Parametric memory recall measures unprompted model knowledge using token-level log-probabilities without search tools, whereas grounded memory evaluation assesses facts established via real-time web retrieval against model recall to identify brand gaps and inaccuracies.