A marketing manager's prompts keep producing off-target results, even though the prompts seem clear to her. According to Anthropic's prompt-engineering guidance, what is the best way to test whether a prompt is actually clear?