Yeah, the LLMs “thought” process does not distinguish between internal monologue and presentation well.

That’s why you funnel it into a template.