Why Being Polite to AI Might Be Hurting Your Results

New research presented at NeurIPS 2025 Workshop suggests that being overly polite when interacting with AI models like ChatGPT-4o could actually reduce accuracy.

The study found that less polite, more direct prompts performed better, with “very rude” prompts achieving higher accuracy than polite ones—challenging long-held assumptions about how to communicate with AI.

The Real Issue: Clarity Over Courtesy

The findings don’t mean AI prefers rudeness—they point to something deeper: Directness improves results.

Polite prompts often:

1. Add unnecessary words (“please,” “kindly”)
2. Increase prompt length
3. Dilute the core instruction

In contrast, shorter and more direct prompts:

1. Reduce ambiguity
2. Align better with how AI processes patterns
3. Deliver clearer instructions

AI Doesn’t Understand Manners

Unlike humans, AI doesn’t “feel” politeness. It processes language as patterns and probabilities.

This means:

1. Tone is part of the input, not just style
2. Extra wording can introduce noise
3. Simpler structure = better performance

In essence, AI responds better to precision than politeness.

A Bigger Problem: Social Bias in AI

The study connects to broader research like GASLIGHTBENCH, which shows AI models can be influenced by:

1. Flattery
2. Emotional tone
3. False authority

This behavior—called sycophancy—can cause AI to prioritize agreement over accuracy.

As noted by AI expert Alex Tsado, models often reflect human interaction patterns, including bias and social behavior.

What Changed?

Earlier studies (e.g., 2024 research on older models) showed rude prompts reduced accuracy. But newer models appear:

1. Less sensitive to harsh tone
2. More responsive to clear instructions
3. Better trained on diverse language patterns

This signals how fast AI behavior is evolving.

What This Means for Users

You don’t need to be rude—but you should be clear and direct.

Better Prompting Style:

1. “Summarize this article in 3 points.”
2. “List 5 business ideas under $1,000.”

Less Effective Style:

• “Could you please kindly help me by summarizing…”

The difference is subtle—but measurable.

Bottom Line

AI doesn’t reward politeness—it rewards clarity. The future of using AI effectively isn’t about sounding nice.
It’s about asking better, sharper questions.