[IAmGraydon]

The em-dash is one marker, but I’ve read that most LLMs create small but statistically detectable biases in their output to help them avoid reingesting their own content.