Positioning GenAI in Academic Research Discourse: Layered Grammatical Patterns and the Limits of a Single Agency Gradient

Abstract

This study examines how generative artificial intelligence (GenAI) is grammatically positioned in Discussion sections of GenAI‐related academic writing research. From 134 eligible articles, 50 were selected through year‐by‐topic stratified probability sampling, yielding 969 AI‐related occurrences and 1026 grammatical relations. We analysed three layers: rank, clause function, and Subject status. Valence was unrelated to the nominal–clausal boundary and to Participant–Circumstance status, but was associated with Clause versus Embedded_clause positioning; this association remained in paper‐robust and within‐paper analyses. Positive and Negative claims did not differ reliably in Subject status. Concordance analysis showed that embedded AI relations could occur within propositions asserted in ranking clauses. The findings show that grammatical dimensions of AI positioning pattern differently across evaluative contexts and should not be collapsed into a single agency gradient.