What is AI training but “parsing content” for “delivering generalized insights”? They intentionally use slippery language that can defend their practices.