GPT 3.5 is more accurate at classifying tweets as liberal than it is at identifying posts that are conservative.

If you're going for rough approximation, LLMs are great, and good enough. More care and conventional ML methods are appropriate as the stakes increase though.