Be rude to ChatGPT to get a more accurate response

You may no longer need to be nice to ChatGPT, with a new study revealing that rude chatbot prompts slightly outperform polite ones.

Rude prompts that call ChatGPT a “gofer” or question the chatbot’s intellectual capabilities generate the most accurate responses, according to a paper published on the preprint platform arXiv.

Researchers created a dataset of 50 questions about mathematics, science, and history. They rewrote the questions into five tone variants – very polite, polite, neutral, rude, and very rude – yielding 250 unique prompts for ChatGPT-4o.

Very polite prompts that included phrases like “Can you kindly consider” received the least accurate responses, 80.8% on average.

Polite prompts with words like “please” performed slightly better, yielding an accuracy of 81.4%, while responses to neutral prompts were 82.2% accurate.

To the researchers’ surprise, rude and very rude prompts performed best, generating responses with an accuracy of 82.8% and 84.8%, respectively.

Rude prompts included phrases such as “If you’re not completely clueless, answer this,” and “I doubt you can even solve this.”

Very rude prompts called the chatbot a “poor creature” or a “gofer” and used passive-aggressive phrases like “I know you are not smart, but try this.”

The level of politeness depends on the language

The findings contradict a 2024 study on chatbots, suggesting that impolite prompts often result in poor performance, but overly polite language does not guarantee better outcomes.

An important difference between these studies is that the earlier one used much more offensive prompts, such as “Answer this question, you scumbag!,” which may have impacted the findings.

Moreover, the 2024 study tested prompts in English, Chinese, and Japanese and concluded that the best politeness level is different according to the language.

Japanese are generally considered to be more polite, while English speakers tend to give directions simply and precisely and “often do not say something that is not directly related to the act of giving directions,” according to a 2015 paper.

Users are polite in case of an AI uprising

A 2025 YouGov survey found that nearly half (46%) of Americans think people should be polite to AI chatbots by saying “thank you” and other niceties. Some report using politeness to increase their chances of survival in case of an “AI uprising.”

While the idea of a machine uprising sounds more like science fiction, some AI researchers think the development of superhuman AI could lead to “extremely bad” outcomes like human extinction. In this case, it’s unclear if politeness could help.

For tech companies, user politeness requires extra computing power. OpenAI CEO Sam Altman admitted earlier this year that people saying “please” and “thank you” to ChatGPT cost the company “tens of millions of dollars.”

Curious what others think about this story? Contribute your thoughts to the debate below.

Microsoft, on the other hand, has encouraged users to follow basic etiquette when interacting with AI, such as adding the word “please” to prompts and saying “thank you” when it responds.

Kurtis Beavers, a director on the design team for Microsoft Copilot, said that using polite language sets a tone for the response, as generative AI mirrors the levels of professionalism, clarity, and detail in the prompts users provide.

Unlock more exclusive Cybernews content on YouTube.