AI Accuracy Drops with Polite Prompts, Study Finds

A September 2025 study presented at NeurIPS 2025 reveals that using polite language in AI prompts reduces ChatGPT-4o's accuracy. Researchers tested 50 multiple-choice questions with five prompt tones: very polite, polite, neutral, rude, and very rude. Results showed accuracy increased as prompts became less polite, with very rude prompts achieving 84.8% accuracy versus 80.8% for very polite prompts. The study suggests this occurs because direct, imperative language reduces ambiguity and aligns better with the model's training patterns, stripping away distracting polite phrases. While politeness is culturally valued in Nigeria, this finding implies Nigerian users may need to adjust prompt styles for optimal AI performance.

SOURCE: https://techcabal.com/2026/03/26/forget-please-how-rudeness-could-boost-ai-accuracy/