AI Recommends Different Brands by Language: Act Now

TL;DR: A controlled study of 1,650 AI responses across five languages found that Claude, ChatGPT, and Gemini recommend entirely different brands depending on which language a user types in — same model, same question, same day. For operators running multilingual acquisition funnels in high-CAC verticals, English-only AI visibility audits are dangerously incomplete. The brand that wins in English may not even appear in Arabic or Russian.

Same Question, Radically Different Winners

The test was straightforward: six buying scenarios, five languages (English, Arabic, Russian, Simplified Chinese, Persian), 40 runs each, using Claude Sonnet 5 with web search off. That produced 1,200 core responses, with another 450 from web-search-on and cross-model comparison tests for a total of 1,650 data points.

The airline scenario produced the starkest numbers. When asked to recommend a return flight from London to Tenerife, Claude picked British Airways 52.5% of the time in Persian — and exactly 0% of the time in Russian. In Russian, easyJet dominated at 72.5%. That is a 60-point swing on a single brand across two languages. Hotel recommendations showed the same pattern: Hotel Botánico captured 32.5% of English recommendations and zero in Russian. Bahía del Duque won 43.6% of Chinese recommendations but only 7.7% of Arabic ones.

This is not a translation issue. The AI does not translate its English reasoning into another language. It reasons from scratch, in that language, drawing on different training emphases, cultural weights, and implicit values. Anthropic’s own July 2026 research on Claude’s values across languages — sampling 310,000 real conversations in 20 languages — found the model leans toward warmth in Hindi and Arabic, and toward rigor in English and Russian. Different values produce different recommendations.

Why AI Reasons Differently Across Languages

The mechanism here matters for operators. When a user types a buying query in Simplified Chinese, the model is not running an English inference and rendering the output in Chinese. It is composing a response shaped by the linguistic and cultural associations embedded in that language’s training corpus. Regional norms around brand trust, risk tolerance, and product expectations all influence what the model presents as the “obvious” recommendation.

This extends beyond Western AI models. ChatGPT and Gemini are inaccessible in China. Claude has no official presence in mainland China, Russia, or Iran. In those markets, regional models like Baidu’s ERNIE, Naver (South Korea), and Mistral (France) carry significant referral weight, each trained on a national corpus with its own cultural identity. For operators with any exposure to those markets, brand visibility in English-language AI outputs is essentially irrelevant.

Even within the United States, the English-only assumption breaks. Los Angeles County alone has roughly 74,000 Persian speakers, and more than half of residents speak a language other than English at home. There is no evidence they switch to English when opening ChatGPT. If your brand appears consistently in English queries but drops out in Persian queries, you are losing a meaningful slice of a high-intent domestic audience.

What This Means for High-CAC Verticals

The implications are sharpest where customer acquisition costs are highest. Consider the verticals where a single converted lead is worth hundreds or thousands of dollars.

In iGaming acquisition, operators running campaigns across the Middle East and Southeast Asia already deal with multilingual creative — but almost none of them test whether their brand appears in Arabic or Malay AI responses. The assumption is that AI visibility is an English SEO problem. It is not. It is a language-by-language recommendation problem, and the audit methodology has to match.

For forex broker acquisition, the same gap is acute. High-intent retail traders in Kazakhstan, Saudi Arabia, and Singapore — markets where ChatGPT alone drives roughly three-quarters of AI chatbot referrals according to Statcounter’s August 2026 data — are querying in Russian, Arabic, and Simplified Chinese. If a broker has invested heavily in English AI visibility and done nothing in those languages, competitors who have not are still potentially winning those referrals by default.

Legal operators targeting Spanish-dominant markets in the U.S., or crypto exchanges with a significant Russian-speaking user base, face the same structural problem. The law firm marketing playbook has always accounted for Spanish-language outreach on paid channels. It now needs to account for Spanish-language AI referrals too. Same logic applies to crypto exchange acquisition in Eastern European and Central Asian markets.

The English-Only Audit Is the Wrong Baseline

Most brands doing any kind of AI visibility work are checking their presence in English. They are asking ChatGPT “what CRM should I use?” and noting whether their product gets named. That is a starting point, not a strategy.

A real multilingual AI visibility audit requires: selecting the languages that index against your actual acquisition markets, running structured query sets in each language, counting brand mention rates and recommendation rates separately, and comparing across languages to surface the gaps. The methodology used in this study — fixed scenarios, fixed run counts, text-matching pipelines for brand names including local-language brand forms — is replicable and should be part of any serious performance marketing audit for operators with international or multilingual audiences.

One complication: branded terms do not always appear in Latin script. In the airline test, roughly 5% of Chinese responses rendered British Airways as “英国航空” and easyJet as “易捷航空.” Some Arabic responses used phonetic transliterations. Any text-matching pipeline that only looks for Latin-script brand names will undercount — or miscount — visibility in non-Latin-script languages.

Web search toggle also matters. The study ran a subset of scenarios with web search enabled and compared those results against the closed-model outputs. Operators should understand that their AI visibility picture shifts depending on whether the model is pulling live web data or reasoning from training alone — and that both states occur in the real world depending on how users configure their AI tools.

Building a Multilingual AI Visibility Program

Operators who want to close this gap need to approach it as an ongoing measurement program, not a one-time check. Here is what that looks like in practice.

First, identify the languages that matter for your specific acquisition markets. This is not a generic list — it is derived from where your current paid traffic comes from, what languages your highest-LTV customers speak, and which AI models are dominant in those geographies. A trucking operator recruiting CDL drivers in South Texas has a Spanish-language AI visibility problem that is entirely different from an iGaming operator targeting Southeast Asia.

Second, build a query set that reflects real buying intent in each language. Do not translate English queries word-for-word. Native speakers phrase buying questions differently, and the AI’s reasoning is sensitive to those phrasings. Precision audience targeting at the paid level already accounts for cultural and linguistic nuance — your AI visibility program should too.

Third, run your queries at scale and at regular intervals. AI models update. Training data shifts. A brand that appears consistently in Russian responses today may drop out after the next model update. This is not a set-and-forget exercise.

Fourth, connect AI visibility findings to your content and authority-building strategy. The brands that appear most consistently in AI recommendations across languages tend to have authoritative, language-specific content that the model has encountered in training. English-only content libraries are a structural disadvantage in multilingual AI recommendation contexts. This is where paid performance programs and organic authority-building need to run in parallel — visibility in AI responses is partly a function of how much credible, language-appropriate content exists about your brand across the open web.

Finally, do not assume that model behavior is consistent across ChatGPT, Gemini, and Claude. The study found variation across models as well as across languages. Your brand’s AI visibility profile is a matrix: model by language by query type. Operators who map that matrix have a competitive advantage over those who are still spot-checking in English.

Originally reported by Search Engine Journal, October 2026.