When I was in formation a couple of years ago, I showed our homiletics instructor a ChatGPT-generated homily for our assigned text. He read through it and put his head on the desk. Then he handed it back to me and said it was as good as good as anything you'd hear from the ambo that Sunday.

By this, he meant that it was ok-but-not-great, and there's a lot of weak preaching out there. And your point is dead on: the text and the assembly are the primary considerations. I preach on the same readings to 4 different masses, but the 4:30 Saturday Vigil folks are a different group than the 11:30 Sunday Morning crowd, so the message is tuned accordingly. Different emphases, different touchstones, differing exhortations, etc.