Simulating Subjectivity: Discursive Ethos in Human-Authored and LLM-Generated Autobiographical Texts

Generative large language models (LLMs) reproduce many linguistic markers associated with subjectivity, yet the status of the textual voice they construct remains contested. This exploratory qualitative study compares two canonical autobiographical excerpts – Rousseau’s (2009) Les Confessions and Chateaubriand’s (2001) Mémoires d’outre-tombe – with two French-language autobiographical passages retrieved from the ChatGPT interface (OpenAI, 2026) and analysed as model-generated text. Drawing on an enunciative and discourse-analytical framework encompassing deixis, modalisation, evaluative and affective subjectivemes, and discursive ethos, the study examines deictic anchoring and evaluative and modal markers across the four passages. All four texts use first-person forms and evaluative resources. Within this corpus, however, the human-authored excerpts display denser biographical and sociohistorical anchoring, greater identity differentiation, and more conflictual modalisation, whereas the model-generated passages construct coherent but comparatively more generalised textual personae with more balanced evaluative patterning. The article proposes an operational framework for comparing discursive ethos in human-authored and LLM-generated autobiographical discourse while identifying limitations related to corpus design, prompt documentation, output selection, and historical and genre differences.