Cognitive dissonance in large language models is neither cognitive nor dissonant

Lehr et al. (2025; hereafter LSHB) claim that GPT-4o exhibits “an analog form of humanlike cognitive selfhood,” based on its behavior in a classical cognitive dissonance paradigm (p. 1). In this brief commentary, we argue that this interpretation is a fundamental mischaracterization of large language models (LLMs), and that the observed results do not entail "cognitive" nor "dissonant" effects. We highlight that seemingly-irrelevant prompts features can influence LLM responses in extreme and unintuitive ways, and that such effects do not require an analog to a sense of self to explain. We additionally demonstrate that ChatGPT-4o exhibits effects of essay valence on its evaluations independent of the authorship of the essay, illustrating suggesting a general sensitivity to the content of prompts, rather than a specific dissonance between its authorship of an essay and its evaluation. We conclude that the alleged cognitive dissonance observed by LSHB in ChatGPT is neither cognitive nor dissonant.