Why your brand is invisible in AI answers in your own language
Andrei Nekrasov
TL;DR
- AI names a completely different set of companies depending only on the language of the question, not on how good a business actually is.
- A well-known local brand has a real advantage over global competitors in AI answers, but only once the question is asked in its home language.
- A website that exists only in English gets excluded from most AI answers in other languages before anyone judges its quality.
- Translating a website is necessary for AI to cite it in another language at all, but getting mentioned on other local-language websites improves visibility further than translation alone does.
- Being invisible in AI answers in a customer's own language costs more in some industries than others, and more when the real competitors are local brands, not global ones.
A customer now asks AI before they ask you. If AI doesn't name your brand in their own language, you don't get that customer, even if your product is the best one on the market. I'll show you why the language gap happens in AI answers and exactly how to fix the problem.
Section 1
Being named in English doesn't mean you're named in your own language
Ask an AI chatbot the same question in 2 languages, and it doesn't hand you a translated answer. It builds a different answer from scratch, naming different companies, depending only on which language you used.
In a test by SALT.agency, asking about a flight from London to Tenerife in Russian got British Airways named in 0% of the answers, while easyJet was named in 72.5%. Ask the exact same question in Persian, and the 2 swap places: British Airways jumps to 52.5%, easyJet drops to 12.5%.
The same swap (a different company wins once the language changes) turns up in categories that have nothing to do with each other: travel, electronics, accounting software.
| Category | AI tested | One language | Another language |
|---|---|---|---|
| Flights, London–Tenerife | Claude | Russian → easyJet, 72.5% | Persian → British Airways, 52.5% |
| Headphones | Claude | English → Sony, 50% | Russian / Persian → Sony, 100% |
| Accounting software | ChatGPT | English → FreshBooks, QuickBooks | Norwegian → Fiken, Tripletex |
Sources: SALT.agency, Entrepreneur
None of this is a handful of lucky examples stitched together. One study, reported by Entrepreneur, tested it directly: it asked an AI the same question 234 times, changing only the language, or only the brand. Changing the language changed the answer 26.5% of the time. Changing the brand barely changed anything: just 1.5% of the time.
If a customer asks about your category in their own language, they are not reading a translated version of what your English-speaking customers see. They are reading a different shortlist, with different companies on it; whether your name is one of them has so far had less to do with how good your business is than with which language the question was asked in.
Section 2
Local businesses beat global competitors in AI answers whenever the customer asks in their own language
Being well known in your own country is worth more inside an AI answer than being well known everywhere. Ask about a product or service in a brand's home language, and a name nobody outside that market has heard of can outrank a famous global brand.
One study tested the pattern across 66 real brands, in 11 countries and 12 languages, collecting 35,640 answers. Each time the same question moved from English to a brand's home language, a well-known local brand's chances of getting recommended jumped about 5 times more than a global brand's chances did.
The test covered 4 unrelated language families, not one corner of Europe
A brand that checks its AI visibility only in English will never see its own local-language edge, because English is exactly where that edge disappears. The researchers called the effect "a measurable language blind spot" around local brands.
The home-language advantage does not apply to every category. In a related test, run 234 times, one category showed no such effect at all: local suppliers existed, but the AI never named them, in any language. The reason had nothing to do with language. It came down to rules about who is legally allowed to offer that service.
If you are a known local name, AI answers in your own language are the one place your size does not work against you. A global competitor cannot buy the local-language edge by translating its own site. The edge only exists for a brand that real people already talk about in its own language, not just a brand whose own website has been translated. And it only exists in a category where AI names suppliers at all. In some categories, often regulated ones, AI names nobody, in any language, no matter how well known the brand is.
Section 3
AI cannot name you from an English-only website when your buyer asks in their own language
AI can exclude an English-only page from a non-English answer before anyone judges its content, for 2 separate reasons.
Reason 1 sits inside the AI itself: even when you ask a question in French, part of its search still runs in English, and part does not. One analysis of 10 million ChatGPT prompts found that 43% of the background searches behind a non-English question still ran in English, and 78% of non-English conversations had at least one hidden English search. That English slice is real, but it stays the minority of the search. So an English-only page can only be found through that English slice, while a page in the buyer's own language gets found through most of the search.
Reason 2 shows up once the AI actually writes the answer: it quotes pages that already carry a sign of the local language far more often than pages that don't. In one study of German-language questions, 65% of the pages the AI quoted carried a German signal: a .de domain, or a German section of a larger site. The same method, applied to French-language questions, found 49% of quoted pages carried a French signal. That leaves an English-only page, with no local signal at all, competing only inside the smaller share left over in each country.
An English-only website can be well written and still lose, because the problem is not quality. The problem is existence: whether the site has a page on a local domain or a local path, a country domain or a /de/, /fr/ style section, covering the exact topic the buyer is asking about, not just a translated homepage. AI skips a site without that page twice: once when it searches, and once when it decides what to quote.
Section 4
Translating your website gets you partway, but it does not guarantee AI will say your name
Translating a website is a real, measurable move. What it buys is access to a language AI did not reach before, not a guaranteed mention in your own.
One study of 1.3 million AI citations in Spain and Mexico compared untranslated Spanish-language sites against sites with both a Spanish and an English version. AI cited untranslated sites 17,094 times in Spanish, against just 2,810 in English: it mostly knew them in one language only. On sites with both versions, the imbalance nearly disappeared: 10,046 Spanish citations against 8,048 English. So translating a site does not add citations out of nowhere, it rebalances which language sees the site at all.
AI citations in Spain and Mexico, by site language and by version
Untranslated sites
Spanish citations
English citations
Sites with a Spanish and an English version
Spanish citations
English citations
Translating a site changes which language AI cites it in. It does not, by itself, make other people write about a brand in that language. In one study of German-language questions, associations and industry bodies made up 8.9% of citations, against 5.3% in the same study's American data. That means outside mentions, press, reviews, industry directories, move a brand's visibility more directly than translating its own homepage does.
Translating a website is a necessary step. Without it, AI mostly cannot cite a site in that language at all. But translating is not the step that gets a brand's name said. That comes from other people, writing about the brand in that language, on sites the brand does not own.
Section 5
Missing from AI answers in your customers' language costs more in some industries than in others, and more if your rivals are local brands
How much it costs a business to be invisible in a customer's own language depends on 2 things: the industry, and whether the competition is a known local name or a global one.
The same German-language study behind the 65% figure in Section 3 broke the number down by industry. Health questions had the strongest local pull: 73% of quoted pages carried a German signal. Finance and insurance followed at 69%. Industrial and manufacturing questions dropped to 37%, and education platforms came in lowest at 31%. The gap between the strongest and the weakest industry is 42 percentage points: 73% against 31%.
Share of pages AI quoted that carry a German signal, by industry
Section 2 already measured the other variable: a well-known local brand gains about 5 times more than a global brand does, from the same switch to its home language. So the 2 variables compound. A well-known local health brand in Germany sits inside the best possible combination of the 2; a global education brand sits inside the worst.
No single "AI visibility" number applies to every business. Industry and competitor type decide, together, whether this problem is urgent or barely worth worrying about. A local health brand checking only its English visibility is hiding its best advantage. A global education brand has little local advantage to find in the first place.
Section 6
Closing the language gap in AI answers takes 4 moves: the order matters
Sections 1 to 5 point to the same 4 moves, and the order matters: doing step 3 or 4 before step 1 or 2 wastes the work.
Check what AI says in each language separately.Test the brand's own category in every language a real customer uses, not only English. Section 1 showed AI gives a genuinely different answer in each language, not a translation of the same one.
Work out where the business actually sits.Which industry it's in, and whether the real competitors are local names or global ones. Section 5 showed that the share of local sources in AI answers swings from 73% down to 31%, depending on the industry.
Translate the site onto a real local address.A country domain or a local path, not just a language switcher on the same English domain. Section 3 showed an untranslated page is excluded from most of the answer before anyone judges its content.
Get mentioned on other local-language sites.Press, reviews, industry directories, not just the brand's own pages. Section 4 showed that outside mentions improve visibility more than translating a homepage does.
Steps 3 and 4 only pay off once steps 1 and 2 say where to spend them. Fixing every language, in every market, at the same time, is the slowest way to see a result.
Closing thought
Fixing the language gap in AI answers is not complicated. It is a matter of checking the right things in the right order, nothing more exotic than that. Do it now, and a brand becomes the answer AI gives by default, in languages its competitors never thought to check.