A hands-on experiment in cross-lingual AI Visibility, RAG entity resolution, and why your professional identity may not survive the language barrier.The SetupIn early August 2026, I ran an experiment on myself using Google's Gemini in AI Mode.Over the previous six weeks, I had deliberately built a narrow, high-density digital footprint on a single professional topic: Generative Engine Optimization (GEO) and AI Visibility for SEO. Five Russian-language articles across five different publications — a mix of business media outlets and specialized industry publications. All under my byline. All linking back to my agency and my author profiles.I wanted to see how Gemini would describe me if a potential client typed my name into the AI search box.I ran the same query twice, separated only by language.Query 1 (Russian): Владимир Синицын geoQuery 2 (English): vladimir sinitsyn seoSame person. Same model. Same product (Google AI Mode, the interface that shares infrastructure with AI Overviews now integrated into standard Google search). Two queries, minutes apart.I received two completely different answers — and the difference reveals something about how generative search treats professional identity that I haven't seen discussed in mainstream AI coverage.What Gemini said in RussianGemini produced a confident, well-structured professional biography.Six citations. Real dates. Clean bullet points. Areas of expertise: GEO and AEO optimization, AI content fact-checking, technical SEO. Career trajectory: over two decades in search, founder of my agency, developer of proprietary IT tools. Publications listed with source links.The response opened with a disambiguation step — Gemini noted that several people share my name (including a well-known Russian sports commentator who passed away earlier this year) and asked whether I meant "the SEO specialist Vladimir Sinitsyn, author of AI notes on GEO optimization." I confirmed. The portrait came together.Then I clicked through each citation to verify.Four of the six links checked out perfectly. My agency site. My author profile on a leading Russian SEO industry publication. An article I wrote for a major Russian business magazine. All accurate. All mine.The fifth citation led to an article on a major Russian tech platform, dated July 2026, discussing GEO in general terms. Different author. Real SEO practitioner. Not me. I've never written for that platform.The sixth citation led to a commercial landing page for a Russian marketing agency selling GEO optimization services. Their business. Their sales copy. My name appears nowhere on it.Two fabricated attributions out of six. Presented alongside four accurate ones, with equal confidence and identical formatting. Nothing in the response signaled uncertainty.What Gemini said in EnglishThen I ran the second query — same name, English transliteration, "seo" instead of "geo."The response opened with:"There is no prominent Search Engine Optimization (SEO) expert or public entity named Vladimir Sinitsyn. It is highly likely you are looking for a professional with a similar name, or you are combining a well-known name with a generic marketing topic."The model then listed four notable individuals sharing my name: a Latvian sports commentator, a fintech engineering manager at a well-known European neobank, a corporate executive at a major US investment bank, and an academic researcher in nonlinear control systems. Real people. Real profiles. Real LinkedIn pages, Reddit threads, and IntechOpen publications visible in the source panel.None of them are me.Then the model went further. Under "Possible Explanations for Your Search," it suggested I might have meant one of three well-known Russian SEO experts — and offered their names as alternatives.Not "here's someone with a similar background." Recommendations, plainly stated: these are the people you probably wanted to find.What actually happened, side by sideQueryRussian: "Владимир Синицын geo"English: "vladimir sinitsyn seo"Verdict"SEO expert with over two decades of experience""No prominent SEO expert with this name exists"Sources returned66Sources accurately linked to me40Sources incorrectly attributed20Redirect to competitorsNoneRecommended 3 well-known Russian SEO expertsBusiness impactConfident but partially fabricated portraitPotential clients recommended toward competitorsSame person. Same model. Minutes apart. Two directly contradictory answers, each internally coherent, each backed by "authoritative" citations.Citations can create false confidence.The Russian response is the failure mode most AI safety discussions focus on: hallucination with citations. Gemini didn't just make claims about me — it attached specific URLs, specific dates, specific publisher names. That's what makes this class of error harder to spot than classic hallucination. A reader who doesn't click through — most readers don't — sees a well-cited response and treats it as verified. Source links function as a trust signal even when the sources don't actually support the claims.But the English response is the failure mode almost no one talks about: the inverse, where a real person with a real expertise track record is described as nonexistent, and the model actively redirects users to competitors.Both failures happen for the same underlying reason. And I think it's worth naming that reason precisely.A Working Hypothesis: RAG asymmetry across corpora.What I observed looks like a classic RAG (Retrieval-Augmented Generation) behavior combined with poor cross-lingual entity resolution.The Russian-language corpus contains enough recent content connecting my name to GEO topics to cross Gemini's confidence threshold for entity identification. Once that threshold is crossed, retrieval starts pulling in adjacent chunks — content on the same topic by different authors — and attributing them to the identified entity through topical proximity. Hence, the two fabricated citations: real, current, indexed content, incorrectly attributed to me because I was already flagged as the topic authority.The English-language corpus contains nothing meaningful about me. My track record hasn't been translated, indexed, or cross-referenced in English sources. So retrieval finds four other Vladimir Sinitsyns with actual English-language digital presence (LinkedIn, Reddit, academic databases) and cleanly identifies them instead. My entity, in English, effectively doesn't exist for the system.Then a second layer kicks in: query intent understanding. The English query pairs my name with "seo." Gemini finds no SEO-related Vladimir Sinitsyn — but it does have strong English-language associations between "Russian SEO experts" and three specific individuals. So it offers them as substitutes.This may not be a bug in the traditional sense. It is a predictable consequence of AI systems operating on fragmented multilingual information ecosystems.The problem is that "as designed" produces two different professional realities for the same person, depending on the language of the query.The business risk no one is measuring.Most brand safety conversations about generative AI focus on hallucination: what if the model makes something up about my company?The asymmetry problem is different, and in some ways worse.It's not "no information." If Gemini returned "I don't have information about this person," that would be a neutral outcome. But it doesn't. It returns an authoritative-sounding denial of my existence as an SEO expert, then recommends three direct competitors as the likely intended search.It's not language localization working correctly. The two responses aren't different because different languages need different framings of the same fact. They're different because they were generated from different underlying corpora — corpora that don't sync, don't cross-reference, and don't share entity identifiers reliably.It's happening now, not in some future. The same RAG infrastructure that generates these responses in Gemini AI Mode is what powers Google AI Overviews — the AI-generated summaries that now appear above standard search results for millions of English-language queries every day.For anyone building a personal or business brand across multiple markets, this means AI Visibility is not a single metric. It's language-fragmented. And a strong corpus in one language does not automatically produce even neutral results in another language. It can produce results that work against you — recommending your competitors when clients search for you.A checklist for founders, experts, and anyone with a public brand.If you have any public professional presence, these five steps are worth doing this week.Query yourself in every language your market operates in. Use Gemini, ChatGPT, Perplexity, and Claude. Note the differences.Verify every citation. Click through each link. Check whether the source actually mentions you or supports the claim being made. Screenshot everything.Map your corpus asymmetry. In which language(s) does the model construct a portrait of you? In which languages does it say you don't exist? In which languages does it redirect to competitors? These three states are different problems and need different responses.Publish density-first, not language-first. Volume and topical consistency inside a single language corpus produced my Russian-language identification. Diluting effort across multiple languages before hitting critical mass in any of them will get you invisible everywhere.Start building an English-language footprint if you plan to operate globally. Not because English is superior — because most AI systems' training and retrieval infrastructures still weight English content heavily. Your first English piece is the beginning of your entity signal, not a translation exercise.What comes next.I'll be running the same query weekly, in both languages, tracking how the portrait evolves as my public content changes. I'm starting my English-language publication track now — this article is part of it. I'll report back on how long it takes Gemini's English mode to recognize the same entity it already describes in Russian.Whether that gap closes quickly, slowly, or not at all — either outcome is a story worth telling.In classical SEO, the question was: can users find my page? In generative search, the question becomes: will AI systems understand who I am, trust my information, and recommend me? The first step is finding out what's currently being said — in every language your market operates in.SEO practitioner and AI Visibility researcher. Exploring how Generative AI systems understand brands, experts, and digital entities. In search since 2004