MacroSnaps3 October 2026
No. 84

AI learns to write by reading the web. After English, the web's biggest languages are Russian, German and Japanese. Hindi, spoken by 611 million people, is a fifth of one percent of it.

Share of web pagesShare of the world who speak it
English
41.9%
18% of people
Russian
6.7%
3% of people
German
5.7%
2% of people
Japanese
5.2%
2% of people
Spanish
4.7%
7% of people
French
4.7%
4% of people
Chinese
4.2%
14% of people
Arabic
0.67%
4% of people
Hindi
0.21%
7% of people
Bengali
0.11%
3% of people

Orange: each language's share of 2.2 billion pages in Common Crawl, the free copy of the web most AI training starts from, September 2026. Grey: share of the world's 8.3 billion people who speak it.

Why this is happening
  • The rich world got online first. America, Europe, Russia and Japan have been writing web pages for 30 years. Most Hindi, Bengali and Swahili speakers came online in the last ten, mostly on phones.
  • Much of India's web is in English. Newspapers, government and business publish in it, so Indians who write online often do it in English, not Hindi.
  • Some text is hard to count. Lots of Hindi is typed in English letters, and Arabic speakers often write their local dialect, so the real figures are somewhat higher. The gap is still huge.
The take

German has a fifth as many speakers as Hindi but 27 times as many web pages: almost one page per German speaker, one per 131 Hindi speakers. AI models are weaker in languages they have read less of, and their makers' own tests show it. Hundreds of millions of the newest internet users will get a worse AI in their own language than in English, unless someone pays to close the gap.

Science & techmacrosnaps.app

Use this chart

Use this chart anywhere, including commercially, and change it if you need to. The one condition is that you credit MacroSnaps and link back to this page. Licence: Creative Commons Attribution 4.0 International (CC BY 4.0).

Embed it

A picture of the chart, linked back. Works anywhere, including in newsletters and Markdown. The credit line comes with it.

Cite it

Cites MacroSnaps for the chart. For the numbers themselves, cite the source below.

Permanent link: https://macrosnaps.app/c/ai-languages

The underlying data

MacroSnaps drew this chart. The numbers in it are published by Common Crawl, Ethnologue, UN (2026). Their terms, not ours, govern the underlying data.