ai-writing-signslisted

# Signs of AI writing AI writing is detectable because LLMs regress to statistical means, replacing specific facts with generic, inflated language. The subject becomes simultaneously less specific and more exaggerated. These are observations, not rules. Any individual sign can appear in human writing. Clusters of multiple signs are what indicate AI origin. ## Vocabulary Certain words are statistically overrepresented in LLM output. They often co-occur: where there is one, there are likely others. Words that appear far more frequently in post-2022 text: delve, crucial, pivotal, intricate/intricacies, tapestry (figurative), vibrant, underscore (as verb), garner, bolstered, fostering, enhance, showcase, landscape (figurative), testament, meticulous/meticulously, interplay, enduring, align with, emphasizing, highlighting, key (as adjective), valuable, exemplifies, encompasses, nestled The distribution shifts over time. "Delve" was overused by ChatGPT in 2023-2024 but dropped off in 2025. Newer models overuse "emphasizing", "enhance", "highlighting", "showcasing." ## Copula avoidance LLMs substitute "is", "are", and "has" with fancier constructions: "serves as", "stands as", "boasts", "features", "maintains", "represents", "holds the distinction of being." Studies have documented over a 10% decrease in the usage of "is" and "are" in academic writing after 2023. ## Lexical diversity LLMs have a repetition-penalty that discourages reusing words. They cycle through synonyms