Walkable Intelligent Parks: Can a Large Language Model Turn Urban Park Audit Findings into Actionable Recommendations?

Park audits can inform neighbourhood-scale urban design strategies that support health. However, interpreting park audit data is time-consuming and requires specialised expertise. Recent advances in large language models suggest potential for assisting with structured data interpretation, but their application to park audits remains unexplored. This study examined whether a large language model (ChatGPT-5) can assist in interpreting park audit data from public parks in Dhaka City, Bangladesh. Using a modified park audit tool, 59 parks were audited, of which 26 met the predefined completeness threshold and were included in the analysis. ChatGPT-5 and human experts received identical park audit datasets and instructions for interpretation. Outputs were evaluated using content analysis and a pre-defined scoring framework to assess accuracy and comprehensiveness. Paired t-tests compared ChatGPT-5 outputs with the expert benchmark. ChatGPT-5 showed no statistically significant difference from the expert benchmark in overall accuracy in this sample (p = 0.13). The unadjusted analysis showed a higher recommendation-match count for ChatGPT-5 (p < 0.05), but this result should be interpreted as exploratory because multiple criteria were tested. No statistically significant differences were detected across accuracy domains or other comprehensiveness criteria. These findings suggest that ChatGPT-5 may assist with preliminary interpretation of structured park audit data. However, the results do not establish equivalence with expert assessment. Expert review remains necessary to assess feasibility, contextual relevance, and alignment with local planning standards.