Behavioral patterns in cultural heritage engagement with general-purpose conversational AI

<p dir="ltr">This dataset contains behavioral coding data from the study 'Behavioral Patterns in Cultural Heritage Engagement on General-Purpose AI Platforms.' The dataset includes conversation-level relevance screening results for 1,146 heritage-related conversations extracted from three public AI conversation corpora (WildChat, LMArena, ShareChat), and turn-level behavioral category assignments for 1,454 user turns across 396 relevant conversations.

Two independent coders applied an eight-category deductive behavioral coding scheme comprising Factual Inquiry, Explanatory Inquiry, Guidance Seeking, Content Production, Creative Engagement, Evaluation and Correction, Personal Expression, and Meta-Interaction/Off-Topic. Both coders' independent assignments are provided to enable reliability verification. The complete codebook, including category definitions, decision rules, positive and negative examples, and the keyword list used for heritage conversation extraction, is included.