Influence Tactics in LLM Code Generation: A Study on Prompt Framing

A recent preprint on arXiv (2608.11513) explores the potential of psychology-driven influence strategies to enhance prompt framing for Large Language Models (LLMs) in coding tasks. The research applies eight influence techniques from Yukl & Falbe's framework—including rational persuasion, ingratiation, and exchange—into reproducible prompt formats. These formats were tested on five prominent open-weight LLMs using two benchmarks: LiveCodeBench and SWE-bench Verified. The generated code was evaluated based on four software quality criteria, although the abstract does not specify these dimensions. This study delves into the previously unexamined link between psychological persuasion methods and LLM efficacy in software engineering, indicating that prompt phrasing can significantly influence model outcomes. The results may guide developers toward better prompt engineering strategies.