Does Personality Persuade? Investigating Personality Trait Prompting and Persuasion in Large Language Models
* Equal contribution
Preprint
Abstract
Large language models (LLMs) can be prompted to express different personality profiles, but it remains unclear how reliably they follow these prompts and whether the personality they express affects their persuasiveness. We address these questions in two evaluations. First, we assess Big Five personality prompting with the IPIP-NEO-120, comparing prompt formats across repeated generations and measuring trait control, cross-trait spillover, and unconditioned model profiles. Second, we conduct a within-subjects study in which participants discuss morally ambiguous decisions with LLM agents prompted to exhibit personality profiles aligned to the participant, mismatched, neutral, or unconditioned. Our calibration experiments show that prompted personality profiles can be highly reproducible and strongly responsive to requested trait levels, although manipulating one trait can also cause smaller changes in other traits. We further find that unconditioned LLMs do not constitute personality-neutral baselines: the models examined exhibit systematic and highly consistent personality profiles even without explicit personality instructions. In the user study (N=32), personality alignment did not significantly affect perceived persuasiveness, stance switching, or changes in decision confidence. Participants changed their initial stance in only 8 of 128 conversations, and we found no significant associations between their OCEAN personality traits and susceptibility to persuasion. However, higher self-reported AI persuasion literacy was associated with less frequent changes in decision confidence. An exploratory analysis further revealed that agents predominantly relied on logical appeals, with other persuasion strategies varying across conditions and dilemmas. Our findings highlight the importance of distinguishing between an LLM's ability to express a prompted personality and its actual influence on users, and suggest that AI persuasion literacy may influence how users respond to persuasive AI systems.
Citation
@misc{boone2026doespersonalitypersuadeinvestigating,
title={Does Personality Persuade? Investigating Personality Trait Prompting and Persuasion in Large Language Models},
author={Ruben Boone and Gilles Eerlings and Sebe Vanbrabant and Davy Vanacken and Kris Luyten},
year={2026},
eprint={2611.06572},
archivePrefix={arXiv},
primaryClass={cs.CY},
url={https://arxiv.org/abs/2611.06572},
}



