Have you gotten better at doing this without the LLM, maybe even extemporaneously? Wondering if enough exposure to that kind of modeling also serves an educational role.
Sentiment analysis did that for me. I ran Watson on some challenging emails and worked to remove tones of anger and contempt. After a couple of times, I internalised it.
Oh yeah, since there is a built in feedback loop with the person I am interacting with, I was able to start recognizing patterns of how to shift ways that I inherently think/phrase something to be better received by others.