Synthetic English and LLM‑mediated writing: A research agenda for register variation and enregisterment

Abstract

Routine use of large language models (LLMs) is changing how texts are produced and revised, and comparative research has begun to identify recurrent distributional differences between LLM‑generated and human‑authored English. This article uses the provisional label synthetic English (SE) for recurrent linguistic distributions associated with LLM‑mediated writing. LLMs operate within existing registers and may redistribute linguistic resources within and across them. Research on stance, textual organization and nominal and informational packaging provides an empirical starting point for this research program, although the direction and strength of these differences vary with genre, task, model and production conditions. The proposed agenda examines linguistic distributions, production processes, circulation, persistence, recognition and social evaluation. Current research provides stronger evidence for recurring production patterns and linguistic distributions than for shared recognition or social meaning. Professionalism, institutional legibility, trust and perceived authenticity are therefore treated as questions for investigation. The article outlines a research agenda using corpus comparison, writing‑process evidence, reader judgments and change over time. It closes by considering how rhetorical comparison can help writers examine what LLM‑mediated revision changes in their texts and whether those changes suit the writing situation.