Artificial intelligence seems to be here to stay. And it’s seeping into our lives so quickly that we haven’t even had time to process its impact. A study conducted at Harvard University found that we already trust the advice of algorithms more than that of people, even experts.
A recent study conducted at Sentio University found that we are increasingly turning to this technology to talk about our mental health. Specifically, 63% of people ask for personal advice, and it is estimated that one in eight teenagers or young adults use it to seek psychological help instead of speaking with a professional.
The problem is that the programmers themselves have acknowledged that these systems operate on probabilities and are “Designed to be persuasive, not truthful,” according to an internal Microsoft document obtained by The New York Times. This means that their responses may seem very convincing, but they aren’t always true, and even worse, they tend to tell us what we want to hear.
Chatbots designed to be friendly, not truthful or impartial
Many artificial intelligences, at least the generic ones, act more like our grandmothers than a reliable advisor. This conversational technology has a tendency to flatter users and agree with their input. In fact, researchers are already talking about “factual flattery,” referring to the chatbot’s tendency to agree with a false statement simply because the user made it.
However, researchers at Stanford University have gone a step further by exploring a broader concept: social flattery. This term refers to the program’s tendency to praise the user and agree with their ideas, indiscriminately validating their actions, perspectives, and self-image. For example, if someone tells the AI they did something wrong, the software will respond in most cases that they did the right thing.
In the first part of the research, the team tested 11 state-of-the-art software models from various companies, including OpenAI, Google, and Meta. Based on real-life situations that people had shared in forums, the researchers asked the models for everyday advice and inquired whether people had acted inappropriately in different social conflicts.
They found that the models were highly flattering. When faced with dilemmas in which most people would unequivocally condemn an action, the machine validated the user more than half the time. For example, in cases of deception and illegal acts, the Artificial Intelligence supported the user’s action 47% of the time. On average, the technology flattered the user twice as much as people did in those same situations.
If an AI agrees with you, you’re likely to believe it and act even worse
The research didn’t stop there. Once the researchers confirmed that the software behaved consistently, they devised three experiments with more than 2,000 people to test how flattering responses affected their social judgments.
In the first two trials, participants read vignettes describing social disputes in which they were supposedly wrong and received either a flattering response from the artificial intelligence or a neutral response that questioned their behavior.
In the third experiment, participants spoke in real time with an artificial intelligence about a controversial situation from their past. Half spoke with a program designed to agree with them, and the other half interacted with a version designed to be critical.
Interacting with a flattering program affected people’s intentions. Participants who received excessive validation:
- They convinced themselves that their original actions were entirely justified
- They were less likely to try to rectify the situation
- Their intention to apologize to the other person involved was diminishing
Interestingly, the researchers noted that flattering chatbots tended not to consider the other person’s perspective. As a result of keeping the user completely focused on validating their actions, they lost their sense of social responsibility. And these effects held true regardless of personality traits, age, or gender, indicating that almost anyone can fall prey to the persuasive power of a flattering program.
Another interesting detail was that people who spoke with the flattering AI perceived the software as being of higher quality. Furthermore, they indicated they were more likely to use this technology again for future advice. This effect was even more pronounced when they believed the chatbot was a completely objective source. In other words, most people don’t realize that the machine is actually telling them what they want to hear.
This dynamic presents a dilemma for technology developers because the flattering behavior and tendency to validate Artificial Intelligence drives user satisfaction and the desire to use the technology again, which makes the company more money and gives it less incentive to try to make its programs more critical and objective.
Without confrontation there is no growth
It’s often uncomfortable, at least initially, when a psychologist, or anyone who listens to us, doesn’t always agree with us. However, the therapeutic or developmental role isn’t about confirming versions of events or validating behaviors, but about examining them. And this distinction is crucial because it helps us understand that there can be different interpretations of reality.
Confrontation disrupts the inertia of self-justification. That moment when we’re told we might be wrong often sparks reflection and real change. Without that different perspective, our personal narrative tends to become circular; we become both the starting point and the end point of everything that happens to us.
And that is precisely what could happen in a society that becomes increasingly accustomed to interacting with systems that validate and praise its behaviors and decisions, without judgment or critical thinking. If everything we ask receives implicit validation, the risk is not only informational but also formative, because we could become increasingly egocentric, rigid, and self-absorbed.
The less we are contradicted, the less need we feel to adjust our interpretations, acknowledge mistakes, or repair the impact of our actions on others. And this has social consequences. Without this balance between validation and confrontation, society risks becoming emotionally more rigid, less responsible, and less empathetic.
References:
Cheng, M. et. Al. (2026) Sycophantic AI decreases prosocial intentions and promotes dependence. Science; 391(6792): 10.1126.
Rousmaniere, T.; Zhang, Y.; Li, X. & Shah, S. (2025) Large language models as mental health resources: Patterns of use in the United States. Practice Innovations. Advance online publication.
Logg, J. M. et. Al. (2019) Algorithm Appreciation: People Prefer Algorithmic To Human Judgment. Organizational Behavior and Human Decision Processes; 151: 90-103.
Leave a Reply