In my experience models aren't very good at following such prompts. Smart "non-reasoning" models like Claude 3.5 could, but would generate so much text when thinking they ran out of context.