A new analysis by Copyleaks compared sample outputs from two leading generative AIs, Grok and ChatGPT, to determine if one model’s training data makes it easier for an AI detector to catch. The article explains that AI text is detectable through signals known as perplexity and burstiness. Low perplexity means highly predictable word choices, while burstiness measures how much sentence length and structure change throughout the text. Humans are naturally bursty, moving between short and long sentences, whereas AI tends to stick with a narrow subset of sentence length and structure, making lower burstiness likely AI-generated. Additionally, AI tools depend on “fillers, repetitive stereotypical expressions, and structured but reduced linguistic subtlety,” including words like “in addition,” “furthermore,” and “it is important to note.” While both Grok and ChatGPT use these transition words, each uses them in different patterns, creating a unique digital signature. The article notes that the Grok vs ChatGPT comparison cannot be limited to features because the two models differ in many other ways, but it does not provide the specific results of the analysis or state which model was flagged more.