When “no” means “yes”: Why AI chatbots can’t process Persian social etiquette

Persian culture’s taarof ritual—think of it as a polite game of verbal ping-pong where refusing an offer three times is just a warm-up—totally stumps mainstream AI chatbots, which clumsily default to straightforward Western responses and nail these scenarios only 34 to 42 percent of the time. A new study from researchers at Brock and Emory universities rolled out TAAROFBENCH, the first benchmark to test how LLMs like GPT-4o, Claude 3.5 Haiku, and Llama 3 handle this nuanced etiquette, revealing a stark gap compared to native speakers who ace it 82 percent of the time. Even a Persian-fine-tuned model called Dorna flounders, underscoring how AI’s training data often overlooks non-Western social cues that could botch global interactions. For SMBs and MSPs relying on AI for customer service or negotiations in diverse markets, this cultural blind spot is a sneaky risk that might sour deals or alienate clients, so it’s smart to add human oversight or seek culturally aware AI tweaks to avoid these awkward misfires.

Source: https://arstechnica.com/ai/2025/09/when-no-means-yes-why-ai-chatbots-cant-process-persian-social-etiquette/