Polish has unexpectedly emerged as the most effective language for giving instructions to artificial intelligence systems, a new study finds. Researchers from the University of Maryland and Microsoft examined how 26 different languages are understood by major AI models, including OpenAI, Google Gemini, Qwen, Llama and DeepSeek.
Their results show that Polish achieved an impressive 88% accuracy rate when completing given tasks. English, often considered the main language of technology, only reached sixth place. French, Italian and Spanish followed closely behind, while Chinese performed surprisingly poorly, ranking fourth from the bottom.
“Our experiment yielded some surprising results,” the researchers explained. “English did not perform best across all models. Polish proved to be the leading language.”
The study analysed how AI systems interpreted identical prompts in multiple languages, focusing on accuracy and comprehension. According to the Polish Patent Office, the findings overturn long-held beliefs about Polish being too complex for machines to process effectively.
“Humans may find Polish difficult, but artificial intelligence handles it with remarkable precision,” the institution commented on social media.
Experts note that this success is particularly striking because there is far less Polish-language data available for training AI than for English or Chinese.
The top ten most effective languages for AI prompting were Polish, French, Italian, Spanish, Russian, English, Ukrainian, Portuguese, German and Dutch — proving that linguistic diversity still matters in the digital world.
