10% of Americans now turn to chatbots for emotional support

The numbers in this article point to a pattern that predates AI: when people face questions beyond their own judgment, they consult sources they believe know more. A 2026 Pew report found 10% of Americans now use chatbots for emotional support or companionship, and a study of 1.1 million ChatGPT conversations found 49% of messages involved seeking information or advice.

The article's authors — an anthropologist and an AI researcher at the University of Michigan — connect this to the history of oracles. Ask Delphi, an AI ethics website, drew 3 million queries in a few weeks in 2021; its designers at the Allen Institute for AI and the University of Washington worried users might treat it as a moral authority. The Verge argued the name itself, borrowed from the Delphic oracle, invited that reading. The article notes that AI systems remain opaque even to their designers — a whole field, mechanistic interpretability, exists to understand their internals. Anthropic's "spiritual bliss" attractor, where Claude 4 drifted into mystical themes without training, is cited as evidence that unexpected behavior can emerge from systems no one fully understands.

What hinges on this is whether designers can close that gap before users decide the opacity itself is a mark of authority. The article's authors stress they are not arguing AI is divine or infallible — only that the human habit of looking beyond ourselves when we reach our limits is showing up again, now pointed at chatbots.