Multilingual AI Search Optimization: A 2026 Guide

Short answer: Multilingual AI search optimization means giving AI assistants a real page in each language your customers ask in. Publish every language at its own URL, link the versions with hreflang, write the content natively rather than as a thin translation, and keep your business facts identical across languages. An engine cannot recommend you in Spanish from an English-only page.

This guide is for businesses whose customers ask ChatGPT, Gemini, Claude or Perplexity in more than one language: a clinic serving Spanish and English speakers, a hotel taking bookings from abroad, a shop selling into several countries. It covers the technical setup, the content, and the off-site signals, in that order.

Why does language matter for AI search?

Language matters because AI assistants answer in the language of the question and draw on sources that match it, so a business with no content in that language has little for the engine to cite. When someone asks in Portuguese for a dentist who takes walk-ins, the assistant searches, reads what it finds, and writes a Portuguese answer. Pages that exist in Portuguese and say the right things are the easiest to use.

The reach is already wide. In May 2025 Google announced that AI Overviews had expanded to more than 200 countries and territories and more than 40 languages, adding Arabic, Chinese, Malay and Urdu among others. For a business with customers outside its main language, that means AI answers are being written in their language whether or not you have content for them.

How do AI engines tell what language a page is in?

Search engines decide a page's language mainly from its visible text, so a page has to actually be written in that language to count. Google is explicit about this in Managing Multi-Regional and Multilingual Sites: it uses the visible content of the page and does not use code-level signals such as the lang attribute or the URL to determine language.

Two consequences follow. A page with a translated menu and footer but English body text is still an English page. And Google recommends one language per page, because mixing languages makes the page harder to match to anyone. The AI engines that ground answers in search results, which is most of them for local and commercial questions, inherit these decisions from the indexes they search. Our explainer on training data vs live retrieval covers why the search index matters so much.

How do you set up a multilingual site for AI search, step by step?

There are seven steps, starting with giving each language its own URL and ending with checking what the assistants say in each language. Do them in order; the later steps depend on the earlier ones.

1. Give each language its own URL. Use subdirectories such as /es/ and /fr/, subdomains, or country domains. Google does not recommend URL parameters, and it advises against switching language with cookies or browser settings on a single URL.
2. Do not auto-redirect by language or location. Google says automatic redirects between language versions can stop users and search engines from seeing all versions. Offer a visible language switcher instead.
3. Add hreflang annotations. In the HTML head, HTTP headers or the sitemap, list every language version on every version, including the page itself, with fully qualified URLs. Add an x-default for visitors whose language you do not cover.
4. Translate the whole page, not the template. Headings, body copy, FAQs, prices, opening hours and calls to action all need to be in the target language.
5. Localise the facts. Use the local currency, date format, phone format and the words customers really use for your service in that language.
6. Mark up languages in structured data. Add schema to each version in its own language, and state the languages you serve customers in.
7. Check the answers in each language. Ask ChatGPT, Gemini, Claude and Perplexity your customers' questions in each language and note who is named and which pages are cited.

What are the hreflang rules that most sites get wrong?

The rule most sites break is reciprocity: every language version must list itself and all the others, or Google ignores the tags. Google's page on localized versions of your pages says that if two pages do not both point to each other, the tags will be ignored.

The other common mistakes are smaller but just as fatal to the annotation.

Can AI crawlers see your other languages?

Crawlers can only see your other languages if each one is reachable at a stable URL without depending on the visitor's location or browser settings. Google's page on locale-adaptive pages notes that Googlebot's default crawl comes from IP addresses that appear to be in the USA and sends no Accept-Language header, so content that changes by location or browser language can be missed. Google recommends separate locale URLs with hreflang instead.

AI crawlers have the same problem in a stronger form. Most fetch a page once, from one place, and many do not run JavaScript. If your Spanish content only appears after a script reads the browser language, an AI crawler may never see it. Our guide to JavaScript rendering and AI crawlers explains how to check what a crawler receives, and the AI crawlers and robots.txt guide covers making sure each language path is not blocked.

How should structured data handle multiple languages?

Structured data should be written in the language of each page and should state which languages you serve customers in. Put the LocalBusiness or Organization markup on each language version with the name, description and services in that language, and keep the address, phone and identifiers identical.

To say which languages your staff speak, Schema.org defines availableLanguage as a language someone may use with the item, service or place, using IETF BCP 47 codes. It applies to types such as ContactPoint, so the usual pattern is a contactPoint on your business with availableLanguage set to, for example, "en" and "es". Our schema markup for AI search guide shows the base markup to start from.

Do off-site signals need to be in each language too?

Yes, off-site signals count for more when they exist in the language the customer is asking in, because the assistant looks for agreement among sources it can read in that language. A business with Spanish-language reviews, a Spanish description on its Google Business Profile and mentions in Spanish-language local directories gives the engine sources that confirm the facts on its own Spanish page.

• Reviews. Let customers review in their own language, and reply in the language they wrote in.
• Directories. List the business in directories and community sites used by speakers of that language, with the same name, address and phone as everywhere else.
• Profiles. Say on your Google Business Profile and social profiles which languages you serve customers in.
• Consistency. A different phone number or address on the Spanish page than on the English one makes the engine less sure which is right.

These are the same consistency rules that apply in one language, described in our guide to directory listings and NAP citations, applied once per language. Businesses with several branches should also read AI visibility for multi-location businesses.

How do you check whether it is working?

You check by asking the AI assistants your customers' questions in each language and recording whether your business is named and which of your pages are cited. Use the phrasing a native speaker would use, not a literal translation of your English prompts, and repeat the check over several weeks because answers vary from one run to the next.

Alphaa's free 60-second AI check shows what ChatGPT, Gemini, Claude and Perplexity say about your business today, which is a quick baseline before you add languages. Alphaa itself is an AI agent that monitors those answers weekly and drafts fixes such as FAQ pages and structured data for the owner to approve, from $99 a month. It does not guarantee any recommendation in any language.

What else do businesses ask about multilingual AI search?

Is machine translation good enough for AI search?

Machine translation is a reasonable first draft, but a native speaker should review it before it goes live. Awkward phrasing and wrong local terms make a page less likely to match how customers ask, and errors in prices or hours spread into AI answers.

Should I use subdirectories or separate domains for each language?

Subdirectories such as /es/ are usually the simplest choice for a small business, because they are easy to set up and maintain on one site. Google lists country domains as the clearest geotargeting signal, but they cost more and target only one country each.

Does the lang attribute help AI engines?

Google says it does not use the lang attribute to determine a page's language, so it is not a ranking lever there. It is still worth setting correctly for accessibility and for other tools, but the visible text is what counts.

Do I need a separate llms.txt file for each language?

No, one llms.txt file at the root of the site is enough, and it can link to the key pages in each language. Our guide to creating an llms.txt file shows the format.

Which language should I add first?

Add the language your existing customers already use most after your main one. Your call logs, booking forms and review languages usually show it, and it is the language where you can get real reviews fastest.