Should you say “please” and “thank you” to ChatGPT or other artificial intelligence models? It's a question that may seem strange, but scientific studies have also been conducted on it.
According to a BBC analysis, much of the advice circulating online about how to talk to chatbots – from praise, threats to role-playing – is more myth than method that works.
There are no "magic words"
Researchers have tested dozens of ways to improve artificial intelligence responses. They have complimented the models, asked them to "think carefully," and even had them imagine they were in the Star Trek universe.
The results were surprising: almost none of these techniques produced lasting improvements. Only in a few very specific cases, such as basic math exercises, did role-playing yield slightly better results.
Politeness doesn't make AI smarter
In 2025, a user on the X network humorously asked how much money OpenAI had spent on electricity just because people said “please” and “thank you” to ChatGPT.
OpenAI director Sam Altman responded with irony: "Tens of millions of dollars, well spent. You never know."
However, experts emphasize that there is no convincing evidence that politeness makes artificial intelligence more accurate.
Some early experiments have even shown the opposite: previous ChatGPT models have given more accurate answers even when users addressed them in offensive tones. But the researchers caution that these results are not consistent and the models are constantly changing.
What really works?
According to Professor Jules White from Vanderbilt University, the problem lies not in the choice of words, but in the way the request is formulated.
Experts recommend some simple practices:
* ask for several alternatives and not just one answer;
* provide concrete examples of the style or outcome you want;
* allow the chatbot to ask you questions one after another to gather information;
* avoid questions that pre-emptively lead it towards a specific answer.
Be careful with role-playing games.
It was once believed that if you told AI that it was a math professor, a doctor, or a lawyer, it would give more accurate answers.
Now researchers warn that this can have the opposite effect, especially when exact facts are required. The models can become more confident in themselves and produce incorrect answers with complete conviction – a phenomenon known as “hallucination”.
However, role-playing remains useful for creative tasks, such as simulating a job interview, brainstorming, or developing ideas.
So, should we be polite?
From a technical standpoint, the answer is not necessarily. Politeness does not make artificial intelligence smarter.
But psychologically, it can have value. Speaking politely makes many people feel more comfortable using technology and helps maintain communication habits.
As philosopher Immanuel Kant argued, the way we behave, even toward beings who have no feelings, affects our character.
Ultimately, artificial intelligence is neither offended nor happy. But perhaps it's worth continuing to say "please" and "thank you"—not for its sake, but for ours.
