Over half of active ChatGPT consumer users now primarily speak a language other than English. That is OpenAI's own Signals program data. The top three non-English languages are Spanish, Portuguese, and Arabic. The fastest-growing user languages inside ChatGPT are Uzbek, Kazakh, and Burmese. AI search visits in Europe are doubling year over year and projected to reach a quarter of all organic traffic by late 2026. And nearly every AEO playbook currently on the web is written in English, tested against English queries, and optimized for English citation graphs that look nothing like the graphs those non-English users are actually generating.
This is the guide for the other half. Eighth in the vertical AEO series, covering per-engine non-English behavior, regional AI ecosystems that Western engines do not reach, and the schema stack that keeps a brand resolvable as one entity across languages.
The Non-English Majority Nobody Optimizes For
English is 49.5% of websites whose content language is known, but only 26% of the global internet population primarily reads English. Chinese speakers make up 19.4% of internet users. Spanish 7.9%. Arabic 5.2%. Each of those is a large market with a distinctive AI citation graph and comparatively little competition inside it.
The gap is real and it is measurable. Temso.ai's March 2026 study of 3.25 billion citations across seven AI models and 14 countries produced a per-language citation-rate table that is worth internalizing:
| Language | Google AI Overview | Copilot | ChatGPT | Grok |
|---|---|---|---|---|
| Italian | 89.9% | 76.5% | 73.8% | 54.0% |
| German | 89.8% | 75.9% | 66.0% | 49.3% |
| Swedish | 85.1% | 60.4% | 57.4% | 47.1% |
| Spanish | 83.4% | 83.5% | 74.6% | 55.3% |
| French | 82.1% | 87.2% | 72.3% | 58.9% |
| Dutch | 81.2% | 66.4% | 62.2% | 38.3% |
The numbers are the percentage of citations that come from local-language sources when the query itself is in that language. Google AI Overview cites local sources 85% of the time on average. Grok cites local sources 52% of the time. ChatGPT sits in the middle, and drops significantly for Germanic languages (Dutch, Swedish) that have smaller native-language web corpora.
The Faux Polyglot Problem
The most important behavior to understand is the one researchers named "Faux Polyglot" in an arXiv paper cited by Advanced Web Ranking. ChatGPT often generates its response in the query language while grounding the answer in English-language sources. A Spanish query returns a Spanish paragraph built from English content. A Japanese query returns Japanese output with English sources behind it.
That is why the intuitive move (translate your English content into ten languages and add hreflang tags) does not work. The AI grounding layer operates upstream of the URL-serving logic where hreflang functions in classic search. Localized URLs are frequently omitted during generative AI retrieval even when the user query is in a specific language. The AI is not asking Google, "what's the German version of this page?" It is asking a training-derived corpus for "the answer to this question," and defaulting to the language its training corpus has the most of.
✈️ Where AI Search Sends Traffic: 10-Market Patterns for Your Global AI Search Strategy 👇 The outcome of my latest International AI search traffic analysis: There’s a comfortable narrative around AI search right now: AI assistants surface the biggest brands, ecommerce https://t.co/eww8H2MBUn
Aleyda Solis 🕊️@aleydaApr 29, 2026
Aleyda Solis's April 2026 international AI search traffic analysis across 10 markets landed on the same finding by different means. AI search does not automatically surface the biggest global brands. Traffic and citation patterns are visibly different per market, and the markets where brands invest in genuinely native content see disproportionate lift. The corollary is stark: brands that treat multilingual AEO as translation-plus-hreflang will lose to smaller local competitors that treat it as native content plus regional citation graphs.
How Each Western Engine Handles Non-English Queries
Per-engine defaults matter more than most global AEO teams realize:
- Gemini has the deepest multilingual training corpus of the Western engines, 70+ languages, and cites local-language sources 85.4% of the time. Gemini 2.5 Pro and Flash rank among the best for Chinese, Portuguese (Brazil), Ukrainian, and French translation.
- ChatGPT with Browse enabled defaults to English-language sources for ambiguous queries regardless of user locale. GPT-4o offers 32% better context understanding in non-English languages than GPT-3.5.
- Perplexity follows Bing index coverage in non-English. That single sentence is the whole strategy: Bing Webmaster Tools is as important as Google Search Console for non-English Perplexity visibility.
- Claude supports all major world languages with particularly strong performance in German, Japanese, Korean, Dutch, and Italian. Claude Sonnet 3.7 outperforms GPT-4 in Southeast Asian languages (81% Thai sentiment analysis accuracy versus GPT-4's 68%). Claude also maintains dialogue context across five-plus language switches per turn.
- Grok is the least localized at 51.7% average local-language citation rate.
Know exactly what AI says about your competitors.
RankControl's Recon Agent monitors competitor citations across ChatGPT, Perplexity, Claude, and Gemini. See where they show up and you don't.
The Citation Graph Changes Per Language Market
Query language reshapes the citation graph itself, well beyond changing which single source gets cited. Profound's analysis of Google AI Overviews behavior by market:
- Spanish (LATAM): TikTok jumps to 16% of AI citations in Spanish queries, a 5x increase over the English baseline. Instagram and TikTok collectively dominate social citations in Mexico and Spain.
- Portuguese (Brazil): YouTube concentrates further at 65% of citations. Reddit collapses to 7%. Instagram triples to 17%.
- Arabic: Instagram reaches 29% of citations, YouTube falls to 26% (nearly inverting the English baseline where YouTube dominates). LinkedIn roughly doubles.
For a global brand, that means the off-site presence stack must be rebuilt per market. TikTok is a citation channel in Spanish and Portuguese in a way it simply is not in English. Instagram carries an outsized share in Arabic. Reddit's outsized citation share in ChatGPT and Perplexity is an English-market phenomenon and does not carry into most European or Asian markets.
The Regional AI Ecosystems Western Engines Do Not Reach
Chinese, Russian, and Korean AI search operate on entirely different citation ecosystems.
Baidu ERNIE / Wenxin. Baidu merged the Ernie App and Wenxin Bot into the Baidu Wenxin Assistant on June 25, 2026, unifying its AI search entry point for hundreds of millions of Chinese users. The citation priority stack is Baidu Baike (their Wikipedia equivalent) first, then Zhihu high-vote answers, then B2B tech media like 36Kr, Huxiu, and Leiphone, then WeChat Public Account articles, then CNKI and WanFang for academic content. A brand's own website carries minimal weight. DeepSeek cites heavily from Zhihu long-form Q&A and technical documentation in Simplified Chinese. Qwen (Alibaba) surfaces results from Taobao, Tmall, and 1688 for commercial queries, which means e-commerce brands need aligned platform presence.
Yandex Alice and YandexGPT. Yandex is integrating generative AI across 100% of its consumer-facing products by end of 2026, and YandexGPT has deep understanding of Russian language, legislation, and post-Soviet cultural context. Citation sources are Russian Wikipedia, Yandex Q for Q&A, Kinopoisk for entertainment, and Russian media like RBC and Kommersant.
Naver HyperCLOVA X. HyperCLOVA X powers Naver's "Cue:" AI search. Primary citation sources are Naver Q&A, Naver Blog, and Naver News. A brand without a Naver Blog presence is nearly invisible to Korean AI search regardless of how strong its English content is. SK Telecom's KoGPT is a smaller Korean alternative for younger demographics.
Doubao (ByteDance). Doubao is connected to Douyin, Toutiao, and Xigua Video, and functions as a TikTok-adjacent citation ecosystem for brand awareness among mid-market and SMB audiences in China.
Wikipedia Is Language-Specific
Wikipedia accounts for an estimated 47.9% of ChatGPT's top-10 source share and 7.8% of total individual citations. It is the single most important AI citation source across engines. The critical fact for international AEO is that non-English Wikipedia has different content depth, different editorial gatekeepers, and different article eligibility requirements than English Wikipedia. A brand with a solid English Wikipedia article is not automatically cited in Japanese, German, or Arabic AI responses.
The AEO implication: to earn AI citation in a target language, the entity needs a Wikipedia article in that language with its own citations from local sources. A translation stub of the English article does not qualify. That is a substantial editorial project. It is also often the single highest-ROI move an international brand can make.
15+ content types. Published on your domain. Matched to your brand.
Guides, comparisons, listicles, case studies, and more. RankControl generates content that gets cited by ChatGPT, Claude, Perplexity, and more.
Hreflang, Schema, and Multilingual Entity Resolution
Hreflang tags do help with cross-lingual entity resolution, but they matter more for classic Google search than for AI grounding. Ship them anyway. Three delivery patterns:
<link rel="alternate" hreflang="de" href="https://example.com/de/page/">in the HTML head (most common)- HTTP headers for non-HTML content
<xhtml:link rel="alternate" hreflang="es" href="..."/>in the XML sitemap
Include an x-default tag or you break entity resolution when AI cannot match a locale. Ensure schema markup references locale-specific URLs, not English defaults.
The schema stack that actually moves multilingual AI citation:
inLanguageon every page using the ISO 639-1 code ("de","ja","ar","zh-Hans","zh-Hant"). Must match the page's actual language, not the site's default.sameAslinking every language version to the same Wikidata entity plus the language-specific Wikipedia article plus regional business directories. Wikidata items with multilingual labels in every operating language function as semantic anchors that prevent AI systems from treating regional pages as unrelated brands.areaServedto distinguish regional entity from global brandparentOrganizationto link regional entity back to global brandaudiencewithgeographicAreafor targeted local content
For the underlying AEO terminology that governs how each of these interacts with the retrieval layer, our terminology cheat sheet covers 65 terms in one-line definitions. Enterprises running this across 12 markets should also review the enterprise AEO playbook for the governance and procurement layer.
Translation vs Cultural Localization
Machine translation produces grammatically correct output. It fails at entity localization, cultural context adaptation, and market-specific terminology mapping. AI citation engines detect semantic distinctiveness between regional variants.
The classic reference numbers, cited across McKinsey localization research and multiple practitioner sources: keyword-translated content saw 45% lower engagement than culturally contextualized content. Airbnb documented a 20% year-over-year booking increase in new markets after fully localizing support content, moving from basic translation to local idioms, region-specific search intent, and regionally adapted keywords.
The practical distinction for AEO. Translation moves words from one language to another. Localization moves entities: the local competitors named, the local news sources cited, the local tax code referenced, the local city names spelled the way natives actually spell them, the writing systems (Japanese has three: kanji, hiragana, katakana) selected to match the native reading pattern for that content type. AI models score 64% accuracy on Arabic and Hindi, 79% on Japanese and Korean, versus 90%-plus on English, French, and Spanish. Content that reads as machine-translated to a native ear reads as machine-translated to a model trained on native content. Both fail the citation test.
The half of the ChatGPT user base that primarily speaks a language other than English is currently under-served by AEO tooling built for the English graph. The brands that move first inside a target language (native content, language-specific Wikipedia, regional Wikidata entity, off-site presence in the platforms that actually feed each ecosystem) are inheriting citation share while everyone else waits for the English playbook to translate itself. RankControl tracks citation share across ChatGPT, Perplexity, Claude, and Gemini per language and per market, so the Spanish citation drop or the Naver Blog absence surfaces the same week rather than the same fiscal year.
Your competitors are getting cited by AI. You're not.
Every day without citation tracking is a day your competitors pull ahead in ChatGPT, Perplexity, and Claude.
