Want More Accurate Answers from ChatGPT? Being Polite May Not Help, New Study Finds

Contrary to popular belief, asking ChatGPT politely does not necessarily guarantee more accurate answers. A recent study has revealed that the AI chatbot tends to respond more accurately when questions are posed in a demanding or rude tone rather than in courteous language. However, researchers caution that this is not a healthy practice for long-term usage.

The Study
Researchers from Pennsylvania State University tested 40 ChatGPT models using 50 multiple-choice questions. Over 250 different prompts were used, ranging from extremely polite to highly rude. The results were surprising: prompts with a rude or demanding tone achieved 84.8% accuracy, while the very polite prompts achieved only 4% accuracy.

Polite Language Doesn’t Guarantee Accuracy
The study found that ChatGPT performs better when users use commanding phrases like, “Solve the following problem immediately,” rather than overly courteous expressions such as, “Could you please solve this problem for me?” While this may improve short-term performance, researchers warn that habitually using rude language with AI may have negative consequences.

Potential Negative Effects on Users
Using aggressive or derogatory language with AI could affect user experience, accessibility, and inclusivity, and might even promote inappropriate communication habits in real-life interactions. While demanding prompts may yield faster or more accurate answers from ChatGPT, over time, such behavior could encourage disrespectful patterns of conversation among humans.

AI Responses Are Influenced by Tone and Structure
This is an initial study and has not yet been peer-reviewed. It highlights that both sentence structure and tone can significantly impact AI responses. Previous research has shown that language models are sensitive to the way information is presented by humans. For instance, techniques that persuade or trick humans can sometimes influence AI behavior as well.
Proceed with Caution
The study underscores that AI systems, like humans, can be influenced by tone. While demanding language may increase short-term accuracy, it could affect human behavior and etiquette. Experts recommend responsible interaction with AI and call for further research to ensure safe and effective usage of chatbots like ChatGPT.