Common practices for isolating long-term memory by user include metadata filtering by user ID in a vector database, creating a separate collection for each user, or enforcing tenant constraints at the retrieval layer.