Researchers find AI understands Polish and Russian better than English

Researchers from the University of Maryland and Microsoft have found that AI models respond more accurately and thoroughly to queries in languages other than English, particularly Polish and Russian. Contrary to expectations, English ranked only sixth in terms of effective interaction with AI systems.

The team tested several leading AI models, including OpenAI, Google Gemini, Qwen, Llama, and DeepSeek, across 26 languages. Each model received identical prompts, and the responses were evaluated for accuracy, completeness, and consistency, especially when processing long-form text.

The highest performance was observed in Polish, with an accuracy score of 88%. Russian followed closely at 87%, ahead of French (86%), Italian (85%), and Spanish (85%). English scored 84%, placing it sixth overall.

The researchers noted that the outcome was unexpected. While most modern AI systems are trained primarily on English-language data, the tests show that these models can analyze and structure queries more effectively in several European languages.

One possible explanation is the linguistic structure of these languages: features such as case systems, morphology, and clearer syntactic rules may reduce ambiguity, allowing the models to interpret meaning more precisely.

In other words, despite being considered the default language of artificial intelligence, English may not be the one AI understands best, with Polish and Russian now taking the lead.