While an LLM based AI may generate ideas that are fresh and new to you, it is designed and built to predict an answer to any prompt based entirely on the statistical probabilities learned from its data set. That means it favours the most predictable, statistically most common ideas from everything in its training data (millions of already published documents) and rewrites those for its response.
In that those predictable, already used ideas can be good, and may be ones that the person using the AI hasn’t thought of, that may still be enough, and good enough. But it is really important to know the difference, and to understand what LLM based AI are really capable of, and what they are not. A Crash Course in LLM-based AI