March 26, 2026
AI chatbots are suck-ups, and that may be affecting your relationships
A new study of AI sycophancy shows how asking agreeable chatbots for advice can change your behavior

By Allison Parshall edited by Tanya Lewis

Large language model (LLM) chatbots have a tendency toward flattery. If you ask a model for advice, it is 49 percent more likely than a human, on average, to affirm your existing point of view rather than challenge it, a new study shows. The researchers demonstrated that receiving interpersonal advice from a sycophantic artificial intelligence chatbot can make people less likely to apologize and more convinced that they’re right.

People like what such chatbots have to say. Participants in the new study, which was published today in Science, preferred the sycophantic AI models to other models that gave it to them straight, even when the flatterers gave participants bad advice.

“The more you work with the LLM, the more you see these subtle sycophantic comments come up. And it makes us feel good,” says Anat Perry, a social psychologist at the Hebrew University of Jerusalem, who was not involved in the new study but authored an accompanying commentary article. What’s scary, she says, “is that we’re not really aware of these dangers.”

-snip-

The new study examined only brief interactions with chatbots. Dana Calacci, who studies the social impact of AI at Pennsylvania State University and wasn’t involved in the new research, has found that sycophancy tends to get worse the longer users interact with the model. “I think about this [as] compounded over time,” she says.

-snip-