The term "AI slop" has become a common way to describe low-quality AI-generated text, yet there is no agreed-upon definition or method for measuring it. A new arXiv paper takes on this gap by developing a taxonomy for what counts as slop, offering a structured way to identify and quantify its occurrence.
The authors argue that without a clear definition, discussions about slop remain vague and hard to evaluate. Their taxonomy is intended to give researchers and practitioners a shared framework for classifying text quality, which could support more consistent measurement across different contexts and applications.
This work is a step toward turning a fuzzy concept into something testable. While the abstract does not detail the taxonomy's specific categories, the proposal itself signals a move from casual usage toward systematic analysis of AI-generated text quality.