Can an English page be cited for a question asked in another language?

Yes, it happens, especially when no good page exists in the asker's language. The engine may translate or summarize the English source. That can still bring visibility, but the visitor lands on a page they may not read comfortably, which hurts conversion. A strong localized page gives the engine a better option in the asker's own language.

Do AI crawlers read hreflang?

There's no clear public documentation that AI-specific crawlers use hreflang directly. Search indexes that many answer engines rely on do use language signals, including hreflang, to route the right version to the right query. Keeping hreflang correct costs little and supports the retrieval step, so we treat it as required rather than optional.

Should I create pages only in English to concentrate authority?

Not if you sell to non-English markets. Buyers increasingly research in their own language, and an English-only site gives answer engines nothing native to cite. Subfolders on one domain let every language share authority anyway. The better trade-off is fewer languages done properly rather than English only, or many thin translations.

Is machine translation penalized in AI search?

There's no public evidence of a specific penalty for machine translation in AI answers. The practical issue is quality: unreviewed machine translation tends to be generic and occasionally wrong, which makes it a weak source to quote. Machine translation reviewed and adapted by someone who knows the market is a reasonable way to produce supporting content.

How often should I run a multilingual prompt audit?

Monthly is a sensible rhythm for most startups, with an extra run after any major content or structure change. AI answers vary from one run to the next, so look for patterns across several months rather than reacting to one result. Keep the prompts the same between runs so the comparison is fair.

All posts

Web design & development

Localizing for AI Search: Do Answer Engines Cite Translated Pages?

Do AI answer engines cite translated pages? An honest early take on multilingual AI search, what we know, what we don't, and how to test it yourself.

By Mehdi HassanOct 29, 20267 min read

The short answer

In multilingual AI search, answer engines can and do cite localized pages, but only when those pages are good search results in their own language: crawlable, server-rendered, properly tagged and genuinely useful to that market. No public evidence shows translation alone earns citations, so treat each language as its own content and measure it separately.

Key takeaways

AI engines mostly retrieve from search indexes, so localized pages need to rank and be crawlable first.
Thin or unreviewed machine translations are weak candidates for citation in any language.
No special markup is needed: Google says AI features use normal SEO, and tests found no schema lift.
Test it yourself: run the same buyer prompts in each language every month and log who gets cited.
Evidence on this topic is thin in 2026, so be wary of anyone selling certainty.

Do AI answer engines cite translated or localized website pages?

Yes, AI answer engines cite pages in many languages, and a localized page can be cited when someone asks in that language. What's not established is that translating a page by itself earns citations. The pages that get cited seem to be the ones that would also be good search results: clear, specific, crawlable and written for the person asking.

This is an opinion piece, and we want to be upfront about that. Public research on multilingual AI citations is thin as of late 2026, and we haven't found a controlled study we'd trust on the question. What follows is our reasoning from how these systems retrieve content, what has been tested about AI citations in general, and how we'd run the test on your own site. For the structural basics, start with our multilingual website guide.

How does multilingual AI search actually find pages?

Most AI answer engines find pages through a search step: they run queries, retrieve results from an index, then write an answer from what they retrieved. Google's AI features draw on Google's own index, and other assistants use web search behind the scenes. That means multilingual AI search inherits most of the rules of multilingual search.

If a German-language question triggers a German query, the candidates are German pages that the index understands as German and as relevant. A page that exists only as an English URL with client-side translation, or that hreflang points away from, may never be a candidate. This is why we start every conversation about AI visibility with indexing and structure rather than with AI-specific tactics.

Factor	Matters for multilingual AI citations?	Why
Each language on its own crawlable URL	Yes, it's the foundation	Retrieval can't cite a page it can't find or tell apart
Server-rendered or static HTML	Yes	Many AI crawlers don't run JavaScript, so client-only content is invisible
Correct hreflang and lang attributes	Likely	Helps search systems route the right version to the right query
Content written for the market	Likely, the biggest lever	Specific local answers are better retrieval matches than literal translations
AI crawler access in robots.txt	Yes, if you want to be cited	Blocking a crawler blocks that engine's direct fetches
llms.txt files	No measured effect	SE Ranking's 2026 study of 300,000 domains found no citation lift
Extra schema markup	No measured effect	Ahrefs' controlled test found no significant citation lift from adding schema

Why aren't translated pages enough for AI citations?

Translated pages aren't enough because answer engines pick passages that answer the question well, and a literal translation of English copy often answers the wrong question for that market. The German buyer asks about data hosting and invoicing; the translated page talks about speed and growth.

Unreviewed machine translation adds a second problem. It tends to produce awkward phrasing and generic statements, which are poor material to quote. Google's general guidance treats mass-produced, low-value content as a quality problem regardless of how it was made. Localized pages, with market-specific examples, pricing in local currency and answers to local objections, are simply better candidates. Our comparison of localization vs translation explains which pages deserve the extra work.

An answer engine can only quote what you actually said to that market.

Do you need special markup for AI search in other languages?

No. Google's documentation on its AI features says no special markup or AI-specific files are needed to appear in them, and our research found no measured lift from llms.txt or from adding schema. Standard technical SEO, done per language, is the work that matters.

That means each locale needs translated titles and descriptions, self-canonical URLs, correct hreflang, and the page language declared in the HTML. Structured data is still worth keeping clean for rich results, as our piece on schema markup and AI citations explains, but it's hygiene, not a citation strategy. The structure of the content itself matters more: answer first, then detail, as covered in our guide to answer-first page structure.

How can you test whether AI engines cite your localized pages?

Test it by running the same set of buyer questions, written natively in each language, through the major answer engines every month, and logging which of your URLs get cited. It's manual, but it's the only measurement that answers your actual question for your actual site.

Write prompts natively
Have a fluent speaker write 15 to 30 real buyer questions per language. Don't translate your English prompts; buyers in each market phrase things differently.
Run them across engines
Ask each question in Google's AI features, ChatGPT, Perplexity and any assistant your market prefers. Use a clean session and note the location and language settings.
Log citations by URL
Record whether you're mentioned, whether a URL is cited, and which language version it is. An English page cited for a German question is a useful signal in itself.
Track referrals by locale
Set up an AI referral channel in analytics and break it down by landing-page locale, so you can see which language versions actually receive visits.
Change one thing at a time
Localize one money page properly, then compare the next few monthly runs against pages you haven't touched. Keep a dated log of every change.

Hypothetical example: a first audit for a two-language SaaS site

For example, suppose a SaaS company has English and German sites, and wants to know whether its German pages appear in AI answers.

Its German sales lead writes 20 questions the way customers actually ask them, such as how to compare tools for a specific use case and what hosting options exist. The team runs them across three engines in German and logs results in a spreadsheet.

The first audit shows the company mentioned in a few answers, but usually with its English pricing page cited, not the German one. Checking that page reveals a canonical pointing at the English version and no German FAQ on hosting. The team fixes the canonical, adds German answers about hosting and invoicing, and reruns the same 20 questions monthly. Whatever happens next is evidence about their site, not a general rule.

What should you do about multilingual AI visibility now?

Fix the foundations per language, localize the few pages that sell, keep them fresh, and measure with a monthly prompt audit. That's the honest playbook while the evidence is still thin, and it's the same work that helps traditional search in each market.

Make sure every locale is server-rendered, crawlable and allowed in robots.txt for the AI crawlers you want, as in our guide to an AI-search-ready website.
Localize the home page, pricing and product pages first, with answers to that market's real questions.
Update localized pages when the source changes, not months later. Our note on content freshness for AI search covers how often.
Don't publish hundreds of unreviewed machine-translated pages hoping for citations.

Frequently asked questions

Can an English page be cited for a question asked in another language?: Yes, it happens, especially when no good page exists in the asker's language. The engine may translate or summarize the English source. That can still bring visibility, but the visitor lands on a page they may not read comfortably, which hurts conversion. A strong localized page gives the engine a better option in the asker's own language.
Do AI crawlers read hreflang?: There's no clear public documentation that AI-specific crawlers use hreflang directly. Search indexes that many answer engines rely on do use language signals, including hreflang, to route the right version to the right query. Keeping hreflang correct costs little and supports the retrieval step, so we treat it as required rather than optional.
Should I create pages only in English to concentrate authority?: Not if you sell to non-English markets. Buyers increasingly research in their own language, and an English-only site gives answer engines nothing native to cite. Subfolders on one domain let every language share authority anyway. The better trade-off is fewer languages done properly rather than English only, or many thin translations.
Is machine translation penalized in AI search?: There's no public evidence of a specific penalty for machine translation in AI answers. The practical issue is quality: unreviewed machine translation tends to be generic and occasionally wrong, which makes it a weak source to quote. Machine translation reviewed and adapted by someone who knows the market is a reasonable way to produce supporting content.
How often should I run a multilingual prompt audit?: Monthly is a sensible rhythm for most startups, with an extra run after any major content or structure change. AI answers vary from one run to the next, so look for patterns across several months rather than reacting to one result. Keep the prompts the same between runs so the comparison is fair.

Mehdi Hassan · CEO & Design Director

Copywriter, designer and developer, and founder of Designify Studio. Over five years, 60+ projects across 14+ industries, for startups that have raised $300M+.

Want us to do this for you?

We take craft seriously over at Designify Studio. We do deep research and do what we do best - pay attention to the smallest details and make sure we speak to the right audience.

Schedule Call