Parametric memory represents what an AI model recalls strictly from its internalized training weights, whereas grounded memory reflects what the model discovers and verifies using live web search and source retrieval.