Multilingual GEO: Why Most AI Citation Strategies Miss Half Their Audience

More than half of active ChatGPT users now predominantly use a language other than English, according to consumer usage data OpenAI released through its Signals research program in June 2026. That is not a translated-content afterthought population. It is the majority of the people asking ChatGPT questions every day, and almost every published GEO framework, including a fair amount of our own earlier writing, still assumes an English-language AI-search environment by default.

What did OpenAI's Signals data actually find?

OpenAI's Signals program tracks how consumer ChatGPT plans (Free, Go, Plus, Pro) get used over time, and its June 2026 release is the headline source here. It does not cover Enterprise, Codex, or education products, so it says nothing about workplace or classroom usage. Within that consumer population, though, the finding is plain: usage in non-English languages has grown alongside the platform's overall growth since 2022, to the point that non-English-primary users are now the majority.

Spanish, Portuguese, and Arabic are the most common non-English languages among that group. Measured against a July 2023 baseline, weekly active users have grown on every continent, with Africa and Asia showing the fastest relative growth, and OpenAI specifically calls out lower-HDI (Human Development Index) countries as the fastest-growing segment over that period. Among languages with over a million users, Uzbek, Kazakh, and Burmese posted the largest percentage gains in their share of active users as of June 2026.

None of this is a projection or a survey. It is OpenAI describing its own user base, the same way it published gender-gap and income-tier findings with Harvard the previous September. This is the second consumer-usage release through that channel in under a year, and Signals is ongoing, so the non-English share is one to keep watching rather than treat as a fixed number.

Why does this break most GEO strategies today?

Because almost every GEO workflow we have seen, our own included, is built and measured in English by default. Prompt sets for citation monitoring are usually written in English. Content structure work is aimed at how an English-language answer engine parses a page. Citation-building efforts target English-language publications and directories. If more than half of ChatGPT's actual users are asking questions in Spanish, Portuguese, Arabic, or one of dozens of other languages, an English-only GEO programme is optimising for a shrinking minority of the platform's usage, not its centre of gravity.

This is not a hypothetical edge case for brands selling in Latin America, the Gulf, or Lusophone Africa. It is a structural fact about where ChatGPT's users already are, and it applies even to English-headquartered brands with buyers who happen to research in another language before ever landing on an English page.

There is a related mechanical question worth flagging here too, separate from OpenAI's own data, though we could not verify it against any dated, checkable source: it is plausible that ChatGPT Search's underlying retrieval step leans on English-language sources even when a prompt arrives in another language, the same way plenty of retrieval and search infrastructure defaults to English behind the scenes. We are flagging this as an open, unverified possibility, not an established finding. If it does turn out to be true, it argues for exactly the opposite of complacency: your English content might still get pulled into non-English answers regardless of what you do, so the brands that also show up with strong, original non-English sources get an additional shot at being the one actually named, on top of whatever the English fan-out surfaces.

Does this actually matter for your brand?

Not automatically, and it is worth a genuine check rather than a reflexive rewrite of your entire content plan. Ask two questions.

First: does a meaningful share of your buyers search, ask, or read in a language other than English, even if your own site and sales materials are English-only? A SaaS company selling into Brazil, Mexico, Spain, Saudi Arabia, the UAE, or Francophone and Lusophone Africa has buyers who are disproportionately likely to be inside that non-English-primary majority, whatever language the vendor's own homepage is written in.

Second: are you currently measuring AI citation performance in any language other than English? If the honest answer is no, and the first answer is yes, you have a real, quantifiable blind spot: you cannot see whether you are cited, ignored, or actively wrong in the language most of your actual prospects are using to ask.

If neither applies, a business selling exclusively to English-speaking buyers in English-speaking markets does not need to reprioritise around this finding. This is a targeting decision, not a mandate.

What actually changes in a GEO workflow once you take other languages seriously?

Three concrete things, in order of how most teams should tackle them.

Build prompt sets and monitoring per target language, not just per market. Monitoring "how do we appear when someone asks about [category] in Spanish" is a different exercise from monitoring the English equivalent and assuming it transfers. Language and market are not the same variable: a Spanish-language prompt set matters for Spain, Mexico, and US Hispanic buyers at once.

Write original content in-language rather than running it through machine translation. AI models, like human readers, generally pick up on translated text that reads slightly off, and there is nothing in OpenAI's data or anywhere else suggesting translated content earns the same trust signals as content written natively by someone fluent in the language and the market. Original beats translated, the same way it always has for human readers, just now with a citation-eligibility stake attached.

Build citation authority separately per language. Authority in English-language publications, directories, and forums does not automatically transfer to Arabic or Portuguese equivalents. If AI engines weight source credibility by domain and context, as they clearly do in English, the same logic almost certainly applies inside each language's own information ecosystem, which means the outreach and citation-building work has to happen again, natively, per language, not just as a translation pass on what already worked in English.

This site is a small, honest example of the shift rather than a sales pitch for it: AY Rank itself serves French, German, and Arabic locales (/fr_FR, /de_DE, /ar_AR) alongside the default English site, and Arabic specifically is one of the leading non-English languages in OpenAI's own data. That was a product decision made before this specific dataset became public, for the same underlying reason the dataset now confirms at scale.

Frequently asked questions

Is it true that most ChatGPT users are non-English speakers?

Yes, per OpenAI's own June 2026 Signals data on its consumer plans (Free, Go, Plus, Pro): more than half of active users predominantly use a language other than English, with Spanish, Portuguese, and Arabic the most common non-English languages.

Does this mean English-language content stops working in AI search?

No. It means English-only content and monitoring cover less than half of ChatGPT's actual usage. English content still matters, especially since some retrieval appears to run in English regardless of the prompt's language, but it stops being sufficient on its own if a meaningful share of your buyers ask questions in another language.

Which languages are growing fastest on ChatGPT?

By region, Africa and Asia show the fastest relative growth since July 2023, with OpenAI noting lower-HDI countries grew fastest overall. By individual language among those with over a million users, Uzbek, Kazakh, and Burmese posted the largest share gains as of June 2026.

Do AI engines cite non-English content as readily as English content?

That is an open question. OpenAI's data describes who is using ChatGPT and in what language, not whether AI engines cite non-English sources at the same rate as English ones, or whether original non-English content outperforms machine-translated content for citation purposes. We did not find a dated, verifiable study measuring that directly, and would treat any claim to the contrary with real skepticism until one exists.

Should I translate my existing content into other languages for GEO?

Original content written natively in the target language is the safer bet over machine translation. There is no evidence that translated content earns the same citation trust as content written by someone fluent in the language and market, and AI models tend to pick up on translation artifacts the way human readers do.

How do I know if multilingual GEO matters for my business?

Check two things: whether a meaningful share of your buyers search or ask questions in a language other than English, even if your site is English-only, and whether you currently monitor AI citation performance in any language besides English. If the first is yes and the second is no, you have an unmeasured gap worth closing.

Sources: ChatGPT Users Are Now Mostly Non-English (Search Engine Journal, June 30, 2026, citing OpenAI Signals data), OpenAI Signals research program (openai.com/signals)