When College of Natural Resources and Environment geospatial data scientist Junghwan Kim asked an artificial intelligence (AI) image generator to create a picture of Blacksburg, the result wasn't quite right.
"The image looked generic," Kim said. "It didn't capture what makes Blacksburg unique."
But when he asked the same system to create images of larger cities such as Richmond, Virginia Beach, and Washington, D.C., the results looked much more recognizable.
The images included familiar landmarks, waterfronts, and city features that reflected the character of those places.
That observation sparked a research question: Does AI do a better job representing large cities than smaller communities?
A new study from researchers at Virginia Tech, Hong Kong University of Science and Technology (Guangzhou), and the University of Alabama found the answer is yes.
The team discovered that AI-generated images were consistently better at representing larger metropolitan areas than smaller towns such as Blacksburg. The findings raise questions about how generative artificial intelligence tools portray places and whose communities are most visible online.
The study, published in Technology in Society , examined how OpenAI's DALL·E 2 image generator created images of three Virginia localities — Blacksburg, Richmond, and Virginia Beach — and Washington, D.C.
Researchers then asked residents to evaluate how realistic and recognizable the images are.
As generative AI tools become more common in travel planning, urban design, marketing, and public communication, Kim said these gaps in representation matter.
"People are increasingly relying on AI-generated content to learn about places," said Kim, assistant professor in the Department of Geography in the College of Natural Resources and Environment and the director of the Smart Cities for Good research group. "If smaller cities are not well represented in the data used to train these systems, then the images people see may not reflect the real identity of those communities."
The research team surveyed 129 participants, asking them to evaluate AI-generated images based on how realistic they appeared and how well they captured each city's identity. The images focused on elements such as landmarks, districts, paths, and waterfronts using urban design principles developed by planner Kevin Lynch.
The study found that AI struggled most with landmarks and culturally significant features. In Blacksburg, for example, there was no Hokie Stone featured on any of the university buildings.
Researchers also found that long-term residents were more critical of the AI-generated images than newer residents. Kim said that suggests people with stronger local knowledge are more likely to notice inaccuracies or missing details.
The findings point to a broader issue in artificial intelligence systems: communities with less online representation may also receive less accurate AI-generated content.
"AI systems learn from enormous amounts of online data," Kim said. "Larger cities tend to have far more images, media coverage, and digital documentation available online. Smaller towns often do not have the same level of representation."
Kim said the research highlights the importance of building more geographically comprehensive datasets and incorporating local perspectives into AI development. Without those efforts, AI-generated imagery could reinforce uneven representation between large urban centers and smaller communities.
The work also contributes to a growing conversation about the ethical use of artificial intelligence in planning and design. While AI tools can help generate ideas quickly and expand access to design technologies, researchers said the systems still have important limitations.
"Generative AI can be a powerful tool," Kim said. "But we also need to understand where it falls short and who may be left out."
Original study : doi.org/10.1016/j.techsoc.2026.103360
