Ageism in AI: corpus-based analysis of AI-generated essays

Abstract

Age-related language plays an increasingly important role in shaping societal assumptions about older people and may contribute either to the reinforcement or mitigation of ageist beliefs. The detrimental effects of ageist biases on wellbeing and social equity underscore the need for systematic examination from a linguistic perspective. Ageist biases have been explored in various contexts, such as social, intuitional and cultural. However, the linguistic realization of ageism has received less academic attention to date. The present study addresses this gap by investigating the lexical portrayal of age and later life in the corpus of nine AI-generated essays created by three large language models (ChatGPT (based on OpenAI’s GPT-5.2 model), Claude Sonnet 5 (developed by Anthropic), and Gemini 3 (developed by Google DeepMind)).

The analysis integrates quantitative and qualitative approaches. The former applies AntConc, a corpus analysis toolkit, to study the presence and frequency of age-related terms (blue-hair, codger, dinosaur, fossil, geezer, old fart, old hag, old-timer, senile, wise, old and age) in the nine AI-generated texts, while the latter employs S. Hunston’s evaluative and non-evaluative distinction to examine the evaluative meaning of age-linked terms and their contribution to ageism in the essays under examination.

The findings suggest that the target openly ageist and non-ageist terms are absent from the corpus. The only present age-focused term is age. It appears 27 times and depicts later life and retirement either neutrally or negatively. The research results also indicate that although the AI-generated texts do not possess explicitly ageist language items, they tend to generate implicit ageism in the representation of chronological age.

Keywords: ageism, age representation, age-related stereotypes, AI language, corpus-based analysis.