Here is the direct answer: AI assistants answer in the language the question was asked in, and they lean on sources in that language to do it. If every page you publish is in English, you are effectively absent from every question asked in Spanish, Portuguese, Vietnamese or Tagalog — including questions about your own business, in your own city.
Most people never check this, because they never think to ask. They test their visibility in English, see themselves named, and conclude they are visible. They are visible in one language.
Why the answer changes with the language
A model assembling an answer in Spanish will preferentially retrieve and cite Spanish-language sources. Not exclusively — it can and does draw on English material and translate the substance — but a competitor with a real Spanish page is holding a much stronger hand than you are with nothing.
There is a second effect that matters more. Language is not just vocabulary; it is a different set of questions. The Spanish-speaking version of your customer may be asking about things the English-speaking version never raises, framing the problem differently, or using a category term that does not translate cleanly. You cannot answer a question you have not noticed is being asked.
Who this actually matters for
Not everyone. Be honest about the market before spending on this.
It matters most for local and regional businesses in places with a large non-English-speaking population, where the service is one people genuinely research before buying — clinics, legal services, trades, schools, gyms, financial services. It matters for businesses that already get walk-in or phone enquiries in another language, which is the strongest possible signal that the demand is there and currently being handled by nobody online.
It matters much less for a narrow B2B product whose buyers all work in English regardless of where they live.
The cheap test: ask an assistant, in the target language, the question a customer in your area would ask, and see who gets named. If it names three competitors and not you, you have found a gap. If it names nobody at all and gives a generic answer, you have found something better than a gap — an empty category.
Translation is not the strategy, but it is the start
Machine translation has crossed the threshold where a competent translated page is genuinely useful. That is a real change from a few years ago, and it lowers the cost of entry here considerably.
But a translated page inherits the limits of the original. If the English page answers the questions English speakers ask, translating it produces a Spanish page answering the questions English speakers ask. That is better than nothing and worse than the thing that actually wins.
The approach I would take, in order:
- Translate your three or four highest-intent pages first. Services, pricing, location, and whatever page answers your single most common pre-purchase question. Not the whole site. Depth beats breadth here, and a sprawling half-maintained second language is worse than four excellent pages.
- Have a native speaker read them. Not retranslate — read. They will catch the term your customers actually use, which is routinely not the dictionary translation, and that term is what people type.
- Write one or two pieces natively, answering a question that only comes up in that language. This is the part that produces an answer nobody else has, and it is the part everyone skips.
- Mark the language properly. Declare the page language, and use hreflang if you have genuine parallel versions. Machines should not have to infer this.
The mistake that undoes the work
Half-translating. A page where the headings are in Spanish and the body drifts back to English, or where the translated page links only to English pages, reads as abandoned — to a person and to a model. Partial translation can leave you worse off than none, because it creates a page that looks like an answer and is not one.
Decide what you are supporting and support it completely. Four finished pages beat forty abandoned ones.
What to expect
My read, not a guarantee: in a market where nobody has bothered, a handful of genuinely good pages in the second language can move a language-specific answer faster than the equivalent effort in English, because the English answer is crowded and the other one frequently is not. That asymmetry does not last forever, but right now it is real and widely unexploited.
Start by running the test. Ask the question in the language your customers speak and look at who gets named. That five-minute check tells you whether any of this is worth your money, which is a far better basis for deciding than my opinion or anyone else's.
Key takeaways
- AI answers in the language of the question and leans on sources in that language.
- Testing your visibility only in English tells you about one language, not your market.
- Another language is a different set of questions, not just different words for the same ones.
- Translate three or four high-intent pages properly rather than the whole site badly.
- Have a native speaker check terminology — the dictionary translation is often not what customers type.
- Half-translated pages read as abandoned and can leave you worse off than none.
Frequently asked questions
About the author
Scott Tischler is the Founder & Chairman of AIrecommend.ai and a practitioner-authority on AI search and Answer Engine Optimization. With 20+ years in marketing technology — including American Express, MetLife, and UBS — and executive and professional study at Wharton, Harvard, and Oxford, he helps businesses become the ones AI recommends.
Want to be the business AI recommends?
See how AIrecommend.ai builds the entity authority answer engines reward.
Explore AIrecommend.ai