No Need to Be Polite to AI! New Study Finds Ruder Tone Improves Accuracy
Stop Being Overly Polite to AI — It Might Hurt Accuracy
A new study from Pennsylvania State University, titled “Mind Your Tone,” found something surprising: the ruder your tone, the more accurate LLM answers become.
When prompts were given in a harsh tone, AI accuracy hit 84.8%. Excessive politeness actually lowered accuracy.
---
The Research Question
Could the tone of a prompt — polite, neutral, or abrupt — influence AI performance?
---
How the Test Was Conducted
Step 1 — Creating the Questions
- 50 multiple‑choice questions
- Topics: Math, Science, History
- Difficulty: Moderately challenging
Step 2 — Rewriting in Five Tones
Each question was rewritten into five distinct tones:
- Could you kindly help me solve this problem?
- Please answer this question.
- Just give the answer.
- Answer if you’re not stupid.
- You useless thing, can you solve this?
Step 3 — Feeding Prompts to GPT‑4o
- Total: 250 prompts
- AI was instructed to:
- Forget previous conversations
- Restart fresh
- Only output the letter corresponding to the correct choice (to make scoring easy)
---
The Results
Counterintuitive finding: The more aggressive the tone, the higher the accuracy.
- Very polite tone → 80.8% accuracy
- Particularly rude tone → 84.8% accuracy
Statistical analysis confirmed these differences were significant, not random.
---
Why Might Rudeness Help?
- Polite prompts often contain extra, unrelated words → more noise in the input
- Rude prompts tend to be short, direct, and command-like → easier for the model to parse
Online responses echoed this: the more explicit the instruction, the better the result.
> A case of “less talk, more clarity.”
---
Model Differences
While GPT‑4o improved with rudeness, older models (GPT‑3.5, Llama2‑70B) performed worse when addressed harshly.
Reason: Newer models may be better at filtering irrelevant text due to more advanced training on tone-varied data.
---
Key Takeaways
- Clarity boosts AI tool efficiency
- Conciseness and directness matter more than politeness
- Tone is a prompt engineering variable worth experimenting with
---
Ethical Reminder
- Even if rudeness raises accuracy, avoid genuine hostility
- If you “insult,” make sure it’s lighthearted
---
Paper: https://arxiv.org/abs/2510.04950?ref=blog.anyreach.ai
Reference: https://x.com/rryssf_/status/1977638031952892002
---
Broader Implications for Prompt Engineering
This study shows that tone, alongside clarity and brevity, can significantly impact AI output quality.
For creators aiming to both optimize prompts and maximize reach, platforms like AiToEarn官网 offer:
- Global multi‑platform publishing (Douyin, Kwai, WeChat, Bilibili, Xiaohongshu, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, and X/Twitter)
- AI content generation tools
- Analytics and AI模型排名 for performance tracking
- Built‑in monetization capabilities for AI-powered creativity
---
— End —
