The value might not be in the ability of the model to correctly classify people but in it's ability to generate discussions about the classification.

"I'm a Gemini which fits because I am X but unlike most Gemini I like Y" tells you X and Y about someone.

Saying "Gemini and Aeries shouldn't date because of..." isn't necessarily a conversation about whether any given individuals should date. It may instead actually be a conversation about whether people who conform to certain sets of traits should date.