Early LLMs were so much better for this kind of stuff.

I had a really good thing going with the old ones generating plausible random data. But now they generate the same 15 things over and over, so I went back to concatenating randomized content from lists.