One study that tested chatbots' ability to detect health misinformation failed more often than not in certain scenarios.