Your authority doesn’t translate itself
Structure and markup refreshed for current answer engines; the original analysis is preserved.
This month Bard reached the European Union and Brazil, conversing in more than forty languages. As the answer engines go multilingual at scale, a hard truth follows for anyone who sells across borders: the reputation a model recognizes is built language by language. Being well known in English does not make you known in Spanish — and a model answering in Spanish will tell you so.
the short answer
A model answering in Spanish leans on Spanish-language sources, not a translation of its English answer. So if your reputation lives almost entirely in English, a Spanish query has thinner material to draw on and is likelier to leave you out. Your English standing is not worthless across the border — multilingual models share some understanding — but the corroboration the model reaches for first is local to the question’s language. Translation is not standing. Earn depth where your growth buyers ask, in their language; that is the visibility that does not transfer for free.
key takeaways
- This month Bard reached the EU and Brazil in more than forty languages — the answer engines are going multilingual at scale, and that changes where your visibility lives.
- A model answering in Spanish leans on Spanish-language sources. If your reputation lives almost entirely in English, a Spanish query has thinner material to work from and is likelier to leave you out.
- Your English standing is not worthless across the border — multilingual models share some understanding — but the corroboration the model reaches for first is local to the question’s language.
- Translation is not standing. Being named confidently comes from being corroborated as a real entity by trusted sources, and those sources are largely language-specific; your own content is a claim, not corroboration.
- With limited resources, go deep where your growth buyers ask, in their language — earned authority in the one or two languages that map to revenue beats a shallow translated footprint everywhere.
the same brand, two languages
Nothing about the company changed between the two panels. What changed is the body of evidence the model could reach for: rich in one language, sparse in the other. That gap is the whole subject of this note.
Why a translation feature is a visibility problem
It is tempting to read "Bard now speaks forty languages" as a user convenience and nothing more. For a company that sells in more than one of those languages, it is something sharper: a quiet redrawing of where your visibility has to be earned. The moment a generative surface answers fluently in Spanish to hundreds of millions of people, the question "do the engines name us?" splits into one question per language, each with its own answer. A buyer in São Paulo or Madrid or Mexico City asking in their own language is not querying the same evidence base as a buyer asking in English, so the comfortable assumption that your hard-won English reputation covers you everywhere stops being safe. It was always a little false; multilingual reach at this scale is what makes it expensive.
To be fair to the technology, this is not a wall between languages. Modern multilingual models genuinely share representations across tongues, so a brand that is enormous in English is not invisible the instant a question switches to Spanish — some of that standing carries. The claim here is narrower and, we think, more useful: the carry is partial and unreliable, and the evidence a model weights most heavily for a Spanish question is Spanish. So treating your languages as one undifferentiated market will leave you strong where you did the work and quietly weak where you did not — and "quietly" is the dangerous part, because nothing tells you that you are missing from answers you never see, in a language you may not be watching.
The shift, in three parts
The answer engines just went multilingual; the model trusts corroboration, and corroboration is local; so earn depth in the languages that map to revenue. Open each part for where it changes the plan.
01 The answer engines just went multilingual
Until now the generative surfaces that mattered most spoke mostly English, which let a lot of companies quietly assume their AI visibility was a single thing. This month that assumption gets harder to hold. Bard’s biggest expansion to date puts it in front of hundreds of millions of new people across the European Union and Brazil, conversing in more than forty languages — Spanish, German, Arabic, Hindi, Chinese, and many more. The mechanics of how a model decides whom to name do not change at a border; what changes is the pool of material it draws on, because a question asked in Spanish pulls most strongly on what exists in Spanish. So the same model that confidently names you for an English question can come up blank for the equivalent Spanish one, not out of inconsistency but because it is working from a different, language-specific body of evidence. The era when "are we visible in AI?" had one answer is ending; the honest version of the question now has a language attached, and the answer in each one depends on what you have actually earned there.
02 The model trusts corroboration, and corroboration is local
The reason this bites is the same reason earned authority matters at all: a model is far more confident naming a provider that independent sources describe as credible than one that only describes itself. That corroboration — the reviews, comparisons, mentions, references, and write-ups produced by others — is overwhelmingly language-specific. The Spanish-speaking corner of your category has its own publications, its own communities, its own reference points, and they are what a Spanish answer leans on. If you have spent years earning that kind of standing in English and none of it in Spanish, you arrive at a Spanish query with a strong claim about yourself and almost nothing behind it from anyone else. It is worth being precise here: a multilingual model does carry some cross-language understanding, so your English reputation is not nothing. But it is thin gruel compared with being corroborated in the language of the question, and thin gruel is what gets you left out of the shortlist a model offers when someone asks in that language.
03 So earn depth in the languages that map to revenue
The wrong response to this is to panic-translate everything into forty languages and call it coverage; a shallow translated footprint everywhere earns you little, because translation is not corroboration. The right response is to be deliberate. Find the languages where your actual growth depends on being found — usually one or two, the markets where the buyers you want are asking in their own tongue — and concentrate on earning genuine, corroborated presence there. That means getting written about, reviewed, compared, and referenced by independent voices in that language, in the places its speakers actually look, in addition to offering your own content localized and correctly marked up so the model can attribute it to you. Depth in the languages that matter will beat breadth that is only skin-deep, every time, because the thing being rewarded is reputation others vouch for, and that is earned in a specific market. This is the discipline the AC Group has worked in for 27 years: authority is not a translation job, it is built where your buyers are, in the words they use.
One company, two markets
A software company has spent five years building a real reputation in its English-speaking home market: it is reviewed on the English-language sites buyers trust, compared favourably in English roundups, quoted by English trade press, and discussed in English communities. Ask an English-speaking model who the credible options are in its category and it names the company without hesitation, because the evidence is everywhere. The company now wants to grow in Spanish-speaking markets, and it does the obvious thing: it translates its website, publishes Spanish versions of its pages, and waits.
A buyer in Mexico asks a model, in Spanish, for the leading options in that category — and the company is not among them. Not because its Spanish pages are bad, but because nobody in Spanish has vouched for it: no Spanish-language reviews, no comparisons in Spanish publications, no mentions in the Spanish communities the model leans on. Its own translated pages are a claim about itself; what is missing is everyone else. The competitor that does get named is not necessarily better — it is corroborated in Spanish, written about by independent Spanish voices, treated as a known quantity in that language. The first company has authority that stops at the language border, and translating its site did nothing to move it across, because the thing that needed translating was never the company’s own words — it was other people’s belief in the company, and that has to be earned again in each language it cannot be copied into.
What to do with this
Start by making the invisible visible: pick the languages that actually matter to your growth, and for each one, ask a model in that language the questions your buyers would ask to find a provider like you. Read the answers as a local buyer would. Where you are named and well described, you have earned standing; where you are absent or thin, you have found the gap that your home-language confidence was hiding. Do this honestly per language rather than assuming your best market reflects the rest, because the whole point is that it does not.
Then build depth, not breadth. Resist the urge to translate yourself into every language at once; choose the one or two that map to real revenue and earn genuine corroboration there — independent reviews, comparisons, mentions, and references in that language, in the places its speakers look — alongside your own localized, properly marked-up content so the model can attribute your material to the right entity. Keep that entity consistent across languages so the model knows the Spanish you and the English you are the same company. None of this is fast, and that is the point: the visibility that transfers for free is not worth much, and the visibility that matters is earned market by market, language by language. That is the work the AC Group has done for ' + years + ' years — building authority where the buyers actually are, in the language they actually use, and never mistaking a translation for the reputation behind it.
Check it language by language
The first move is cheap and uncomfortable in the right way: audit yourself in each language that matters, separately, instead of trusting a single impression. Take the handful of questions a buyer would actually ask to find a provider like you — what the options are, who the credible names seem to be, what to watch for — and ask them in each target language, phrased the way a native speaker would phrase them. Then read the answers the way a local buyer would: are you named, are you described accurately, are you in respectable company or relegated to an afterthought? Do it more than once per language, and note not just whether you appear but how — the framing in one language is often quietly different from another.
What you are looking for is the shape of the gap, not a single verdict. A company can be the confident default in English, a weak maybe in Spanish, and entirely absent in Portuguese, and only a per-language check will show you that. The temptation is to extrapolate from your strongest market because it feels representative; resist it, because the entire lesson here is that it is not. Treat each language as its own scoreboard, write down where you stand on each, and you will have turned an invisible problem — missing from answers you never see, in languages you were not watching — into a concrete list of places to earn standing.
Keep one entity across the languages
There is a subtlety worth getting right, because it is easy to overcorrect. The fact that authority is earned per language does not mean you should become a different company in each one. The opposite: the model needs to understand that the Spanish you and the English you are the same entity, so that whatever cross-language understanding it does have actually accrues to you rather than splitting into two half-known strangers. That means a consistent name, consistent identifiers, and the technical signals — hreflang, language-correct localized pages, the same clean entity references — that tell a model your localized sites are facets of one organization, not unrelated ones. The earned, off-site reputation is what differs by language; the entity that reputation attaches to should be unmistakably singular.
Get both halves right and they reinforce each other. A clean, single entity means the partial cross-language carry has somewhere to land, so the corroboration you earn in Spanish strengthens a brand the model can already connect to its English self, rather than building a second, disconnected reputation from zero. Get the entity wrong — different names, sloppy references, no hreflang — and you make the model’s job harder in every language at once. So the discipline is two layered things at once: one entity, made unambiguous everywhere, carrying many language-specific reputations, each earned where its speakers actually are.
Multilingual AI visibility: quick answers
Why would my AI visibility differ by language?
Because a model answering in Spanish is not simply translating its English answer — it is drawing, with a heavy lean toward the language of the question, on what it has read in that language. When someone asks in Spanish who the credible providers in a category are, the model is most influenced by Spanish-language sources: the write-ups, comparisons, mentions, and references that exist in Spanish. If your reputation lives almost entirely in English — your own English site, English reviews, English press — a Spanish query has thinner material to work from, and you are more likely to be underweighted or left out. It is not that your English standing counts for nothing across the border; multilingual models do share some understanding across languages. It is that the corroboration the model reaches for first is local to the question’s language, so a gap in that language shows up as a gap in the answer.
Can’t the model just translate my English content?
It can translate, and it does carry some knowledge across languages, but translation is not the same as standing. The thing that makes a model name you confidently is not that your words can be rendered in another language; it is that you are corroborated as a real, credible entity by sources the model trusts — and those sources are largely language-specific. A glowing English case study does not become a Spanish-language industry reference just because the model can read both. What earns you a place in a Spanish answer is being talked about, reviewed, compared, and cited in Spanish, by others, in the places a Spanish-speaking buyer’s questions reach. You can and should make your own content available in each language, but your own content is your claim about yourself; the corroboration that backs the claim has to exist in that language too.
Where should a company with limited resources focus?
Where the buyers you actually want are asking, in the language they actually ask in. The mistake is to assume your strongest market is the one that needs the work; often it is the opposite — you are already well corroborated in your home language and invisible in the one where you are trying to grow. So look at where your growth depends on being found, identify the language of those buyers, and check honestly whether a model answering in that language knows you. If it does not, that is where earned authority pays off most: not spreading thin across forty languages, but building genuine, corroborated presence in the one or two that map to real revenue. Depth in the languages that matter beats a shallow translated footprint everywhere, because the model rewards corroboration, and corroboration is something you earn in a specific language and market, not something you spray across all of them at once.
Is this just international SEO under a new name?
It shares roots with international SEO but the emphasis moves. International SEO has always cared about serving the right content to the right region, with the right hreflang and localized pages, and all of that still matters because it helps a model find and attribute your own material correctly. What changes is the weight on off-site, language-specific reputation. Ranking a translated page is a necessary baseline; being named in a generative answer in that language depends more on whether independent voices in that language treat you as a known quantity. So the technical international-SEO hygiene is the floor, not the ceiling. Above it sits the harder, slower work of being corroborated in each language by sources other than yourself — which is exactly the kind of earned authority that does not transfer for free when you cross a linguistic border.
A note on sources and timing
This is written in July 2023, just after Google brought Bard to the European Union and Brazil and to more than forty languages, including Spanish — its largest expansion to that point. We have described only what was public as of this writing. The durable point does not depend on which engine led the way: as generative answers become fluent in many languages, the reputation those answers reflect is built per language, because a model leans hardest on the sources written in the language of the question. Multilingual models do share some understanding across tongues, so this is a matter of weight and reliability, not an absolute wall — but the practical consequence stands: being corroborated in English does not make you corroborated in Spanish, and the authority worth having is earned in each market you want to win. That is the standard the AC Group has held for 27 years — authority lives where your buyers are, in the words they use, and it does not translate itself.