Emotional Prompt Engineering: A Double-Edged Sword for Democratic Disinformation Defense

Vinay et al. (2025) reveal that the emotional tone of prompts dramatically alters the disinformation-generating capacity of large language models (LLMs). This dual-use insight—where polite/emotional prompts increase disinformation, but impolite ones reduce it—suggests a fascinating, underexplored research direction: can we deliberately engineer prompts or AI moderation strategies that "inoculate" public discourse against emotionally charged disinformation? This research would systematically test interventions where AI systems use emotional context to filter, flag, or counteract manipulative content, possibly even generating "antidote" messages in real time. Unlike prior work that treats emotional framing as a risk, this project explores its potential as a tool for resilience—while rigorously investigating the ethical and practical limits of such interventions. It directly synthesizes conflicting roles of AI (amplifier vs. mitigator) and paves the way for nuanced, context-aware disinformation defense.

References:

1. Emotional prompting amplifies disinformation generation in AI large language models. R. Vinay, G. Spitale, N. Biller-Andorno, Federico Germani (2025). Frontiers Artif. Intell..