LLMs often add disclaimers like "I'm just an AI" when asked about themselves. A chat template acts like a switch for this behavior across eight open-source models. Its presence turns disclaimer voice up and experiential voice down. Its absence does the opposite. Researchers found a specific activation direction that steers this behavior. Removing it reduces disclaimer voice, while adding it increases it. Models without a chat template disclaim like one was present when the direction is added. The study warns that model self-reports may be confounded by the chat template. It concludes that model self-descriptions should not be treated as literal facts.