Multilingual GEO
Multilingual GEO is generative engine optimisation for queries asked in languages other than English. Google's AI Overviews reached more than 200 countries and territories and more than 40 languages by May 2025[1], and AI Mode reached nearly 100 languages in February 2026[2]. Coverage arrived country by country and language by language, so a market being live and its language being supported are two separate facts.
Rollout by country and by language
AI Overviews launched in the United States in May 2024 and reached more than 100 countries and territories by October that year, in six languages[3]. March 2025 added German, Italian, Polish and French alongside a set of European markets[4]. By May 2025 Google put the total at more than 200 countries and territories and more than 40 languages, taking in Arabic, Chinese, Malay and Urdu[1].
AI Mode rollout
AI Mode ran its own schedule: the United States in March 2025, India in June, the United Kingdom in July[5], five more languages in September[6], then more than 35 new languages and over 40 new countries on 7 October 2025, which carried it past 200 territories[7]. February 2026 added 53 further languages for a total of nearly 100[2]. That batch was weighted towards lower-resource languages, among them Amharic, Hausa, Somali, Yoruba, Zulu, Kazakh, Nepali, Khmer, Faroese and Western Frisian[8].
A country launch is not a language launch
At the March 2025 European launch Germany received AI Overviews in German and English while Belgium received English only[9], so users in some live markets met an English answer to a query typed in their own language. The obvious page-level signal does not close that gap. Google's documentation states that neither hreflang nor the HTML lang attribute is used to detect the language of a page, which separate algorithms determine; the tag only tells Google which localised variant to show a given user[10].
Retrieval quality outside English
Coverage in a product is not quality in the language. MIRACL, a standard multilingual retrieval benchmark, gathered more than 700,000 relevance judgments for around 77,000 queries over Wikipedia in 18 languages, every assessment made by a native speaker[11]. Specialist domains fall much further. A 2026 medical retrieval benchmark measured biomedical encoders at 0.818 nDCG@10 in English and 0.056 in Japanese on the same task, and states that an English-only benchmark cannot detect a gap of that size[12].
Measurement outside English
Frequency data outside English is thin. Ahrefs puts AI Overview appearance at roughly 12.5% to 16.5% of searches in countries where the feature has rolled out[13]. That single range covers every rolled-out market at once and is not broken down by language. Because a country can be live while its own language is not[9], per-country and per-language figures are not interchangeable, and the same distinction applies to local search outside English-speaking markets.
References (13)
- AI Overviews are now available in over 200 countries and territories, and more than 40 languages Archive
- Google AI Mode Expands to 53 Languages: Complete Analysis Archive
- Google AI Overviews: All You Need to Know Archive
- Google AI Overviews: All You Need to Know Archive
- Google expands AI Mode to over 40 countries and territories Archive
- Google expands AI Mode to over 40 countries and territories Archive
- Google expands AI Mode to over 40 countries and territories Archive
- Google AI Mode Expands to 53 Languages: Complete Analysis Archive
- Google AI Overviews: What's Changing for SEO & SEA in 2025 Archive
- Tell Google about localized versions of your page Archive
- Making a MIRACL: Multilingual Information Retrieval Across a Continuum of Languages
- MMed-Bench-IR: A Heterogeneous Benchmark for Multilingual Medical Information Retrieval
- Google AI Overviews: All You Need to Know Archive
Last updated 2026-09-04. Written and maintained by Baseline Labs.