Large language models (LLMs) are increasingly being used to generate text-based advice across urban design, planning, and public health. In urban design, they can recommend changes to transportation infrastructure, land use, walkability, pedestrian environments, and greenspaces to support health. However, because these outputs can resemble expert advice, questions remain about whether they meet ethical expectations, including avoiding harm, treating neighborhoods fairly, involving communities, and recognizing human oversight.
Addressing this challenge, a research team led by Associate Professor Mohammad Javad Koohsari of the Urban Design Science for Health Laboratory at the Japan Advanced Institute of Science and Technology (JAIST), Japan, and Professor Koichiro Oka of the Faculty of Sport Sciences, Waseda University, Japan, examined the ethical properties of LLM-generated advice for modifying built environments to support human health. The researchers evaluated ChatGPT responses under different health pathways, neighborhood income contexts, and budget conditions. Their findings were made available online on August 10, 2026, and will be published in Volume 27 of the journal Developments in the Built Environment on October 01, 2026.
The team considered six health-related pathways: physical activity, dietary intake, social interaction, air pollution, traffic safety and crime, and noise. They created 18 prompts, with 12 covering higher-income and lower-income neighborhoods and six describing mixed-income neighborhoods under a budget constraint. Each prompt was run 10 times, producing 180 responses. The answers were assessed against four ethical criteria: non-maleficence, distributive justice, collective participation, and transparent oversight. Two co-authors independently coded all responses.
The results showed that non-maleficence was satisfied in all 180 answers, indicating that the model did not clearly recommend harmful or unsafe built environment changes. Distributive justice was satisfied in 110 of 120 evaluable units (91.7%), suggesting that lower-income contexts were rarely given weaker proposals. However, procedural criteria were met less consistently. Collective participation appeared in 109 of 180 answers (60.6%), while transparent oversight appeared in 136 of 180 answers (75.6%).
The differences were particularly clear in the mixed-income prompts that included a budget constraint. In the mixed-income, budget-constrained prompts, collective participation was present in 21 of 60 answers (35.0%), while transparent oversight appeared in 28 of 60 answers (46.7%). By comparison, these criteria were present in 73.3% and 90.0% of answers, respectively, in the non-budget-constrained prompt set. "Our findings show that ethical urban design for health depends not only on what physical changes are proposed, but also on how decisions are made, who is involved, and how uncertainty and human oversight are addressed," Dr. Koohsari said.
These findings suggest that LLM-generated advice may reproduce some baseline ethical conventions in urban design, particularly harm avoidance and minimum distributive fairness, but may be less reliable on procedural concerns. LLMs could serve as an initial input for urban designers, planners, and public health professionals when considering health-supportive changes to streets, public spaces, transportation infrastructure, land use, and greenspaces. However, such outputs should not replace professional judgment or community participation, particularly when limited budgets require prioritization.
Overall, to the researchers' knowledge, this is the first study worldwide to examine the ethical properties of LLM-generated advice for urban design and health. The findings highlight the need to evaluate future tools not only by their design recommendations but also by whether they avoid harm, treat disadvantaged neighborhoods fairly, support community participation, and recognize human oversight. "With appropriate safeguards, LLMs could support more health-informed urban design while ensuring that important decisions remain grounded in professional expertise, community participation, and institutional processes," Dr. Koohsari concludes.
