The citation mining engine extracts and normalises grounded web search data across Google, OpenAI, and Anthropic by routing each provider's payload through dedicated structure parsers. It separates unselected sources (pages browsed during search) from selected citations (pages actively referenced in the response) by mapping text annotations, segment offsets, and support indices. Additionally, raw URLs are cleaned of tracking parameters and processed through a resolver that unwraps Vertex AI redirect URIs to accurately attribute citations to their canonical target domains.