Ask ChatGPT a complex question in English and you’ll quite often get a correct, well-formulated answer that fits the context. Try the same in Hindi, Bengali or Yoruba, and the response tends to get shorter, vaguer, and occasionally plain wrong. In German, French or Spanish, the answers are more on-point, but they often still fall short of the content and linguistic quality of an English response.<\/p>\n
Generative AI has a language problem. And it doesn\u2019t only affect rare or endangered languages – it hits widely spoken ones too. The Brookings Institution describes the quality gap as a continuum: from English through European languages like German, French and Spanish, all the way to the roughly 7,000 languages spoken worldwide, of which only about 20 are considered \u201cdata-rich\u201d – with the gap widening dramatically as you move down the list. This is a problem that surfaces repeatedly in GenAI projects. Non-English systems struggle with precision, hallucinate more frequently, and simply fabricate content that doesn\u2019t exist.<\/p>\n
If we have language technology that doesn’t work for people in the language that they speak, those communities don’t see the technology boost that other people might have.<\/p>Sanmi Koyejo, Stanford University, 2025<\/cite><\/blockquote>\n
The core problem: these systems were built in English<\/h2>\n
This isn\u2019t a bug that someone forgot to fix. It\u2019s a structural issue baked into the architecture of virtually every language model out there. Models learn from data – and the data is overwhelmingly English. Of the content in the Common Crawl dataset, the backbone of most large language model training, over 40% is in English, while no other language comes close to 7%. Models know what they\u2019ve seen, and most of it is in English.<\/p>\n
\n\n\n
\n \nLanguage<\/th>\n Common Crawl Share<\/th>\n Total Speakers<\/th>\n % of World Population<\/th>\n Web to Speakers Ratio<\/th>\n<\/tr>\n<\/thead>\n \n English<\/td>\n 41.06 %<\/td>\n ~1.53 billion<\/td>\n ~18.7 %<\/td>\n 2.2x<\/td>\n<\/tr>\n \n German<\/td>\n 5.98 %<\/td>\n ~135 million<\/td>\n ~1.6 %<\/td>\n 3.7x<\/td>\n<\/tr>\n \n Chinese<\/td>\n 4.99 %<\/td>\n ~1.18 billion<\/td>\n ~14.4 %<\/td>\n 0.35x<\/td>\n<\/tr>\n \n Spanish<\/td>\n 4.66 %<\/td>\n ~560 million<\/td>\n ~6.8 %<\/td>\n 0.7x<\/td>\n<\/tr>\n \n French<\/td>\n 4.61 %<\/td>\n ~310 million<\/td>\n ~3.8 %<\/td>\n 1.2x<\/td>\n<\/tr>\n \n Italian<\/td>\n 2.38 %<\/td>\n ~85 million<\/td>\n ~1.0 %<\/td>\n 2.4x<\/td>\n<\/tr>\n \n Hindi<\/td>\n 0.22 %<\/td>\n ~610 million<\/td>\n ~7.4 %<\/td>\n 0.03x<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n Sources: https:\/\/commoncrawl.github.io\/cc-crawl-statistics\/plots\/languages (accessed March 30, 2026). Ethnologue 2025 (Eberhard, Simons & Fennig, eds., Ethnologue: Languages of the World, 27th ed., SIL International) – for total speaker counts (L1+L2).<\/span><\/p>\n<\/figure>\n
The practical consequence: complex queries in non-English contexts get answered less precisely – especially in specialist domains like law or public administration. And even though German, French, Spanish and Italian are comparatively privileged as relatively \u201cdata-rich\u201d languages, they still share the same structural problems – just in a milder form. There\u2019s a secondary effect worth flagging too: content moderation instructions – for filtering hate speech or detecting statements that indicate serious mental health risk – are primarily designed and trained in English. In other languages, that precision degrades. Things get missed. Or get flagged when they shouldn\u2019t be.<\/p>\n