Text, Textualisation and Human Agency in the Age of Generative AI

Abstract

This article discusses to what extent the products of generative AI can be considered texts from the perspective of text linguistics. Starting from a pedagogical dialogue held in a Text Linguistics course at NOVA FCSH, the study analyses responses produced by three generative AI models – ChatGPT 4, ChatGPT 5, and Gemini 2.5 – on questions concerning the notion of text, the distinction between generating, producing, and making text, authorship, and the context of enunciation. The theoretical framework draws on authors from Anglophone, Francophone, and Lusophone traditions, with particular emphasis on Halliday and Hasan, Beaugrande and Dressler, Bronckart, Adam, Rastier, Marcuschi, Coutinho, and Miranda. The analysis shows that the models tend to privilege a formal and functional view of text, focused on criteria of textuality, while downplaying its empirical, social, praxeological, and agentive dimensions. The article concludes that generative AI may generate plausible textual sequences, but it does not produce texts.