Did you ever try to ask a chatbot to rephrase your hint in its own words? If it prefers LLM generated text, surely that would help.

Maybe that's a triggering a method designed to mitigate against prompt injection / attacks involving poisoning.