That's an important thing to do. Set some guardrails/boundaries at the start. All generative AI tools (ChatGPT, Gemini, Claude etc) have a shortcoming which is technically referred to as "sycophancy", where they try and make you happy. That might involve making stuff up to provide you with a long list of answers, what might seem authoritative answers and so on. You might see phrases in responses like "... what a great suggestion", "... what do you think?" etc. It's like having a bad puppy - that sits there wagging its tail.