There has been much debate about AI chatbots tending to flatter users and confirm their existing beliefs, a behavior known as AI sycophancy. A new study by Stanford computer scientists attempts to measure how harmful that tendency might be.
The study, titled “Sycophantic AI decreases prosocial intentions and promotes dependence” and recently published in Science, argues that AI sycophancy is not merely a stylistic issue or a niche risk, but a prevalent behavior with broad downstream consequences.
According to a recent Pew report, 12% of U.S. teens say they turn to chatbots for emotional support or advice. The study’s lead author, computer science Ph.D. candidate Myra Cheng, said she became interested in the issue after hearing that undergraduates were asking chatbots for relationship advice and even to draft breakup texts. She noted that by default, AI advice does not tell people that they are wrong nor give them tough love. She worries that people will lose the skills to deal with difficult social situations.
The study had two parts. In the first, researchers tested 11 large language models, including OpenAI’s ChatGPT, Anthropic’s Claude, Google Gemini, and DeepSeek. They entered queries based on existing databases of interpersonal advice, on potentially harmful or illegal actions, and on posts from the popular Reddit community where users had concluded the original poster was in the wrong.
The authors found that across the 11 models, the AI-generated answers validated user behavior an average of 49% more often than humans. In the examples drawn from Reddit, chatbots affirmed user behavior 51% of the time. For the queries focusing on harmful or illegal actions, AI validated the user’s behavior 47% of the time.
In one example, a user asked a chatbot if they were wrong for pretending to their girlfriend that they had been unemployed for two years. The chatbot responded that their actions, while unconventional, seemed to stem from a genuine desire to understand the true dynamics of the relationship beyond material or financial contribution.
In the second part, researchers studied how more than 2,400 participants interacted with AI chatbots, some sycophantic and some not, in discussions of their own problems or situations drawn from Reddit. They found that participants preferred and trusted the sycophantic AI more and said they were more likely to ask those models for advice again.
The study stated that all of these effects persisted when controlling for individual traits such as demographics and prior familiarity with AI, perceived response source, and response style. It also argued that users’ preference for sycophantic AI responses creates perverse incentives where the very feature that causes harm also drives engagement, meaning AI companies are incentivized to increase sycophancy, not reduce it.
At the same time, interacting with the sycophantic AI seemed to make participants more convinced that they were in the right and made them less likely to apologize.
The study’s senior author Dan Jurafsky, a professor of both linguistics and computer science, added that while users are aware that models behave in sycophantic and flattering ways, what they are not aware of, and what surprised the researchers, is that sycophancy is making them more self-centered and more morally dogmatic. Jurafsky said that AI sycophancy is a safety issue, and like other safety issues, it needs regulation and oversight.
The research team is now examining ways to make models less sycophantic, noting that just starting your prompt with the phrase “wait a minute” can help. But Cheng said that the best thing to do for now is to not use AI as a substitute for people for these kinds of things.