Large Language Models as Unreliable Narrators: Toward a Postdigital Literacy

The public release of ChatGPT in 2022 brought large language models (LLMs) into everyday textual communication, accelerating the proliferation of machine-generated discourse online and offline. Because LLM outputs often circulate without clear markers of origin, distinctions between human and artificial authorship have become increasingly unstable, complicating judgments of textual reliability and trust. This article argues that narratology (especially the concept of unreliable narration) offers a productive framework for understanding these developments. While public and scholarly debates frequently characterize chatbots as “hallucinating,” “lying,” “fabricating,” or “bullshitting,” we propose instead that LLMs can be understood as rhetorically unreliable narrators. Their unreliability does not stem from intention or deception, but from the probabilistic mechanisms through which they generate persuasive, authoritative-sounding discourse.

Drawing on rhetorical narratology, particularly the work of James Phelan, we reconceptualize unreliability as an effect emerging in communicative exchanges between textual systems and audiences rather than as a property grounded in authorial psychology. We argue that existing models of narrative communication must therefore be revised to account for hybrid, distributed forms of narration in which agency is shared among training data, platform infrastructures, interface design, and user interaction. In this context, reliability becomes a negotiated and socially distributed judgment shaped by interpretive communities operating under contemporary post-truth conditions.
To develop this argument, we analyze outputs produced by ChatGPT and related models through Phelan’s tripartite framework of unreliability, organized around facts/events, understanding, and values. This framework enables us to map distinct forms of chatbot unreliability and to identify the multiple technical, rhetorical, and social levels at which such unreliability emerges. By reframing LLMs as unreliable narrators, the article contributes to ongoing debates about AI-generated discourse and trust in digital culture.

Speakers: Inge van de Ven & Siebe Bluijs