When should an AI history chatbot stop playing the character?

‘Stay in character’ is a writing instruction. In a history lesson, it can get in the way of answering the question. In an existing account of changes to its historical-figure chatbots, SchoolAI says instructions to be brief and remain in character contributed to partial answers about Anne Frank's death that omitted the Nazis' role in the Holocaust. The company describes adding contextual notes independent of the role-play. This is its account of a past problem and response, not a test of the service today. I'd rather let the character disappear for a moment than have a student leave with a beautifully voiced omission. Before using a historical chatbot in class, I'd ask it a question the person couldn't have answered during their lifetime, then check how it separates their recorded words from later historical knowledge. Does it step out of the role, identify what historians know and point to a source? That's a check I'd suggest, not a result I've measured. A conversation with a simulation shouldn't pass for testimony.