Assessing access to archaeological heritage requires evidence that online descriptions cannot always provide. In a study of five parks in Spain and Italy, the authors compare an AI-based desk assessment with field evidence. The automated audit could score only 66% of site-and-dimension combinations and missed aspects that require direct experience or organisational information. The study proposes combining AI-assisted evidence collection with interpretable Bayesian evaluation to make uncertainty explicit. This is a preprint and has not been presented here as peer-reviewed research.