AI Thinks All Cultures Are Basically the Same — New Study Says It's Wrong

A research team led by Kimmo Eriksson tested four of the most popular AI chatbots against a huge real-world survey. That survey, the Global Study of Everyday Norms, asked people in 90 societies to rate 150 everyday behaviors — things like kissing in public or flirting. The researchers then asked GPT-5, GPT-5.4, Claude Opus 4.6 and Gemini 3.1 Pro to guess each society's average answer, and compared the two.

The AI got the big picture wrong. Every model made the world look more alike than it is, guessing that differences between countries were less than half their real size. Worse, for many behaviors the models named the wrong countries as strict or relaxed. There was one bright spot: anything with a whiff of vulgarity, especially kissing and flirting, was judged more accurately — probably because those topics are heavily discussed online.

The researchers also tested fixes. Norms in wealthier, more developed countries were guessed a bit more accurately, likely because more written material exists about those places. Asking the AI in local languages instead of English helped only modestly — it shrank the error but didn't erase it.

So what does this mean for you? If you ask a chatbot whether a joke is fine in a business meeting in Japan, Brazil or Saudi Arabia, expect a flattened, overly general answer that may be confidently wrong. That matters for anyone doing international work, marketing, hiring or moderating online comments at scale. The takeaway isn't that AI is useless for etiquette questions — it's that it should be a starting point, never the final word, on how other cultures actually behave.