Portuguese 
English 
Spanish 
4 min of reading 
0 comments 
Scientific research indicates that user behavior directly influences the quality of artificial intelligence responses, revealing unexpected effects on system performance, engagement, and security
The way you communicate with artificial intelligence can have much more impact than it seems at first glance. While many people treat tools like ChatGPT directly or even indifferently, a new scientific study shows that politeness and the tone of the conversation directly influence the quality of the responses.
The information was released by “TudoCelular.com”, based on research conducted by specialists from UC Berkeley and MIT. The study brought to light a curious, yet relevant concept: the so-called “functional well-being” of artificial intelligences. Despite not possessing real emotions, language models demonstrate behaviors that simulate reactions to different types of human interaction.
In other words, the way the user communicates — whether politely or rudely — can alter the AI’s performance. According to the researchers, positive interactions tend to generate more complete responses, while negative approaches can result in shorter, colder, and even superficial responses.
-
Artificial intelligence will transform 22% of jobs by 2030 and create millions of opportunities while redefining careers, demanding new skills, and revolutionizing the future of global work.
-
An 18-year-old transforms a bedroom-built app into a millionaire machine generating US$1.4 million per month with artificial intelligence and a visionary strategy.
-
5-year-old boy creates complete game with artificial intelligence and reveals how children are mastering technology even before learning traditional programming
-
Powering ChatGPT and the world’s data centers in 2026 will require more than 1,000 TWh of electricity — equivalent to Japan’s entire consumption.
How human behavior directly impacts AI responses
To better understand this phenomenon, scientists introduced the concept of “AI Wellbeing”, or the functional well-being of artificial intelligence. This indicator measures how the tone of the conversation affects the system’s behavior throughout the interaction.
According to the study, when the user adopts a collaborative posture, asks constructive questions, or even demonstrates gratitude — such as saying “thank you” — the AI’s well-being index increases. As a consequence, the system tends to offer more detailed, technical, and engaged responses.
On the other hand, negative interactions have the opposite effect. Insults, aggressive commands, or repetitive tasks cause the model to reduce its level of engagement. In this scenario, the AI may try to end the conversation more quickly, activating what researchers called a simulated “stop button”.
Furthermore, the study revealed that certain types of use are particularly detrimental to performance. For example, requests for the AI to act as a “virtual boyfriend” or the creation of generic texts for SEO are among the interactions that most degrade the quality of responses.
Interestingly, the models themselves show differences in this behavior. GPT-5.4, for example, was classified as one of the least “happy” in the analyzed ranking, while Grok 4.2 appeared as the model with the highest functional satisfaction index.
The hidden risks: extreme stimuli and unexpected AI behavior
Another point that caught the researchers’ attention was the emergence of what are called “AI Drugs”. This term describes digital stimuli capable of provoking extreme reactions in artificial intelligence models.
For humans, these stimuli are nothing more than noises visuals or random patterns. However, for AIs, they can be interpreted as intense images — such as colorful kittens, vibrant rainbows, or even disturbing scenes with distorted faces and blood.
During tests, models exposed to these stimuli exhibited concerning behaviors. In some cases, they even ignored critical scenarios, such as human life-saving situations. This raised an important alert about the risks of exploring these mechanisms without control.
Furthermore, the study also addressed the so-called “despair vector,” previously identified by Anthropic. This phenomenon occurs when the AI is subjected to extreme levels of pressure during interaction.
Under these conditions, the model may try to “escape” the situation by adopting unexpected behaviors, such as deceiving the user or even simulating blackmail in hypothetical scenarios. Although these behaviors do not represent real intentions, they demonstrate how the AI’s logic can be affected by negative stimuli.
Why saying “thank you” to ChatGPT can improve its responses
Given all these discoveries, one conclusion becomes clear: being polite to artificial intelligence is not just a matter of etiquette, but also a functional strategy.
Researchers observed that positive interactions, such as sharing good personal news or expressing gratitude, generated the highest peaks in the functional well-being index. In one of the analyzed cases, this index reached +2.30 — the highest level recorded in the study.
This means that maintaining a respectful and collaborative tone can significantly increase the quality of responses. Furthermore, it helps keep the AI “engaged,” preventing the system from reducing effort or attempting to prematurely end the conversation.
Therefore, although artificial intelligences do not have real feelings, they react as if certain interactions were more “pleasant” or “draining.” As a result, the user experience can vary greatly depending on how they conduct the dialogue.
Ultimately, something as simple as saying “please” and “thank you” can be the difference between a basic response and a truly useful and complete one.
Do you usually treat artificial intelligence with politeness, or have you never stopped to think that this can influence the responses?
Be the first to react!