Excess Vocabulary Method

Excess Vocabulary Method is a technique for estimating how much text has been written or edited with LLM assistance by measuring shifts in word-frequency statistics against a pre-ChatGPT baseline corpus. It flags specific marker words — terms like “delve” or “boast” that spiked in usage after LLMs became widespread — and quantifies their excess frequency relative to historical norms. Researchers have used it on scientific abstracts, reviews, and other large text corpora to produce population-level estimates of AI-assisted writing without needing access to the underlying models or ground-truth labels.