from the paper: high chatGPT scores on the descriptions are generally correlated with high citation counts (which the paper uses as ground truth for quality)

from the paper: high chatGPT scores on the descriptions are generally correlated with high citation counts (which the paper uses as ground truth for quality) despite being part of the training criteria, novelty claims in book descriptions don't lead to higher chatGPT scores, but they do correlate with higher citation counts. books with indirect writing styles ("the author shows") in the descriptions tend to get lower scores.