A study by Ayda Rastiemadabadi in the 2026 architecture journal Grid compared how built-environment academics and ChatGPT represented Istanbul’s Historic Peninsula, testing whether a language model could summarize public perceptions of the city from online reviews.

The researchers compared a verbal cognitive map drawn from 32 interviews with academics, noting possible sample bias, against ChatGPT’s analysis of 210 citizen reviews from TripAdvisor and Ekşi Sözlük. The model had not been trained on those reviews. Researchers used a structured, multi-step prompting workflow to have it extract, categorize and summarize locations and experiences, guiding and checking each stage themselves.

Using urban theorist Kevin Lynch’s five elements—paths, edges, districts, nodes and landmarks—the study found that ChatGPT recognized familiar features but shifted their prominence. Iconic buildings became less dominant, while comfort, crowding, safety and spatial atmosphere featured more centrally. The researchers say language models can summarize collective city images at scale and offer a reproducible way to connect interview-based cognitive maps with platform narratives, while stressing the need for critical, human-led interpretation in urban design.

A study by Ayda Rastiemadabadi in the 2026 architecture journal Grid compared how built-environment academics and ChatGPT represented Istanbul’s Historic Peninsula, testing whether a language model could summarize public perceptions of the city from online reviews.

The researchers compared a verbal cognitive map drawn from 32 interviews with academics, noting possible sample bias, against ChatGPT’s analysis of 210 citizen reviews from TripAdvisor and Ekşi Sözlük. The model had not been trained on those reviews. Researchers used a structured, multi-step prompting workflow to have it extract, categorize and summarize locations and experiences, guiding and checking each stage themselves.

Using urban theorist Kevin Lynch’s five elements—paths, edges, districts, nodes and landmarks—the study found that ChatGPT recognized familiar features but shifted their prominence. Iconic buildings became less dominant, while comfort, crowding, safety and spatial atmosphere featured more centrally. The researchers say language models can summarize collective city images at scale and offer a reproducible way to connect interview-based cognitive maps with platform narratives, while stressing the need for critical, human-led interpretation in urban design.