A small idea that started as a letter to the editor, but the reviewers were not convinced. 🙂 If lexical markers of LLM-assisted writing change as users and models adapt, can we really use them for longitudinal measurement? I call this the Marker Drift Problem. No large empirical study (yet). For now, a short preprint putting the problem on the table: :doi: https://doi.org/10.31222/osf.io/xs64q_v1 #Metascience #GenAI #LLM #ScholarlyCommunication #ResearchIntegrity #preprint