Better prompting does provide more balanced responses to a certain extent but users looking for validation often subconsciously leave bread crumbs that the more powerful models pick up on.