A research team explored which human languages artificial intelligence understands best, revealing surprising results that challenged long-held assumptions. The University of Maryland (UMD) and Microsoft jointly conducted a study analyzing how 26 different languages influence AI model accuracy. To the astonishment of many, Polish ranked first, while English placed only sixth.
Researchers evaluated major AI models, including OpenAI, Google Gemini, Qwen, Llama, and DeepSeek. Each model received identical prompts translated into 26 languages. The team then measured how precisely each model followed instructions and produced relevant outputs. Polish achieved an impressive 88% average accuracy, making it the most effective language for communicating with AI systems.
The report stated, “Our experiment yielded unexpected and counterintuitive findings. English did not perform best across all models; it ranked sixth when processing long texts, while Polish led convincingly.” The result surprised linguists and technologists who had long assumed that English, due to its dominance in digital content, would maintain a clear advantage in AI interpretation.
Why Polish Outperformed English
The findings indicate that Polish offers exceptional precision in AI prompting. According to the Polish Patent Office, which shared the results online, “Polish appears to give the most exact instructions to artificial intelligence. For years, people have called Polish one of the hardest languages to master. It turns out that humans struggle with it—but AI does not.”
Researchers suggest that Polish’s complex grammatical structure may actually help AI models process meaning more effectively. Its rich system of inflections, cases, and verb conjugations allows commands to be expressed with greater clarity and less ambiguity. Unlike English, which often relies on word order, Polish enables precise intent through morphological cues.
Interestingly, AI systems showed strong proficiency in Polish despite the limited amount of Polish-language data available for training. Compared with English or Chinese, the dataset size for Polish is significantly smaller, yet the models interpreted and executed Polish commands more accurately. This outcome suggests that language structure—not just data quantity—plays a vital role in AI comprehension.
Global Ranking Highlights Linguistic Diversity
The study also measured performance across other major languages. French, Italian, Spanish, and Russian followed closely behind Polish, all exceeding 84% accuracy. English reached 83.9%, narrowly surpassing Ukrainian and Portuguese. Meanwhile, Chinese, despite being one of the most widely spoken languages, ranked near the bottom—fourth from last—demonstrating significant limitations in AI understanding.
The top ten languages for effective AI communication were as follows:
- Polish – 88%
- French – 87%
- Italian – 86%
- Spanish – 85%
- Russian – 84%
- English – 83.9%
- Ukrainian – 83.5%
- Portuguese – 82%
- German – 81%
- Dutch – 80%
Experts believe these results could influence how developers design and train future AI systems. By studying why certain grammatical or linguistic features improve comprehension, researchers can refine models to perform better across languages, not just in English-dominated contexts.
In the broader sense, the findings also highlight how artificial intelligence may redefine global linguistic hierarchies. While English has long served as the default language of technology, the study shows that languages like Polish may offer structural advantages that allow machines to understand human meaning with greater accuracy and nuance.
