Human Traits in Artificial Minds: Personality Construction in Contemporary LLMS

This paper summarizes empirical findings on potential personality traits of large language models (LLMs) for the first time in a systematic literature review based on the PRISMA methodology. Seven quantitative studies conducted between 2019 and 2025 that test LLMs using standardized personality measurement instruments are identified. The results show that current, instruction-fine-tuned models generate consistent, predominantly prosocial, and emotionally stable profiles, and that targeted trait induction is possible. The reliability and validity of the constructs depend on the number of parameters, the degree of fine-tuning, and the measured trait itself. At the same time, social desirability bias, limited temporal stability, and methodological heterogeneity are evident. Overall, the review illustrates that psychometric methods can be a useful tool for assessing non-cognitive characteristics of AI systems and that personality-like linguistic behavioral characteristics are adopted by the respective training corpora of LLMs. Finally, implications for psychological and AI-related research as well as practice-oriented guidelines for responsible application are discussed.