcare and community based organization I work with, and well beyond healthcare, from small businesses to large enterprises that have simply never had the opportunity to use GIS in a meaningful way. The most common complaint, not surprisingly, is that the existing GIS stack is complex, has a steep learning curve, and requires specialized expertise. Many leaders still think GIS is “just a map.” What a missed opportunity.

GIS has always been a specialist’s tool

Geographic knowledge has long belonged to experts. From the Roman surveyors who helped governments measure land, organize taxation, and plan the movement of armies, to today’s technology companies optimizing routes at planetary scale, geographic information has usually been produced and interpreted by professionals.

Modern GIS made that knowledge vastly more powerful. Spatial databases, satellite imagery, remote sensing, network analysis, and demographic modeling turned maps into computational systems. But one thing stayed constant: GIS was built for specialists. Most platforms still assume the user is fluent in projections, spatial joins, buffers, isochromes, and raster data. For a GIS professional, these skills are core to their profession. For a public health director, a nonprofit executive, or a clinic administrator, it is a big wall.

None of those leaders wants to configure a network analysis. They want answers to practical questions: Which communities have the greatest unmet need? Where should services expand? Which providers can reach a particular population? The appetite for geographic reasoning is obvious. The traditional workflow is the obstacle.

Consumer mapping already proved what happens when that big wall comes down. Billions of people use Google Maps and Apple Maps every single day without knowing what a coordinate system is. But consumer tools solve standardized problems, such as directions, nearby places, travel time. Organizational spatial analysis is way messier. Agencies, health systems, insurers, and logistics operators need to combine their own operational data with demographic, environmental, and service-access information, and their questions rarely have a one-step answer. That enterprise territory is where the opportunity remains almost entirely unopened.

Conversation Changes the Interaction Model

Large language models offer something more significant than a faster workflow. They change where the interaction begins and how users interface with GIS. Instead of starting with a map, a toolbar, and a list of geoprocessing functions, a user can start with the question: Where are older adults living more than 30 minutes from memory care? Which counties have growing Alzheimer’s populations but limited provider capacity?

Of course, behind the scenes, the system still has to identify the right datasets, calculate travel times, perform the spatial joins, and weigh service capacity against demand. The spatial science does not disappear, and neither does the expertise behind it. What changes is that the user no longer has to translate a real-world question into a sequence of specialized operations. Conversational GIS does not simplify the science. It simplifies access to it, and its greatest value is unlikely to be helping analysts finish familiar tasks a little faster. It is bringing in an entirely new group, such as public health leaders, aging-services organizations, emergency planners, community groups, small businesses, and local governments. These are people already making decisions shaped by distance, access, and place, even though GIS itself was never within their reach.

But, there are some hard problems

Several hard problems still exist. The first is interaction design. We still know very little about what it takes for an enterprise user to work with GIS, or even a plain map, through conversation in a meaningful way (I call this map-chat interactions). A simple example exposes the depth of it: if a user asks about three locations in a single question, what should the map now show? Keeping the map and the conversation aligned with what the user truly intends, and inferring what they need in that moment from limited information, remains largely a black box.

The second is reliability. LLMs are probabilistic by nature, which is exactly the wrong property for analysis meant to guide where a community invests scarce resources. Multi-agent architectures are emerging as a more robust way to give these systems the right tools and structure, but they do not remove the need for human expertise, especially for quality assurance.

The third is AI evaluation. Without a skilled person who actually knows how do to this in the loop, how does anyone know whether the underlying data, and the answer built on it, are actually correct? For non-experts to trust the output, we need a highly accessible evaluation framework that makes the reliability of an answer legible to the person receiving it.

Anything that can go wrong, will go wrong. A subtly incorrect spatial join, delivered in fluent prose, is more dangerous than an error buried in a toolbar, because it looks authoritative. Data coverage is thinnest in precisely the underserved places we most want to help. And when methodology all of a sudden disappears behind a well-versed conversation, so can the important accountability. The answer to all of this is not to hide the complexity but to keep the work visible: show the data, the assumptions, and the steps, and keep expertise close to the decision in a meaningful and accessible way.

For decades, GIS has asked people to learn the language of geospatial systems. Conversational GIS points toward the opposite, systems that begin to learn the language of the people they are meant to serve. That may be the next real evolution of this field: not only making spatial analysis more powerful, but making spatial intelligence reachable for the organizations that have needed it all along. Getting there honestly means solving the hard parts in the open.